diff --git a/.gitattributes b/.gitattributes index bfd32c6e2a1a5791f53b9e64f1d970dec67349a4..a30acf3f2a5b24d087053d745f654ac1060a8067 100644 --- a/.gitattributes +++ b/.gitattributes @@ -1413,3 +1413,58 @@ ignacia/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text ignacia/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text ignacia/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text ignacia/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +detailer-ce/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +detailer-ce/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +detailer-ce/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +detailer-ce/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +detailer-ce/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +detailer-ce/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +instagram-reality/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +instagram-reality/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +instagram-reality/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +instagram-reality/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +instagram-reality/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +instagram-reality/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +jacques-tati-movies-style/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +jacques-tati-movies-style/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +jacques-tati-movies-style/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +jacques-tati-movies-style/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +jacques-tati-movies-style/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +jacques-tati-movies-style/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +cg-texture-master-iray-skin-rendering-cg-3s/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +cg-texture-master-iray-skin-rendering-cg-3s/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +cg-texture-master-iray-skin-rendering-cg-3s/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +cg-texture-master-iray-skin-rendering-cg-3s/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +cg-texture-master-iray-skin-rendering-cg-3s/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +cg-texture-master-iray-skin-rendering-cg-3s/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +klein-realistic-comic-slider/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +klein-realistic-comic-slider/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +klein-realistic-comic-slider/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +klein-realistic-comic-slider/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +klein-realistic-comic-slider/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +klein-realistic-comic-slider/example_images/06_poster.jpg filter=lfs diff=lfs merge=lfs -text +klein-realistic-comic-slider/example_images/06_video.mp4 filter=lfs diff=lfs merge=lfs -text +claire-redfield-resident-evil-2-remake-in-real-life/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +claire-redfield-resident-evil-2-remake-in-real-life/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +claire-redfield-resident-evil-2-remake-in-real-life/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +claire-redfield-resident-evil-2-remake-in-real-life/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +claire-redfield-resident-evil-2-remake-in-real-life/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +claire-redfield-resident-evil-2-remake-in-real-life/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +aniedit-flux-2-klein/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +aniedit-flux-2-klein/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +aniedit-flux-2-klein/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +aniedit-flux-2-klein/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +aniedit-flux-2-klein/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +aniedit-flux-2-klein/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +hand-drawn-sketch/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +hand-drawn-sketch/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +hand-drawn-sketch/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +hand-drawn-sketch/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +hand-drawn-sketch/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +hand-drawn-sketch/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +the-body-sfw/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +the-body-sfw/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +the-body-sfw/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +the-body-sfw/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +the-body-sfw/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +the-body-sfw/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text diff --git a/aniedit-flux-2-klein/AniEdit9B_v2.safetensors b/aniedit-flux-2-klein/AniEdit9B_v2.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f437f61f356dd6852ce5491ed42291b2f4159a33 --- /dev/null +++ b/aniedit-flux-2-klein/AniEdit9B_v2.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d37d58268f6e704f7d5d5daf3cfc4d69bc82189c71e17494b5e02b6892274927 +size 994079944 diff --git a/aniedit-flux-2-klein/example_images/01_image.jpeg b/aniedit-flux-2-klein/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..0ea246b31027dc5e0cbeef63286f7293793fcada --- /dev/null +++ b/aniedit-flux-2-klein/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6ffee8b1498b87d06829554c7db1eeee5dfac0e7cbca5744a897aa4c2e38048 +size 404423 diff --git a/aniedit-flux-2-klein/example_images/02_image.jpeg b/aniedit-flux-2-klein/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..7ba26d2de3bad43ee8192df1f5b7bcc0aa350a44 --- /dev/null +++ b/aniedit-flux-2-klein/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25b07f427495024f4421ac03c6f41d38f5448884eff5b9cf0f1cd2a79e02b103 +size 4133694 diff --git a/aniedit-flux-2-klein/example_images/03_image.jpeg b/aniedit-flux-2-klein/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..cfdb16af051f59a7285f628531897cd2bfde3a23 --- /dev/null +++ b/aniedit-flux-2-klein/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:98e56ec427e4ecc31cf4af5d9048acda03264caacaf0095a852d566e810fa843 +size 324992 diff --git a/aniedit-flux-2-klein/example_images/04_image.jpeg b/aniedit-flux-2-klein/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..c1f010560e1ef444ebdf6a218e1c32ce748d650b --- /dev/null +++ b/aniedit-flux-2-klein/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:828a4aae2fd1c2abe954d455edf148264f31c768b3f215046dcbfdd39a1c369c +size 291001 diff --git a/aniedit-flux-2-klein/example_images/05_image.jpeg b/aniedit-flux-2-klein/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..1d8d5cb149e37e1f0f4cecf2a21d21f7bdb6d306 --- /dev/null +++ b/aniedit-flux-2-klein/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58920e0a3a0fc79ee5ac78aa11f6ba5a381b430e087a1d23bf7986281f83abf6 +size 2984740 diff --git a/aniedit-flux-2-klein/example_images/06_image.jpeg b/aniedit-flux-2-klein/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..5df1450e8abfe56fb31c64ee6951a8247f0c33de --- /dev/null +++ b/aniedit-flux-2-klein/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f900bc75f13ae9f506c6fbb4f599691ba968715a18fae8e3c19ff5139005ae32 +size 305250 diff --git a/aniedit-flux-2-klein/info.txt b/aniedit-flux-2-klein/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..3fb29d51f6e69ba17a5c9733fc153b201728451d --- /dev/null +++ b/aniedit-flux-2-klein/info.txt @@ -0,0 +1,247 @@ +═══════════════════════════════════════════════════════════════════════════════ + ANIEDIT (FLUX 2 KLEIN) +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model FLUX.2 [klein] 9B + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Visual style + Author memnoid + Source https://civitai.com/models/2332320 + Provenance CivitAI · model 2332320 · version 2642969 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File AniEdit9B_v2.safetensors + Version 9B v2 + Size 948.0 MB + Format SafeTensor / unknown precision + SHA256 d37d58268f6e704f7d5d5daf3cfc4d69bc82189c71e17494b5e02b6892274927 + Published 2026-01-30 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps Distilled + FLUX.2-klein-9B: 4 (HF card example; it is step-distilled to exactly this, + and more steps do not refine further). Base FLUX.2-klein-base-9B: 50 in the + HF card example and in ComfyUI's benchmark; community practice runs lower, + with ~20 for production and 8-12 for fast iteration reported in the CivitAI + consistency-LoRA article., cfg Distilled: guidance_scale 1.0 (HF card). It + is guidance-distilled and its transformer config sets + guidance_embeds=false, so raising CFG degrades rather than improves. Base: + guidance_scale 4.0 (HF card); the CivitAI consistency-LoRA article + recommends 3.5-4.5 for most edits., sampler Not stated by BFL or by + docs.comfy.org, and the official flux2 GitHub repo documents no sampler, + scheduler or shift values either. Community consensus is euler with a + resolution-dependent shifted-Euler / flow-match schedule (ComfyUI's Flux2 + workflows use a scheduler that interpolates shift between base_shift and + max_shift by output resolution). Start from ComfyUI's downloadable Flux.2 + klein workflow template rather than a blog. See unverified., resolution + 1024x1024 in every BFL HF card example and in CivitAI's 9B training recipe + ("resolution": 1024). Non-square ratios work but no official list of + supported dimensions or a stated alignment multiple was found — see + unverified., lora_strength 1.0 is the diffusers/ComfyUI default and what + CivitAI applies to training previews (strength 1.0). Published per-LoRA + guidance varies and should be read from the LoRA, not assumed: the widely + used Klein 9B consistency LoRA's CivitAI article says start at 0.5 within a + 0.2-1.0 range (0.2-0.4 loose, 0.5-0.7 balanced, 0.8-1.0 maximum structural + stability) and warns that too high a strength stops edits from applying; + SOLRICKS' Realistic Detail LoRA card says 0.6-0.8 subtle, 0.8-1.0 stronger, + start 0.8; the vafipas663 distillation-delta LoRA card warns it "produces + static noise at strength > 0.5. And sometimes, you have to go as low as + 0.1.". + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 2,840 + Rating 100% positive (128/128) + Comments 15 + Published 218 days ago + Download rate 8.9/day since release + + Adoption rank #103 of 200 in this collection (Modest) + Momentum rank #68 of 200 by download rate (Steady) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general FLUX.2 [klein] 9B ones, not anything specific to this LoRA. + + General FLUX.2 [klein] 9B rules that apply here (see REFERENCE.txt): + + • Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model + output into model input. There is no CLIP side to route — klein LoRAs + are model-only because the Qwen3 text encoder is not trained (CivitAI's + recipe fixes trainTextEncoder: false). + • Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B + LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, + 5+20 vs 8+24 blocks), not a quality question. + • Mixing a "Flux.2 Klein 9B" LoRA and a "Flux.2 Klein 9B-base" LoRA in + one stack is fine — those strings describe the training checkpoint, not + incompatible tensor layouts. Set the sampler for the checkpoint + actually loaded. + • Run base-trained LoRAs on the distilled checkpoint when you want speed: + HF's klein LoRA guide reports the adapter loads there and typically + gives better results than on base, at 4 steps. + • A base-to-turbo / distillation-extraction LoRA is a whole-model delta, + not a content LoRA. Put it on the base checkpoint only, and treat its + strength as a steps/CFG dial rather than a style knob. Do not add it on + top of the already-distilled checkpoint. + • Read each LoRA's own published strength range instead of defaulting + everything to 1.0 — klein LoRAs in the wild ship very different ones + (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, + under 0.5 and sometimes 0.1 for the vafipas663 distillation delta). + • With a consistency or edit LoRA in the stack, excessive strength + suppresses the prompt: the consistency LoRA's article says that if + edits stop applying, lower it before changing anything else. + • Prefer runtime LoRA patching over baking a LoRA into the checkpoint: + ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein + models (4B or 9B) since a March 2026 update — issue #13637, still open. + • The author never states which weight family this targets. Two different + questions, two different answers. Within 9B (distilled vs base — the + two CivitAI strings): not a real incompatibility. Both load through the + same Flux2KleinPipeline and have the same repo layout; for the 4B pair, + where both transformer configs are public, the distil + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Well established — 2,840 downloads. [source: CivitAI stats] + + Positively received — 128 thumbs-up with no down-votes. [source: + CivitAI stats] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Author references NSFW output — review before using in shared work. + [source: description] + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Large at 948 MB — slower to load and heavier to stack. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Two different questions, two different answers. Within 9B (distilled vs + base — the two CivitAI strings): not a real incompatibility. Both load + through the same Flux2KleinPipeline and have the same repo layout; for the + 4B pair, where both transformer configs are public, the distilled and base + configs are byte-for-byte identical in every shape field, and the 9B pair + is built the same way. HF's own klein LoRA guide states the workflow + directly: "Distilled models are step-compressed for fast inference; you + train against the base checkpoint and the adapter still loads on the + distilled model afterward", and "Applying the LoRA on the distilled model + typically gives better results than the base model, and it's faster" (that + post is written against 4B, but the family structure is the same). So the + two CivitAI strings record which checkpoint the author trained or previewed + against, not a hard compatibility wall. The real risk is a silent settings + mismatch: load a base-trained LoRA on the distilled checkpoint while + keeping the author's 20-50 step / CFG ~4 preview settings and the output + degrades badly (distilled wants 4 steps / guidance 1.0), or the reverse. + Nothing errors — it just looks wrong. Across sizes (9B vs 4B): a genuine + architectural mismatch, from the published transformer configs — different + hidden width (4096 vs 3072), different block counts (8+24 vs 5+20) and a + different text-conditioning dim (12288 vs 7680, following Qwen3-8B vs + Qwen3-4B). A 4B LoRA cannot patch a 9B model: expect missing keys and shape + errors, not degraded output. The exact ComfyUI error text is unverified. KV + variant: same 9B size and same 4-step distillation, but a distinct + checkpoint with its own pipeline; whether standard klein 9B LoRAs patch it + cleanly is not documented — see unverified. + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + A comprehensive finetune of Flux.2-Klein 4B/9B, that works on distilled + versions. Art style transfer using 4 reference images: Workflow + Upscale any anime image, prompt at 2-4MP: Workflow + ControlNet any single subject image: Workflow ...And many other edit tasks: + Real life to anime, change clothes, prompt new pose, change expression, + hairtstyle, pupils, Compose two subjects, convert to 3d model, + anthropomorphize, etc. Dataset (1500 image pairs CONTAINS NSFW /Drawn + nudity) + +─────────────────────────────────────────────────────────────────────────────── + VERSION NOTES +─────────────────────────────────────────────────────────────────────────────── + Doubled the dataset. + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 832x1216) + 02_image.jpeg (image, 1056x1952) + 03_image.jpeg (image, 832x1216) + 04_image.jpeg (image, 1216x832) + 05_image.jpeg (image, 1248x1664) + 06_image.jpeg (image, 832x1216) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + 1. Diffusion model -> ComfyUI/models/diffusion_models/. Distilled: + flux-2-klein-9b-fp8.safetensors. Base: flux-2-klein- + base-9b-fp8.safetensors. (BF16 single-file weights + flux-2-klein-9b.safetensors / flux-2-klein-base-9b.safetensors from the + gated HF repos also work if you have ~29GB; community GGUF builds exist + from unsloth and leejet for lower-VRAM setups.) + 2. Text encoder -> + ComfyUI/models/text_encoders/qwen_3_8b_fp8mixed.safetensors (from Comfy- + Org/vae-text-encorder-for-flux-klein-9b). This is the 9B family's encoder; + the 4B family uses qwen_3_4b.safetensors — do not mix them. + 3. VAE -> ComfyUI/models/vae/flux2-vae.safetensors (shared across klein + sizes). + 4. LoRA -> ComfyUI/models/loras/ as a plain .safetensors. + 5. Load with Load Diffusion Model (UNETLoader) for the transformer, + CLIPLoader for the Qwen3 text encoder, VAELoader for the VAE. + docs.comfy.org ships prebuilt Flux.2 klein graphs under Workflow Templates + for both the distilled and base branches — start from those rather than + wiring by hand, since neither the docs nor BFL's repo state + sampler/scheduler/shift in prose; those values live in the template. + 6. Insert LoraLoaderModelOnly between the diffusion-model loader and the + sampler. Model-only is correct for klein: the Qwen3 text encoder is not + trained by the standard klein trainers, so there is no text-encoder half to + patch. + 7. Set steps and guidance to match the checkpoint you actually loaded, not + the one the LoRA was trained on: 4 steps / guidance 1.0 on distilled, + ~20-50 steps / guidance ~4.0 on base. + 8. Note the 9B repos are gated on HuggingFace — you must accept the FLUX + Non-Commercial License on the repo page before any download, including + scripted ones. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 03:55 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/aniedit-flux-2-klein/metadata.json b/aniedit-flux-2-klein/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..58ecf25740489b873a413fe14338f37cff41d720 --- /dev/null +++ b/aniedit-flux-2-klein/metadata.json @@ -0,0 +1,131 @@ +{ + "name": "AniEdit (Flux 2 Klein)", + "slug": "aniedit-flux-2-klein", + "base_model": "FLUX.2 [klein] 9B", + "media_type": "image", + "model_id": 2332320, + "version_id": 2642969, + "version_name": "9B v2", + "source": "civitai", + "url": "https://civitai.com/models/2332320", + "repo_id": null, + "file_path": null, + "author": "memnoid", + "category": { + "key": "style", + "label": "Visual style" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model output into model input. There is no CLIP side to route — klein LoRAs are model-only because the Qwen3 text encoder is not trained (CivitAI's recipe fixes trainTextEncoder: false).", + "Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, 5+20 vs 8+24 blocks), not a quality question.", + "Mixing a \"Flux.2 Klein 9B\" LoRA and a \"Flux.2 Klein 9B-base\" LoRA in one stack is fine — those strings describe the training checkpoint, not incompatible tensor layouts. Set the sampler for the checkpoint actually loaded.", + "Run base-trained LoRAs on the distilled checkpoint when you want speed: HF's klein LoRA guide reports the adapter loads there and typically gives better results than on base, at 4 steps.", + "A base-to-turbo / distillation-extraction LoRA is a whole-model delta, not a content LoRA. Put it on the base checkpoint only, and treat its strength as a steps/CFG dial rather than a style knob. Do not add it on top of the already-distilled checkpoint.", + "Read each LoRA's own published strength range instead of defaulting everything to 1.0 — klein LoRAs in the wild ship very different ones (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, under 0.5 and sometimes 0.1 for the vafipas663 distillation delta).", + "With a consistency or edit LoRA in the stack, excessive strength suppresses the prompt: the consistency LoRA's article says that if edits stop applying, lower it before changing anything else.", + "Prefer runtime LoRA patching over baking a LoRA into the checkpoint: ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein models (4B or 9B) since a March 2026 update — issue #13637, still open.", + "The author never states which weight family this targets. Two different questions, two different answers. Within 9B (distilled vs base — the two CivitAI strings): not a real incompatibility. Both load through the same Flux2KleinPipeline and have the same repo layout; for the 4B pair, where both transformer configs are public, the distil" + ] + }, + "pros": [ + "Well established — 2,840 downloads. [source: CivitAI stats]", + "Positively received — 128 thumbs-up with no down-votes. [source: CivitAI stats]" + ], + "cons": [ + "Author references NSFW output — review before using in shared work. [source: description]", + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Large at 948 MB — slower to load and heavier to stack. [source: file size]" + ], + "file": { + "name": "AniEdit9B_v2.safetensors", + "size_bytes": 994079944, + "size_mb": 948.03, + "sha256": "d37d58268f6e704f7d5d5daf3cfc4d69bc82189c71e17494b5e02b6892274927" + }, + "stats": { + "downloads": 2840, + "thumbs_up": 128, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 2840, + "thumbs_up": 128, + "thumbs_down": 0, + "comments": 15, + "approval": "100% positive (128/128)", + "age_days": 217.8, + "downloads_per_day": 8.89, + "rank_by_downloads": 103, + "rank_by_momentum": 68, + "of_total": 200, + "adoption_tier": "Modest", + "momentum": "Steady" + }, + "selection_score": 71.3, + "selection_reasons": [ + "2,840 downloads", + "128 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "1056x1952", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "1216x832", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "1248x1664", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + } + ], + "tags": [ + "base model", + "edit", + "anime" + ], + "scraped_at": "2026-09-05T03:55:52+00:00", + "path": "flux2-klein-9b/aniedit-flux-2-klein" +} diff --git a/cg-texture-master-iray-skin-rendering-cg-3s/example_images/01_image.jpeg b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..58485b0959dd4ad76bea5bd9270bc09f1f3011df --- /dev/null +++ b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3bcd73d1be6905898f7defd4259f35ef457f9b3e05732d4753d5d786891cf32d +size 13382397 diff --git a/cg-texture-master-iray-skin-rendering-cg-3s/example_images/02_image.jpeg b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..fbf8a9e497f2d656860f52ef979f3d0989560cb7 --- /dev/null +++ b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3fe391c7bbf23ce0bccff6810f8175050895050ed32a048dfa92d84ba5898a3d +size 7454654 diff --git a/cg-texture-master-iray-skin-rendering-cg-3s/example_images/03_image.jpeg b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..9a2a1b8fad0ce277ac95ff28d994288a379ef6fa --- /dev/null +++ b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4fe74c2b9c69e8ce16c992372c3855162e4483a2318f6b9e0b3cba49b1850ee2 +size 1963976 diff --git a/cg-texture-master-iray-skin-rendering-cg-3s/example_images/04_image.jpeg b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..cba6649588dd6bf9b3bca42c8a3d55542e1cf131 --- /dev/null +++ b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04e1875a9b56bad0d74b501c22e07fbc4bc61f790b32d6973457770d66c6df0d +size 1561205 diff --git a/cg-texture-master-iray-skin-rendering-cg-3s/example_images/05_image.jpeg b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..b0c94da357b1346d95cdc2d64a9828215bed55c4 --- /dev/null +++ b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a38095ad4b5ec98de5ce10fe3b7da4b76985bc88f22b4886e419aeafc053832 +size 1737630 diff --git a/cg-texture-master-iray-skin-rendering-cg-3s/example_images/06_image.jpeg b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..cbd744fab9df65031f100ee7fea2779dc691fc16 --- /dev/null +++ b/cg-texture-master-iray-skin-rendering-cg-3s/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c0d6f074c703e02d466838bbbb583d48567558b4f107184c7ec161d3c27a583 +size 1480552 diff --git a/cg-texture-master-iray-skin-rendering-cg-3s/info.txt b/cg-texture-master-iray-skin-rendering-cg-3s/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..72e7582fd0434fc4de6e59ffff6bca8ee672860c --- /dev/null +++ b/cg-texture-master-iray-skin-rendering-cg-3s/info.txt @@ -0,0 +1,225 @@ +═══════════════════════════════════════════════════════════════════════════════ + CG TEXTURE MASTER & IRAY SKIN RENDERING | 极致CG质感与3S皮肤材质 +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model FLUX.2 [klein] 9B + Generation mode Text-to-image + Checkpoint FLUX.2-klein-9B (distilled) — CivitAI string "Flux.2 Klein 9B" + Category Visual style + Author nnegret + Source https://civitai.com/models/2388820 + Provenance CivitAI · model 2388820 · version 2686105 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File nnegrat_cgstyle_9b_1800.safetensors + Version Klein9B_V1 + Size 158.0 MB + Format SafeTensor / unknown precision + SHA256 d58f7c9ffb43e3c3cd9f72adbec905f06397218fd885aa86e4e90e96def8b980 + Published 2026-02-13 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + › "artwork in cgstyle_9b" + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + Taken verbatim from the author's description: + + Sampler euler + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 1,579 + Rating 100% positive (136/136) + Comments 0 + Published 204 days ago + Download rate 4.0/day since release + + Adoption rank #141 of 200 in this collection (Modest) + Momentum rank #141 of 200 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general FLUX.2 [klein] 9B ones, not anything specific to this LoRA. + + General FLUX.2 [klein] 9B rules that apply here (see REFERENCE.txt): + + • Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model + output into model input. There is no CLIP side to route — klein LoRAs + are model-only because the Qwen3 text encoder is not trained (CivitAI's + recipe fixes trainTextEncoder: false). + • Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B + LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, + 5+20 vs 8+24 blocks), not a quality question. + • Mixing a "Flux.2 Klein 9B" LoRA and a "Flux.2 Klein 9B-base" LoRA in + one stack is fine — those strings describe the training checkpoint, not + incompatible tensor layouts. Set the sampler for the checkpoint + actually loaded. + • Run base-trained LoRAs on the distilled checkpoint when you want speed: + HF's klein LoRA guide reports the adapter loads there and typically + gives better results than on base, at 4 steps. + • A base-to-turbo / distillation-extraction LoRA is a whole-model delta, + not a content LoRA. Put it on the base checkpoint only, and treat its + strength as a steps/CFG dial rather than a style knob. Do not add it on + top of the already-distilled checkpoint. + • Read each LoRA's own published strength range instead of defaulting + everything to 1.0 — klein LoRAs in the wild ship very different ones + (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, + under 0.5 and sometimes 0.1 for the vafipas663 distillation delta). + • With a consistency or edit LoRA in the stack, excessive strength + suppresses the prompt: the consistency LoRA's article says that if + edits stop applying, lower it before changing anything else. + • Prefer runtime LoRA patching over baking a LoRA into the checkpoint: + ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein + models (4B or 9B) since a March 2026 update — issue #13637, still open. + • Step-distilled to exactly 4 inference steps and guidance-distilled; HF + card example is num_inference_steps=4, guidance_scale=1.0. Its + transformer config has guidance_embeds=false. This is the intended + inference target: ComfyUI's blog benchmarks it at ~2s and 19.6GB VRAM + on an RTX 5090. Not offered as a training target by BFL's docs or + CivitAI's recipe. + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Well established — 1,579 downloads. [source: CivitAI stats] + + Positively received — 136 thumbs-up with no down-votes. [source: + CivitAI stats] + + Explicit trigger word(s) — artwork in cgstyle_9b — so the effect can be + turned on and off from the prompt. [source: model metadata] + + Compact at 158 MB — cheap to stack with other LoRAs. [source: file + size] + + Author published concrete recommended settings. [source: description] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + (none recorded) + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Two different questions, two different answers. Within 9B (distilled vs + base — the two CivitAI strings): not a real incompatibility. Both load + through the same Flux2KleinPipeline and have the same repo layout; for the + 4B pair, where both transformer configs are public, the distilled and base + configs are byte-for-byte identical in every shape field, and the 9B pair + is built the same way. HF's own klein LoRA guide states the workflow + directly: "Distilled models are step-compressed for fast inference; you + train against the base checkpoint and the adapter still loads on the + distilled model afterward", and "Applying the LoRA on the distilled model + typically gives better results than the base model, and it's faster" (that + post is written against 4B, but the family structure is the same). So the + two CivitAI strings record which checkpoint the author trained or previewed + against, not a hard compatibility wall. The real risk is a silent settings + mismatch: load a base-trained LoRA on the distilled checkpoint while + keeping the author's 20-50 step / CFG ~4 preview settings and the output + degrades badly (distilled wants 4 steps / guidance 1.0), or the reverse. + Nothing errors — it just looks wrong. Across sizes (9B vs 4B): a genuine + architectural mismatch, from the published transformer configs — different + hidden width (4096 vs 3072), different block counts (8+24 vs 5+20) and a + different text-conditioning dim (12288 vs 7680, following Qwen3-8B vs + Qwen3-4B). A 4B LoRA cannot patch a 9B model: expect missing keys and shape + errors, not degraded output. The exact ComfyUI error text is unverified. KV + variant: same 9B size and same 4-step distillation, but a distinct + checkpoint with its own pipeline; whether standard klein 9B LoRAs patch it + cleanly is not documented — see unverified. + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + 交流Q群：173273201 核心特点 / Key Features - Iray Style Rendering: + 带来真实的物理光影感，画面通透且富有深度。 / Provides a realistic physical sense of light and + shadow, making the image transparent and deep. + + - Ultimate Skin Texture: 重点优化了皮肤的 3S 材质，表现出通透、红润且极其光滑的皮肤触感。 / Specifically + optimized for SSS materials, showcasing translucent, rosy, and extremely + smooth skin. + + - Versatile Transformation: 无论是动漫转 CG 还是真人转 CG，都能保持极高的风格一致性。 / Whether it's + Anime-to-CG or Realistic-to-CG, it maintains high style consistency. + + - Multi-Scenario Support: 完美支持文生图 (T2I) 和图生图 (I2I)。 / Fully compatible with + both Text-to-Image and Image-to-Image workflows. + + 使用建议 / Usage Recommendations - 权重 / LoRA Weight: * Range: 0.75 - 0.9 + - Recommended: 0.85 + + - 采样器 / Sampler: er_sde （stabilize） euler (dynamic) + + 示例展示 / Gallery Tips + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 2688x3456) + 02_image.jpeg (image, 1664x2944) + 03_image.jpeg (image, 1152x1018) + 04_image.jpeg (image, 832x1216) + 05_image.jpeg (image, 832x1216) + 06_image.jpeg (image, 832x1216) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + 1. Diffusion model -> ComfyUI/models/diffusion_models/. Distilled: + flux-2-klein-9b-fp8.safetensors. Base: flux-2-klein- + base-9b-fp8.safetensors. (BF16 single-file weights + flux-2-klein-9b.safetensors / flux-2-klein-base-9b.safetensors from the + gated HF repos also work if you have ~29GB; community GGUF builds exist + from unsloth and leejet for lower-VRAM setups.) + 2. Text encoder -> + ComfyUI/models/text_encoders/qwen_3_8b_fp8mixed.safetensors (from Comfy- + Org/vae-text-encorder-for-flux-klein-9b). This is the 9B family's encoder; + the 4B family uses qwen_3_4b.safetensors — do not mix them. + 3. VAE -> ComfyUI/models/vae/flux2-vae.safetensors (shared across klein + sizes). + 4. LoRA -> ComfyUI/models/loras/ as a plain .safetensors. + 5. Load with Load Diffusion Model (UNETLoader) for the transformer, + CLIPLoader for the Qwen3 text encoder, VAELoader for the VAE. + docs.comfy.org ships prebuilt Flux.2 klein graphs under Workflow Templates + for both the distilled and base branches — start from those rather than + wiring by hand, since neither the docs nor BFL's repo state + sampler/scheduler/shift in prose; those values live in the template. + 6. Insert LoraLoaderModelOnly between the diffusion-model loader and the + sampler. Model-only is correct for klein: the Qwen3 text encoder is not + trained by the standard klein trainers, so there is no text-encoder half to + patch. + 7. Set steps and guidance to match the checkpoint you actually loaded, not + the one the LoRA was trained on: 4 steps / guidance 1.0 on distilled, + ~20-50 steps / guidance ~4.0 on base. + 8. Note the 9B repos are gated on HuggingFace — you must accept the FLUX + Non-Commercial License on the repo page before any download, including + scripted ones. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 03:55 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/cg-texture-master-iray-skin-rendering-cg-3s/metadata.json b/cg-texture-master-iray-skin-rendering-cg-3s/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..5e53e66e8c02d3c1913756d5bd6a47709506407c --- /dev/null +++ b/cg-texture-master-iray-skin-rendering-cg-3s/metadata.json @@ -0,0 +1,139 @@ +{ + "name": "CG Texture Master & Iray Skin Rendering | 极致CG质感与3S皮肤材质", + "slug": "cg-texture-master-iray-skin-rendering-cg-3s", + "base_model": "FLUX.2 [klein] 9B", + "media_type": "image", + "model_id": 2388820, + "version_id": 2686105, + "version_name": "Klein9B_V1", + "source": "civitai", + "url": "https://civitai.com/models/2388820", + "repo_id": null, + "file_path": null, + "author": "nnegret", + "category": { + "key": "style", + "label": "Visual style" + }, + "modalities": [ + "t2i" + ], + "primary_modality": "t2i", + "checkpoint_family": "klein-9b-distilled", + "trigger_words": [ + "artwork in cgstyle_9b" + ], + "recommended_settings": { + "Sampler": "euler" + }, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model output into model input. There is no CLIP side to route — klein LoRAs are model-only because the Qwen3 text encoder is not trained (CivitAI's recipe fixes trainTextEncoder: false).", + "Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, 5+20 vs 8+24 blocks), not a quality question.", + "Mixing a \"Flux.2 Klein 9B\" LoRA and a \"Flux.2 Klein 9B-base\" LoRA in one stack is fine — those strings describe the training checkpoint, not incompatible tensor layouts. Set the sampler for the checkpoint actually loaded.", + "Run base-trained LoRAs on the distilled checkpoint when you want speed: HF's klein LoRA guide reports the adapter loads there and typically gives better results than on base, at 4 steps.", + "A base-to-turbo / distillation-extraction LoRA is a whole-model delta, not a content LoRA. Put it on the base checkpoint only, and treat its strength as a steps/CFG dial rather than a style knob. Do not add it on top of the already-distilled checkpoint.", + "Read each LoRA's own published strength range instead of defaulting everything to 1.0 — klein LoRAs in the wild ship very different ones (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, under 0.5 and sometimes 0.1 for the vafipas663 distillation delta).", + "With a consistency or edit LoRA in the stack, excessive strength suppresses the prompt: the consistency LoRA's article says that if edits stop applying, lower it before changing anything else.", + "Prefer runtime LoRA patching over baking a LoRA into the checkpoint: ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein models (4B or 9B) since a March 2026 update — issue #13637, still open.", + "Step-distilled to exactly 4 inference steps and guidance-distilled; HF card example is num_inference_steps=4, guidance_scale=1.0. Its transformer config has guidance_embeds=false. This is the intended inference target: ComfyUI's blog benchmarks it at ~2s and 19.6GB VRAM on an RTX 5090. Not offered as a training target by BFL's docs or CivitAI's recipe." + ] + }, + "pros": [ + "Well established — 1,579 downloads. [source: CivitAI stats]", + "Positively received — 136 thumbs-up with no down-votes. [source: CivitAI stats]", + "Explicit trigger word(s) — artwork in cgstyle_9b — so the effect can be turned on and off from the prompt. [source: model metadata]", + "Compact at 158 MB — cheap to stack with other LoRAs. [source: file size]", + "Author published concrete recommended settings. [source: description]" + ], + "cons": [], + "file": { + "name": "nnegrat_cgstyle_9b_1800.safetensors", + "size_bytes": 165704168, + "size_mb": 158.03, + "sha256": "d58f7c9ffb43e3c3cd9f72adbec905f06397218fd885aa86e4e90e96def8b980" + }, + "stats": { + "downloads": 1579, + "thumbs_up": 136, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 1579, + "thumbs_up": 136, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (136/136)", + "age_days": 203.6, + "downloads_per_day": 4.0, + "rank_by_downloads": 141, + "rank_by_momentum": 141, + "of_total": 200, + "adoption_tier": "Modest", + "momentum": "Long tail" + }, + "selection_score": 70.9, + "selection_reasons": [ + "1,579 downloads", + "136 thumbs-up", + "documented triggers + usage notes", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "2688x3456", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "1664x2944", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1152x1018", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + } + ], + "tags": [ + "1 girl", + "realistic", + "anime", + "2.5d", + "cg", + "style" + ], + "scraped_at": "2026-09-05T03:55:12+00:00", + "path": "flux2-klein-9b/cg-texture-master-iray-skin-rendering-cg-3s" +} diff --git a/cg-texture-master-iray-skin-rendering-cg-3s/nnegrat_cgstyle_9b_1800.safetensors b/cg-texture-master-iray-skin-rendering-cg-3s/nnegrat_cgstyle_9b_1800.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4aa99e95cfb1cfa7f486f9cb6830d17facc8dbd6 --- /dev/null +++ b/cg-texture-master-iray-skin-rendering-cg-3s/nnegrat_cgstyle_9b_1800.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d58f7c9ffb43e3c3cd9f72adbec905f06397218fd885aa86e4e90e96def8b980 +size 165704168 diff --git a/claire-redfield-resident-evil-2-remake-in-real-life/claireredfieldtailor_flux2-klein_v1a.safetensors b/claire-redfield-resident-evil-2-remake-in-real-life/claireredfieldtailor_flux2-klein_v1a.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4f79643d6dffebed17b7b03d875bfa97e889c145 --- /dev/null +++ b/claire-redfield-resident-evil-2-remake-in-real-life/claireredfieldtailor_flux2-klein_v1a.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a08cadb74bdf244d11d647ee802f969b34f56631dc3d0cf5d7fac19de26f74a4 +size 165704384 diff --git a/claire-redfield-resident-evil-2-remake-in-real-life/example_images/01_image.jpeg b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..ca5f2b11c91705da04e4cf356722d33609b53305 --- /dev/null +++ b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:724ac304ffad3b2a5100ea7a30afabcf432f02f859afa083c5baffd772ad396e +size 1424262 diff --git a/claire-redfield-resident-evil-2-remake-in-real-life/example_images/02_image.jpeg b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..dee438ec589b67719fe697591e8edd1bafa98943 --- /dev/null +++ b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:35611aedfbfd98f77ce204c1eb2685b0fa712c8f23768c273c923fc544101c01 +size 1335036 diff --git a/claire-redfield-resident-evil-2-remake-in-real-life/example_images/03_image.jpeg b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..937bc94af3231f4cdb001a2d2624f77435253650 --- /dev/null +++ b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:baf765ff62b9ba8d65eb87f7fbedbd9028f718e8efbcb3875be29d87c0422552 +size 1082876 diff --git a/claire-redfield-resident-evil-2-remake-in-real-life/example_images/04_image.jpeg b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..43be76c490abfe1ef361a492820dc2b664f67d62 --- /dev/null +++ b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:523ed9b39ce2c569e18e7ca253315d8b0483f79437f76a00d4c266fc5f42c386 +size 1270973 diff --git a/claire-redfield-resident-evil-2-remake-in-real-life/example_images/05_image.jpeg b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..b6db109f2c34c44f08e47555a9f2cc939c4ffe06 --- /dev/null +++ b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f80a47a0645bc16d8a463a23bf0539c28482bb19ea3fd1c410a2a3331af8ee27 +size 1364268 diff --git a/claire-redfield-resident-evil-2-remake-in-real-life/example_images/06_image.jpeg b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..67603ff8c0abf28992081576880e61f3df8f75fc --- /dev/null +++ b/claire-redfield-resident-evil-2-remake-in-real-life/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:853b5689ec4e82d6cd30eea9e1e153c770e90db7b180962f78b1e96ad58ac28d +size 174316 diff --git a/claire-redfield-resident-evil-2-remake-in-real-life/info.txt b/claire-redfield-resident-evil-2-remake-in-real-life/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..d2f54770bff7a3881c75a85a85b66502a6fdebef --- /dev/null +++ b/claire-redfield-resident-evil-2-remake-in-real-life/info.txt @@ -0,0 +1,295 @@ +═══════════════════════════════════════════════════════════════════════════════ + CLAIRE REDFIELD - RESIDENT EVIL 2 REMAKE IN REAL-LIFE +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model FLUX.2 [klein] 9B + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Character / subject + Author SteveTailor + Source https://civitai.com/models/2850469 + Provenance CivitAI · model 2850469 · version 3227435 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File claireredfieldtailor_flux2-klein_v1a.safetensors + Version Flux.2 Klein 9B - v1.0A + Size 158.0 MB + Format SafeTensor / unknown precision + SHA256 a08cadb74bdf244d11d647ee802f969b34f56631dc3d0cf5d7fac19de26f74a4 + Published 2026-08-15 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + › "claireredfieldtailor" + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + Taken verbatim from the author's description: + + LoRA strength 1.0 + Sampler steps 9 + CFG / guidance 1.0 + Sampler res_multistep + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 228 + Rating 100% positive (21/21) + Comments 0 + Published 21 days ago + Download rate 2.7/day since release + + Adoption rank #197 of 200 in this collection (Niche) + Momentum rank #183 of 200 by download rate (Long tail · fresh release) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general FLUX.2 [klein] 9B ones, not anything specific to this LoRA. + + General FLUX.2 [klein] 9B rules that apply here (see REFERENCE.txt): + + • Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model + output into model input. There is no CLIP side to route — klein LoRAs + are model-only because the Qwen3 text encoder is not trained (CivitAI's + recipe fixes trainTextEncoder: false). + • Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B + LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, + 5+20 vs 8+24 blocks), not a quality question. + • Mixing a "Flux.2 Klein 9B" LoRA and a "Flux.2 Klein 9B-base" LoRA in + one stack is fine — those strings describe the training checkpoint, not + incompatible tensor layouts. Set the sampler for the checkpoint + actually loaded. + • Run base-trained LoRAs on the distilled checkpoint when you want speed: + HF's klein LoRA guide reports the adapter loads there and typically + gives better results than on base, at 4 steps. + • A base-to-turbo / distillation-extraction LoRA is a whole-model delta, + not a content LoRA. Put it on the base checkpoint only, and treat its + strength as a steps/CFG dial rather than a style knob. Do not add it on + top of the already-distilled checkpoint. + • Read each LoRA's own published strength range instead of defaulting + everything to 1.0 — klein LoRAs in the wild ship very different ones + (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, + under 0.5 and sometimes 0.1 for the vafipas663 distillation delta). + • With a consistency or edit LoRA in the stack, excessive strength + suppresses the prompt: the consistency LoRA's article says that if + edits stop applying, lower it before changing anything else. + • Prefer runtime LoRA patching over baking a LoRA into the checkpoint: + ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein + models (4B or 9B) since a March 2026 update — issue #13637, still open. + • The author never states which weight family this targets. Two different + questions, two different answers. Within 9B (distilled vs base — the + two CivitAI strings): not a real incompatibility. Both load through the + same Flux2KleinPipeline and have the same repo layout; for the 4B pair, + where both transformer configs are public, the distil + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Explicit trigger word(s) — claireredfieldtailor — so the effect can be + turned on and off from the prompt. [source: model metadata] + + Compact at 158 MB — cheap to stack with other LoRAs. [source: file + size] + + Author published concrete recommended settings. [source: description] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Modest adoption (228 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Two different questions, two different answers. Within 9B (distilled vs + base — the two CivitAI strings): not a real incompatibility. Both load + through the same Flux2KleinPipeline and have the same repo layout; for the + 4B pair, where both transformer configs are public, the distilled and base + configs are byte-for-byte identical in every shape field, and the 9B pair + is built the same way. HF's own klein LoRA guide states the workflow + directly: "Distilled models are step-compressed for fast inference; you + train against the base checkpoint and the adapter still loads on the + distilled model afterward", and "Applying the LoRA on the distilled model + typically gives better results than the base model, and it's faster" (that + post is written against 4B, but the family structure is the same). So the + two CivitAI strings record which checkpoint the author trained or previewed + against, not a hard compatibility wall. The real risk is a silent settings + mismatch: load a base-trained LoRA on the distilled checkpoint while + keeping the author's 20-50 step / CFG ~4 preview settings and the output + degrades badly (distilled wants 4 steps / guidance 1.0), or the reverse. + Nothing errors — it just looks wrong. Across sizes (9B vs 4B): a genuine + architectural mismatch, from the published transformer configs — different + hidden width (4096 vs 3072), different block counts (8+24 vs 5+20) and a + different text-conditioning dim (12288 vs 7680, following Qwen3-8B vs + Qwen3-4B). A 4B LoRA cannot patch a 9B model: expect missing keys and shape + errors, not degraded output. The exact ComfyUI error text is unverified. KV + variant: same 9B size and same 4-step distillation, but a distinct + checkpoint with its own pipeline; whether standard klein 9B LoRAs patch it + cleanly is not documented — see unverified. + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Meet Claire Redfield — the one who rode into Raccoon City looking for her + brother and never quite stopped moving since. This LoRA is a fan tribute + built around her likeness: her face, that level, unimpressed calm she keeps + even when everything around her is on fire, and the warmth underneath it. + One thing up front, because it changes how you prompt her: this is a + character-identity LoRA. It reproduces who she is, not a fixed costume. The + biker jacket, the red vest, the Raccoon City era gear — none of that is + baked in. The dataset is deliberately varied everyday clothing and + locations, so her identity holds up in any setting. If you want the + signature look, describe it in your prompt; the LoRA keeps her face while + you decide the wardrobe. + It is tuned for Z-Image Turbo , so you get a photorealistic Claire at turbo + speed: few steps, low CFG, fast iterations, strong identity. There are also + versions for Illustrious and FLUX.2 Klein 9B . Have fun with her — and + please share what you make. + Recommended Settings - Trigger word: claireredfieldtailor (put it at the + very start of your prompt) + + - Strength: 1.0 (useful range 0.8 – 1.0) + + Z-Image-Turbo - Steps: 9 + + - CFG: 1.0 + + - Sampler: res_multistep + + - Scheduler: simple + + - Shift: 3.0 (via a ModelSamplingAuraFlow node — Z-Image / AuraFlow + specific) + + - Negative prompt: (empty) + + - Base model: Z-Image Turbo. A realistic Z-Image Turbo checkpoint such as + AnalogMadness (Z-Image Turbo) v2.0 or cyberrealistic Z-Image v0.5 gives the + most photographic look. + + Illustrious - Steps: 30 + + - CFG: 4.0-5.0 + + - Sampler: dpmpp_2m, dpmpp_3m_sde + + - Scheduler: karras + + - Models: Realism Illustrious By Stable Yogi, Ultra Realistic By Stable + Yogi, ramthrustsNSFWPINK_alchemyMix176, perfectdeliberate v70, iLustMix + + Flux.2 Klein 9B - Steps: 25-30 + + - CFG: 1.0 + + - Guidance: 3.0-4.0 + + - Sampler: euler / res_multistep + + - Scheduler: simple + + - Negative prompt: (empty) + + - Base model: Flux.2 Klein 9B + FP8, Jib Mix Klein, + + Prompt Examples + Prompt Examples Portrait prompt: A photo of claireredfieldtailor, a woman. + A half-body portrait, three-quarter view, soft window light, warm tones, + shallow depth of field, photorealistic, professional photography. Meaning: + just her, cleanly lit, nothing competing with the face. Use this to verify + the likeness + before you build a bigger scene. + + Training Details - Source: curated character dataset (portraits, full-body, + scenes, outside, inside, ...) + + - Dataset size: 89 images + + - Resolution: 768 px + + Notes - Fan service / disclaimer: This is a fan-made tribute LoRA. It is + not affiliated with, endorsed by, or connected to Capcom or any Resident + Evil rights holder. Claire Redfield is a fictional character owned by + Capcom; this LoRA reproduces the likeness of a fictional character, not of + a real person. Check Civitai's rules and Capcom's position before enabling + commercial use or on-site generation. For personal, non-commercial fan use. + + - Identity, not a costume: the biker jacket, the red vest and the rest of + the game wardrobe were not in the training data. Describe the clothing you + want in your prompt — the LoRA supplies the face. Any accessory has to be + prompted and only makes sense in front or side views. + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 896x1152) + 02_image.jpeg (image, 896x1152) + 03_image.jpeg (image, 896x1152) + 04_image.jpeg (image, 896x1152) + 05_image.jpeg (image, 896x1152) + 06_image.jpeg (image, 832x1216) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + 1. Diffusion model -> ComfyUI/models/diffusion_models/. Distilled: + flux-2-klein-9b-fp8.safetensors. Base: flux-2-klein- + base-9b-fp8.safetensors. (BF16 single-file weights + flux-2-klein-9b.safetensors / flux-2-klein-base-9b.safetensors from the + gated HF repos also work if you have ~29GB; community GGUF builds exist + from unsloth and leejet for lower-VRAM setups.) + 2. Text encoder -> + ComfyUI/models/text_encoders/qwen_3_8b_fp8mixed.safetensors (from Comfy- + Org/vae-text-encorder-for-flux-klein-9b). This is the 9B family's encoder; + the 4B family uses qwen_3_4b.safetensors — do not mix them. + 3. VAE -> ComfyUI/models/vae/flux2-vae.safetensors (shared across klein + sizes). + 4. LoRA -> ComfyUI/models/loras/ as a plain .safetensors. + 5. Load with Load Diffusion Model (UNETLoader) for the transformer, + CLIPLoader for the Qwen3 text encoder, VAELoader for the VAE. + docs.comfy.org ships prebuilt Flux.2 klein graphs under Workflow Templates + for both the distilled and base branches — start from those rather than + wiring by hand, since neither the docs nor BFL's repo state + sampler/scheduler/shift in prose; those values live in the template. + 6. Insert LoraLoaderModelOnly between the diffusion-model loader and the + sampler. Model-only is correct for klein: the Qwen3 text encoder is not + trained by the standard klein trainers, so there is no text-encoder half to + patch. + 7. Set steps and guidance to match the checkpoint you actually loaded, not + the one the LoRA was trained on: 4 steps / guidance 1.0 on distilled, + ~20-50 steps / guidance ~4.0 on base. + 8. Note the 9B repos are gated on HuggingFace — you must accept the FLUX + Non-Commercial License on the repo page before any download, including + scripted ones. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 03:55 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/claire-redfield-resident-evil-2-remake-in-real-life/metadata.json b/claire-redfield-resident-evil-2-remake-in-real-life/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..ee3922abebdd0898e2108deffcdec4af164dc48d --- /dev/null +++ b/claire-redfield-resident-evil-2-remake-in-real-life/metadata.json @@ -0,0 +1,142 @@ +{ + "name": "Claire Redfield - Resident Evil 2 Remake in Real-Life", + "slug": "claire-redfield-resident-evil-2-remake-in-real-life", + "base_model": "FLUX.2 [klein] 9B", + "media_type": "image", + "model_id": 2850469, + "version_id": 3227435, + "version_name": "Flux.2 Klein 9B - v1.0A", + "source": "civitai", + "url": "https://civitai.com/models/2850469", + "repo_id": null, + "file_path": null, + "author": "SteveTailor", + "category": { + "key": "character", + "label": "Character / subject" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [ + "claireredfieldtailor" + ], + "recommended_settings": { + "LoRA strength": "1.0", + "Sampler steps": "9", + "CFG / guidance": "1.0", + "Sampler": "res_multistep" + }, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model output into model input. There is no CLIP side to route — klein LoRAs are model-only because the Qwen3 text encoder is not trained (CivitAI's recipe fixes trainTextEncoder: false).", + "Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, 5+20 vs 8+24 blocks), not a quality question.", + "Mixing a \"Flux.2 Klein 9B\" LoRA and a \"Flux.2 Klein 9B-base\" LoRA in one stack is fine — those strings describe the training checkpoint, not incompatible tensor layouts. Set the sampler for the checkpoint actually loaded.", + "Run base-trained LoRAs on the distilled checkpoint when you want speed: HF's klein LoRA guide reports the adapter loads there and typically gives better results than on base, at 4 steps.", + "A base-to-turbo / distillation-extraction LoRA is a whole-model delta, not a content LoRA. Put it on the base checkpoint only, and treat its strength as a steps/CFG dial rather than a style knob. Do not add it on top of the already-distilled checkpoint.", + "Read each LoRA's own published strength range instead of defaulting everything to 1.0 — klein LoRAs in the wild ship very different ones (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, under 0.5 and sometimes 0.1 for the vafipas663 distillation delta).", + "With a consistency or edit LoRA in the stack, excessive strength suppresses the prompt: the consistency LoRA's article says that if edits stop applying, lower it before changing anything else.", + "Prefer runtime LoRA patching over baking a LoRA into the checkpoint: ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein models (4B or 9B) since a March 2026 update — issue #13637, still open.", + "The author never states which weight family this targets. Two different questions, two different answers. Within 9B (distilled vs base — the two CivitAI strings): not a real incompatibility. Both load through the same Flux2KleinPipeline and have the same repo layout; for the 4B pair, where both transformer configs are public, the distil" + ] + }, + "pros": [ + "Explicit trigger word(s) — claireredfieldtailor — so the effect can be turned on and off from the prompt. [source: model metadata]", + "Compact at 158 MB — cheap to stack with other LoRAs. [source: file size]", + "Author published concrete recommended settings. [source: description]" + ], + "cons": [ + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Modest adoption (228 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "claireredfieldtailor_flux2-klein_v1a.safetensors", + "size_bytes": 165704384, + "size_mb": 158.03, + "sha256": "a08cadb74bdf244d11d647ee802f969b34f56631dc3d0cf5d7fac19de26f74a4" + }, + "stats": { + "downloads": 228, + "thumbs_up": 21, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 228, + "thumbs_up": 21, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (21/21)", + "age_days": 20.8, + "downloads_per_day": 2.69, + "rank_by_downloads": 197, + "rank_by_momentum": 183, + "of_total": 200, + "adoption_tier": "Niche", + "momentum": "Long tail · fresh release" + }, + "selection_score": 70.9, + "selection_reasons": [ + "228 downloads", + "21 thumbs-up", + "documented triggers + usage notes", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "896x1152", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "896x1152", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "896x1152", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "896x1152", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "896x1152", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + } + ], + "tags": [ + "claire redfield", + "character", + "resi2", + "resident evil 2 remake", + "resident evil 2", + "claire", + "redfield" + ], + "scraped_at": "2026-09-05T03:55:48+00:00", + "path": "flux2-klein-9b/claire-redfield-resident-evil-2-remake-in-real-life" +} diff --git a/detailer-ce/Detailer04d_CE_Flux2K9b_AIT_000000150.safetensors b/detailer-ce/Detailer04d_CE_Flux2K9b_AIT_000000150.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d772fe8d495fa720b83fe4048e21eaac7f518a62 --- /dev/null +++ b/detailer-ce/Detailer04d_CE_Flux2K9b_AIT_000000150.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ebd9c7088864987efc067f6925dce369c8fe6964137aff044a5d4c8b6650bfd9 +size 41447688 diff --git a/detailer-ce/example_images/01_image.jpeg b/detailer-ce/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..375e764c4ea81fa255119e8bfbbbf6fef0682fd1 --- /dev/null +++ b/detailer-ce/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7008d4f6b47cf166380aaea170fe936829aac2883831a11519d9cdf0f5137b91 +size 7735109 diff --git a/detailer-ce/example_images/02_image.jpeg b/detailer-ce/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..d113db5458291bd53b97381ed57f3099a8385256 --- /dev/null +++ b/detailer-ce/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d765dd08c401d605c5f06ba5a75e0c8f1317212b94645b6ec2c56f4ed83e3ac9 +size 8343101 diff --git a/detailer-ce/example_images/03_image.jpeg b/detailer-ce/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..d1b645181682a8cbc0a98bbd79f721384c400c9b --- /dev/null +++ b/detailer-ce/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:644eb52254bc3bbfbd720e949b8aaad1c651e980205f2605211f1804c6d088a5 +size 7844384 diff --git a/detailer-ce/example_images/04_image.jpeg b/detailer-ce/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..0e708e37f2ac34692ee77d2aa0ec0cebf77898c7 --- /dev/null +++ b/detailer-ce/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84b1e40b3d5872a252823d3a6de11876d38f79fc5215e7069b4928639d0e9f15 +size 2631674 diff --git a/detailer-ce/example_images/05_image.jpeg b/detailer-ce/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..edd05d62f24ebfa66dce228096d34bb447cdd6a7 --- /dev/null +++ b/detailer-ce/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3260f2006aefe847be6ebf22b2e87d08faac75508f7d3100d87476fd02c5e9c +size 2123732 diff --git a/detailer-ce/example_images/06_image.jpeg b/detailer-ce/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..32b3e3eeb062073288c06132feb9f169eaf1bc8b --- /dev/null +++ b/detailer-ce/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c657656324faa55110ae311fb4e49f98a46113079f853daf200338ced44c083d +size 2003141 diff --git a/detailer-ce/info.txt b/detailer-ce/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..3c423672e571454de2ff957ff83ced5b6e501c9d --- /dev/null +++ b/detailer-ce/info.txt @@ -0,0 +1,238 @@ +═══════════════════════════════════════════════════════════════════════════════ + DETAILER CE +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model FLUX.2 [klein] 9B + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Detail / anatomy fixer + Author CreativeEdge + Source https://civitai.com/models/2574468 + Provenance CivitAI · model 2574468 · version 2896490 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File Detailer04d_CE_Flux2K9b_AIT_000000150.safetensors + Version V04d - Flux.2 Klein 9b + Size 39.5 MB + Format SafeTensor / unknown precision + SHA256 ebd9c7088864987efc067f6925dce369c8fe6964137aff044a5d4c8b6650bfd9 + Published 2026-04-28 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps Distilled + FLUX.2-klein-9B: 4 (HF card example; it is step-distilled to exactly this, + and more steps do not refine further). Base FLUX.2-klein-base-9B: 50 in the + HF card example and in ComfyUI's benchmark; community practice runs lower, + with ~20 for production and 8-12 for fast iteration reported in the CivitAI + consistency-LoRA article., cfg Distilled: guidance_scale 1.0 (HF card). It + is guidance-distilled and its transformer config sets + guidance_embeds=false, so raising CFG degrades rather than improves. Base: + guidance_scale 4.0 (HF card); the CivitAI consistency-LoRA article + recommends 3.5-4.5 for most edits., sampler Not stated by BFL or by + docs.comfy.org, and the official flux2 GitHub repo documents no sampler, + scheduler or shift values either. Community consensus is euler with a + resolution-dependent shifted-Euler / flow-match schedule (ComfyUI's Flux2 + workflows use a scheduler that interpolates shift between base_shift and + max_shift by output resolution). Start from ComfyUI's downloadable Flux.2 + klein workflow template rather than a blog. See unverified., resolution + 1024x1024 in every BFL HF card example and in CivitAI's 9B training recipe + ("resolution": 1024). Non-square ratios work but no official list of + supported dimensions or a stated alignment multiple was found — see + unverified., lora_strength 1.0 is the diffusers/ComfyUI default and what + CivitAI applies to training previews (strength 1.0). Published per-LoRA + guidance varies and should be read from the LoRA, not assumed: the widely + used Klein 9B consistency LoRA's CivitAI article says start at 0.5 within a + 0.2-1.0 range (0.2-0.4 loose, 0.5-0.7 balanced, 0.8-1.0 maximum structural + stability) and warns that too high a strength stops edits from applying; + SOLRICKS' Realistic Detail LoRA card says 0.6-0.8 subtle, 0.8-1.0 stronger, + start 0.8; the vafipas663 distillation-delta LoRA card warns it "produces + static noise at strength > 0.5. And sometimes, you have to go as low as + 0.1.". + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 1,961 + Rating 100% positive (127/127) + Comments 0 + Published 129 days ago + Download rate 4.1/day since release + + Adoption rank #133 of 200 in this collection (Modest) + Momentum rank #136 of 200 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general FLUX.2 [klein] 9B ones, not anything specific to this LoRA. + + General FLUX.2 [klein] 9B rules that apply here (see REFERENCE.txt): + + • Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model + output into model input. There is no CLIP side to route — klein LoRAs + are model-only because the Qwen3 text encoder is not trained (CivitAI's + recipe fixes trainTextEncoder: false). + • Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B + LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, + 5+20 vs 8+24 blocks), not a quality question. + • Mixing a "Flux.2 Klein 9B" LoRA and a "Flux.2 Klein 9B-base" LoRA in + one stack is fine — those strings describe the training checkpoint, not + incompatible tensor layouts. Set the sampler for the checkpoint + actually loaded. + • Run base-trained LoRAs on the distilled checkpoint when you want speed: + HF's klein LoRA guide reports the adapter loads there and typically + gives better results than on base, at 4 steps. + • A base-to-turbo / distillation-extraction LoRA is a whole-model delta, + not a content LoRA. Put it on the base checkpoint only, and treat its + strength as a steps/CFG dial rather than a style knob. Do not add it on + top of the already-distilled checkpoint. + • Read each LoRA's own published strength range instead of defaulting + everything to 1.0 — klein LoRAs in the wild ship very different ones + (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, + under 0.5 and sometimes 0.1 for the vafipas663 distillation delta). + • With a consistency or edit LoRA in the stack, excessive strength + suppresses the prompt: the consistency LoRA's article says that if + edits stop applying, lower it before changing anything else. + • Prefer runtime LoRA patching over baking a LoRA into the checkpoint: + ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein + models (4B or 9B) since a March 2026 update — issue #13637, still open. + • The author never states which weight family this targets. Two different + questions, two different answers. Within 9B (distilled vs base — the + two CivitAI strings): not a real incompatibility. Both load through the + same Flux2KleinPipeline and have the same repo layout; for the 4B pair, + where both transformer configs are public, the distil + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Well established — 1,961 downloads. [source: CivitAI stats] + + Positively received — 127 thumbs-up with no down-votes. [source: + CivitAI stats] + + Compact at 40 MB — cheap to stack with other LoRAs. [source: file size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Needs a reduced strength; the default 1.0 is likely too strong. + [source: description] + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Two different questions, two different answers. Within 9B (distilled vs + base — the two CivitAI strings): not a real incompatibility. Both load + through the same Flux2KleinPipeline and have the same repo layout; for the + 4B pair, where both transformer configs are public, the distilled and base + configs are byte-for-byte identical in every shape field, and the 9B pair + is built the same way. HF's own klein LoRA guide states the workflow + directly: "Distilled models are step-compressed for fast inference; you + train against the base checkpoint and the adapter still loads on the + distilled model afterward", and "Applying the LoRA on the distilled model + typically gives better results than the base model, and it's faster" (that + post is written against 4B, but the family structure is the same). So the + two CivitAI strings record which checkpoint the author trained or previewed + against, not a hard compatibility wall. The real risk is a silent settings + mismatch: load a base-trained LoRA on the distilled checkpoint while + keeping the author's 20-50 step / CFG ~4 preview settings and the output + degrades badly (distilled wants 4 steps / guidance 1.0), or the reverse. + Nothing errors — it just looks wrong. Across sizes (9B vs 4B): a genuine + architectural mismatch, from the published transformer configs — different + hidden width (4096 vs 3072), different block counts (8+24 vs 5+20) and a + different text-conditioning dim (12288 vs 7680, following Qwen3-8B vs + Qwen3-4B). A 4B LoRA cannot patch a 9B model: expect missing keys and shape + errors, not degraded output. The exact ComfyUI error text is unverified. KV + variant: same 9B size and same 4-step distillation, but a distinct + checkpoint with its own pipeline; whether standard klein 9B LoRAs patch it + cleanly is not documented — see unverified. + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Detail slider. On its own, the useful range is anywhere from around -1 to + positive 1, though some of the models top out around +/- 0.8. See the + showcase images for an idea of the results at various settings. When used + in conjunction with other LoRAs, it can be more useful to lower the slider + strength to around 0.4. + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 3072x1024) + 02_image.jpeg (image, 3072x1024) + 03_image.jpeg (image, 3072x1024) + 04_image.jpeg (image, 1024x1024) + 05_image.jpeg (image, 1024x1024) + 06_image.jpeg (image, 1024x1024) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + 1. Diffusion model -> ComfyUI/models/diffusion_models/. Distilled: + flux-2-klein-9b-fp8.safetensors. Base: flux-2-klein- + base-9b-fp8.safetensors. (BF16 single-file weights + flux-2-klein-9b.safetensors / flux-2-klein-base-9b.safetensors from the + gated HF repos also work if you have ~29GB; community GGUF builds exist + from unsloth and leejet for lower-VRAM setups.) + 2. Text encoder -> + ComfyUI/models/text_encoders/qwen_3_8b_fp8mixed.safetensors (from Comfy- + Org/vae-text-encorder-for-flux-klein-9b). This is the 9B family's encoder; + the 4B family uses qwen_3_4b.safetensors — do not mix them. + 3. VAE -> ComfyUI/models/vae/flux2-vae.safetensors (shared across klein + sizes). + 4. LoRA -> ComfyUI/models/loras/ as a plain .safetensors. + 5. Load with Load Diffusion Model (UNETLoader) for the transformer, + CLIPLoader for the Qwen3 text encoder, VAELoader for the VAE. + docs.comfy.org ships prebuilt Flux.2 klein graphs under Workflow Templates + for both the distilled and base branches — start from those rather than + wiring by hand, since neither the docs nor BFL's repo state + sampler/scheduler/shift in prose; those values live in the template. + 6. Insert LoraLoaderModelOnly between the diffusion-model loader and the + sampler. Model-only is correct for klein: the Qwen3 text encoder is not + trained by the standard klein trainers, so there is no text-encoder half to + patch. + 7. Set steps and guidance to match the checkpoint you actually loaded, not + the one the LoRA was trained on: 4 steps / guidance 1.0 on distilled, + ~20-50 steps / guidance ~4.0 on base. + 8. Note the 9B repos are gated on HuggingFace — you must accept the FLUX + Non-Commercial License on the repo page before any download, including + scripted ones. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 03:54 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/detailer-ce/metadata.json b/detailer-ce/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..781dbb0b1f0d2e15d600725954b4f14b0018892e --- /dev/null +++ b/detailer-ce/metadata.json @@ -0,0 +1,132 @@ +{ + "name": "Detailer CE", + "slug": "detailer-ce", + "base_model": "FLUX.2 [klein] 9B", + "media_type": "image", + "model_id": 2574468, + "version_id": 2896490, + "version_name": "V04d - Flux.2 Klein 9b", + "source": "civitai", + "url": "https://civitai.com/models/2574468", + "repo_id": null, + "file_path": null, + "author": "CreativeEdge", + "category": { + "key": "detail", + "label": "Detail / anatomy fixer" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model output into model input. There is no CLIP side to route — klein LoRAs are model-only because the Qwen3 text encoder is not trained (CivitAI's recipe fixes trainTextEncoder: false).", + "Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, 5+20 vs 8+24 blocks), not a quality question.", + "Mixing a \"Flux.2 Klein 9B\" LoRA and a \"Flux.2 Klein 9B-base\" LoRA in one stack is fine — those strings describe the training checkpoint, not incompatible tensor layouts. Set the sampler for the checkpoint actually loaded.", + "Run base-trained LoRAs on the distilled checkpoint when you want speed: HF's klein LoRA guide reports the adapter loads there and typically gives better results than on base, at 4 steps.", + "A base-to-turbo / distillation-extraction LoRA is a whole-model delta, not a content LoRA. Put it on the base checkpoint only, and treat its strength as a steps/CFG dial rather than a style knob. Do not add it on top of the already-distilled checkpoint.", + "Read each LoRA's own published strength range instead of defaulting everything to 1.0 — klein LoRAs in the wild ship very different ones (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, under 0.5 and sometimes 0.1 for the vafipas663 distillation delta).", + "With a consistency or edit LoRA in the stack, excessive strength suppresses the prompt: the consistency LoRA's article says that if edits stop applying, lower it before changing anything else.", + "Prefer runtime LoRA patching over baking a LoRA into the checkpoint: ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein models (4B or 9B) since a March 2026 update — issue #13637, still open.", + "The author never states which weight family this targets. Two different questions, two different answers. Within 9B (distilled vs base — the two CivitAI strings): not a real incompatibility. Both load through the same Flux2KleinPipeline and have the same repo layout; for the 4B pair, where both transformer configs are public, the distil" + ] + }, + "pros": [ + "Well established — 1,961 downloads. [source: CivitAI stats]", + "Positively received — 127 thumbs-up with no down-votes. [source: CivitAI stats]", + "Compact at 40 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "Needs a reduced strength; the default 1.0 is likely too strong. [source: description]", + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]" + ], + "file": { + "name": "Detailer04d_CE_Flux2K9b_AIT_000000150.safetensors", + "size_bytes": 41447688, + "size_mb": 39.53, + "sha256": "ebd9c7088864987efc067f6925dce369c8fe6964137aff044a5d4c8b6650bfd9" + }, + "stats": { + "downloads": 1961, + "thumbs_up": 127, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 1961, + "thumbs_up": 127, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (127/127)", + "age_days": 129.4, + "downloads_per_day": 4.12, + "rank_by_downloads": 133, + "rank_by_momentum": 136, + "of_total": 200, + "adoption_tier": "Modest", + "momentum": "Long tail" + }, + "selection_score": 71.3, + "selection_reasons": [ + "1,961 downloads", + "127 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "3072x1024", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "3072x1024", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "3072x1024", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + } + ], + "tags": [ + "detail slider", + "detail", + "style", + "slider" + ], + "scraped_at": "2026-09-05T03:54:59+00:00", + "path": "flux2-klein-9b/detailer-ce" +} diff --git a/hand-drawn-sketch/example_images/01_image.jpeg b/hand-drawn-sketch/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..9ebc156ab4e79c3b3c5761c498ab7889e42b34bf --- /dev/null +++ b/hand-drawn-sketch/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:234849c7317308b0470ee3a5c751ee5a6156abf6bebdb51693bd7ceec7601488 +size 1429275 diff --git a/hand-drawn-sketch/example_images/02_image.jpeg b/hand-drawn-sketch/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..41d224559c84c7d7a490df444ee4c1bf7dbf1e6b --- /dev/null +++ b/hand-drawn-sketch/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ede42cfccb4ffabd2009cd924fdefba2b6568e05e60db46ca6d41c8816537c4b +size 1079364 diff --git a/hand-drawn-sketch/example_images/03_image.jpeg b/hand-drawn-sketch/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..808261d682a73375c10dff3fedbf58da85a66b78 --- /dev/null +++ b/hand-drawn-sketch/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3a3e84e7acd3216c18d312bc32361e3a1fa72221981dc462205ae257c153506 +size 1381485 diff --git a/hand-drawn-sketch/example_images/04_image.jpeg b/hand-drawn-sketch/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..f8fa1ba02dfdc05cbbe8c5fb967b1a6949f36364 --- /dev/null +++ b/hand-drawn-sketch/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae37ffb03afeef72ef487b842542b070a1476e0c7fd6277e2fd1d0f724e68d66 +size 1445630 diff --git a/hand-drawn-sketch/example_images/05_image.jpeg b/hand-drawn-sketch/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..166d50a0fcf76d7250b22aae495e5f46f1d6cad0 --- /dev/null +++ b/hand-drawn-sketch/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:638f45ddac53db4702b2de07a841532526241476cea17be1342e3ea246ed3da0 +size 1757976 diff --git a/hand-drawn-sketch/example_images/06_image.jpeg b/hand-drawn-sketch/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..ccc998f2f39a1f89c35d06623fbc26f18e0ed29a --- /dev/null +++ b/hand-drawn-sketch/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d18c5d4ba82049e278510bbd6ac868b959f6a47973581c175c1a2723cc01de85 +size 1414860 diff --git a/hand-drawn-sketch/hand_drawn_sketch_klein9b.safetensors b/hand-drawn-sketch/hand_drawn_sketch_klein9b.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5178b83428dd4744a52994a086f05d91bee7de6f --- /dev/null +++ b/hand-drawn-sketch/hand_drawn_sketch_klein9b.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96b1ce30cadd42bc15bb457117814dd0d31e19172e236f3038fc877129e869aa +size 82879480 diff --git a/hand-drawn-sketch/info.txt b/hand-drawn-sketch/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..889f0e282f9241dc376c9f87aef07f153d7ed64a --- /dev/null +++ b/hand-drawn-sketch/info.txt @@ -0,0 +1,217 @@ +═══════════════════════════════════════════════════════════════════════════════ + HAND-DRAWN SKETCH +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model FLUX.2 [klein] 9B + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Visual style + Author Yofaraway + Source https://civitai.com/models/2692445 + Provenance CivitAI · model 2692445 · version 3023139 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File hand_drawn_sketch_klein9b.safetensors + Version Klein9B v1 + Size 79.0 MB + Format SafeTensor / unknown precision + SHA256 96b1ce30cadd42bc15bb457117814dd0d31e19172e236f3038fc877129e869aa + Published 2026-06-16 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + › "hand-drawn sketch" + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + Taken verbatim from the author's description: + + Sampler steps 4 + Resolution 768x1152 + CFG / guidance 1 + Sampler Euler + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 361 + Rating 100% positive (48/48) + Comments 1 + Published 81 days ago + Download rate 4.5/day since release + + Adoption rank #192 of 200 in this collection (Niche) + Momentum rank #127 of 200 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general FLUX.2 [klein] 9B ones, not anything specific to this LoRA. + + General FLUX.2 [klein] 9B rules that apply here (see REFERENCE.txt): + + • Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model + output into model input. There is no CLIP side to route — klein LoRAs + are model-only because the Qwen3 text encoder is not trained (CivitAI's + recipe fixes trainTextEncoder: false). + • Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B + LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, + 5+20 vs 8+24 blocks), not a quality question. + • Mixing a "Flux.2 Klein 9B" LoRA and a "Flux.2 Klein 9B-base" LoRA in + one stack is fine — those strings describe the training checkpoint, not + incompatible tensor layouts. Set the sampler for the checkpoint + actually loaded. + • Run base-trained LoRAs on the distilled checkpoint when you want speed: + HF's klein LoRA guide reports the adapter loads there and typically + gives better results than on base, at 4 steps. + • A base-to-turbo / distillation-extraction LoRA is a whole-model delta, + not a content LoRA. Put it on the base checkpoint only, and treat its + strength as a steps/CFG dial rather than a style knob. Do not add it on + top of the already-distilled checkpoint. + • Read each LoRA's own published strength range instead of defaulting + everything to 1.0 — klein LoRAs in the wild ship very different ones + (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, + under 0.5 and sometimes 0.1 for the vafipas663 distillation delta). + • With a consistency or edit LoRA in the stack, excessive strength + suppresses the prompt: the consistency LoRA's article says that if + edits stop applying, lower it before changing anything else. + • Prefer runtime LoRA patching over baking a LoRA into the checkpoint: + ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein + models (4B or 9B) since a March 2026 update — issue #13637, still open. + • The author never states which weight family this targets. Two different + questions, two different answers. Within 9B (distilled vs base — the + two CivitAI strings): not a real incompatibility. Both load through the + same Flux2KleinPipeline and have the same repo layout; for the 4B pair, + where both transformer configs are public, the distil + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Explicit trigger word(s) — hand-drawn sketch — so the effect can be + turned on and off from the prompt. [source: model metadata] + + Compact at 79 MB — cheap to stack with other LoRAs. [source: file size] + + Author published concrete recommended settings. [source: description] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Thin documentation — expect trial and error. [source: description + length] + - Modest adoption (361 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Two different questions, two different answers. Within 9B (distilled vs + base — the two CivitAI strings): not a real incompatibility. Both load + through the same Flux2KleinPipeline and have the same repo layout; for the + 4B pair, where both transformer configs are public, the distilled and base + configs are byte-for-byte identical in every shape field, and the 9B pair + is built the same way. HF's own klein LoRA guide states the workflow + directly: "Distilled models are step-compressed for fast inference; you + train against the base checkpoint and the adapter still loads on the + distilled model afterward", and "Applying the LoRA on the distilled model + typically gives better results than the base model, and it's faster" (that + post is written against 4B, but the family structure is the same). So the + two CivitAI strings record which checkpoint the author trained or previewed + against, not a hard compatibility wall. The real risk is a silent settings + mismatch: load a base-trained LoRA on the distilled checkpoint while + keeping the author's 20-50 step / CFG ~4 preview settings and the output + degrades badly (distilled wants 4 steps / guidance 1.0), or the reverse. + Nothing errors — it just looks wrong. Across sizes (9B vs 4B): a genuine + architectural mismatch, from the published transformer configs — different + hidden width (4096 vs 3072), different block counts (8+24 vs 5+20) and a + different text-conditioning dim (12288 vs 7680, following Qwen3-8B vs + Qwen3-4B). A 4B LoRA cannot patch a 9B model: expect missing keys and shape + errors, not degraded output. The exact ComfyUI error text is unverified. KV + variant: same 9B size and same 4-step distillation, but a distinct + checkpoint with its own pipeline; whether standard klein 9B LoRAs patch it + cleanly is not documented — see unverified. + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Klein 9B Works better with simple prompts. + Showcase images : + - Distilled model + + - 768x1152 + + - Euler / normal / 4 steps / CFG 1 + + - Lora strength of 1-1.7 + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 768x1152) + 02_image.jpeg (image, 768x1152) + 03_image.jpeg (image, 768x1152) + 04_image.jpeg (image, 768x1152) + 05_image.jpeg (image, 768x1152) + 06_image.jpeg (image, 768x1152) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + 1. Diffusion model -> ComfyUI/models/diffusion_models/. Distilled: + flux-2-klein-9b-fp8.safetensors. Base: flux-2-klein- + base-9b-fp8.safetensors. (BF16 single-file weights + flux-2-klein-9b.safetensors / flux-2-klein-base-9b.safetensors from the + gated HF repos also work if you have ~29GB; community GGUF builds exist + from unsloth and leejet for lower-VRAM setups.) + 2. Text encoder -> + ComfyUI/models/text_encoders/qwen_3_8b_fp8mixed.safetensors (from Comfy- + Org/vae-text-encorder-for-flux-klein-9b). This is the 9B family's encoder; + the 4B family uses qwen_3_4b.safetensors — do not mix them. + 3. VAE -> ComfyUI/models/vae/flux2-vae.safetensors (shared across klein + sizes). + 4. LoRA -> ComfyUI/models/loras/ as a plain .safetensors. + 5. Load with Load Diffusion Model (UNETLoader) for the transformer, + CLIPLoader for the Qwen3 text encoder, VAELoader for the VAE. + docs.comfy.org ships prebuilt Flux.2 klein graphs under Workflow Templates + for both the distilled and base branches — start from those rather than + wiring by hand, since neither the docs nor BFL's repo state + sampler/scheduler/shift in prose; those values live in the template. + 6. Insert LoraLoaderModelOnly between the diffusion-model loader and the + sampler. Model-only is correct for klein: the Qwen3 text encoder is not + trained by the standard klein trainers, so there is no text-encoder half to + patch. + 7. Set steps and guidance to match the checkpoint you actually loaded, not + the one the LoRA was trained on: 4 steps / guidance 1.0 on distilled, + ~20-50 steps / guidance ~4.0 on base. + 8. Note the 9B repos are gated on HuggingFace — you must accept the FLUX + Non-Commercial License on the repo page before any download, including + scripted ones. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 03:55 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/hand-drawn-sketch/metadata.json b/hand-drawn-sketch/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..b3335a02b63534744049b2c19075f59b8bcaa73d --- /dev/null +++ b/hand-drawn-sketch/metadata.json @@ -0,0 +1,140 @@ +{ + "name": "Hand-drawn sketch", + "slug": "hand-drawn-sketch", + "base_model": "FLUX.2 [klein] 9B", + "media_type": "image", + "model_id": 2692445, + "version_id": 3023139, + "version_name": "Klein9B v1", + "source": "civitai", + "url": "https://civitai.com/models/2692445", + "repo_id": null, + "file_path": null, + "author": "Yofaraway", + "category": { + "key": "style", + "label": "Visual style" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [ + "hand-drawn sketch" + ], + "recommended_settings": { + "Sampler steps": "4", + "Resolution": "768x1152", + "CFG / guidance": "1", + "Sampler": "Euler" + }, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model output into model input. There is no CLIP side to route — klein LoRAs are model-only because the Qwen3 text encoder is not trained (CivitAI's recipe fixes trainTextEncoder: false).", + "Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, 5+20 vs 8+24 blocks), not a quality question.", + "Mixing a \"Flux.2 Klein 9B\" LoRA and a \"Flux.2 Klein 9B-base\" LoRA in one stack is fine — those strings describe the training checkpoint, not incompatible tensor layouts. Set the sampler for the checkpoint actually loaded.", + "Run base-trained LoRAs on the distilled checkpoint when you want speed: HF's klein LoRA guide reports the adapter loads there and typically gives better results than on base, at 4 steps.", + "A base-to-turbo / distillation-extraction LoRA is a whole-model delta, not a content LoRA. Put it on the base checkpoint only, and treat its strength as a steps/CFG dial rather than a style knob. Do not add it on top of the already-distilled checkpoint.", + "Read each LoRA's own published strength range instead of defaulting everything to 1.0 — klein LoRAs in the wild ship very different ones (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, under 0.5 and sometimes 0.1 for the vafipas663 distillation delta).", + "With a consistency or edit LoRA in the stack, excessive strength suppresses the prompt: the consistency LoRA's article says that if edits stop applying, lower it before changing anything else.", + "Prefer runtime LoRA patching over baking a LoRA into the checkpoint: ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein models (4B or 9B) since a March 2026 update — issue #13637, still open.", + "The author never states which weight family this targets. Two different questions, two different answers. Within 9B (distilled vs base — the two CivitAI strings): not a real incompatibility. Both load through the same Flux2KleinPipeline and have the same repo layout; for the 4B pair, where both transformer configs are public, the distil" + ] + }, + "pros": [ + "Explicit trigger word(s) — hand-drawn sketch — so the effect can be turned on and off from the prompt. [source: model metadata]", + "Compact at 79 MB — cheap to stack with other LoRAs. [source: file size]", + "Author published concrete recommended settings. [source: description]" + ], + "cons": [ + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Thin documentation — expect trial and error. [source: description length]", + "Modest adoption (361 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "hand_drawn_sketch_klein9b.safetensors", + "size_bytes": 82879480, + "size_mb": 79.04, + "sha256": "96b1ce30cadd42bc15bb457117814dd0d31e19172e236f3038fc877129e869aa" + }, + "stats": { + "downloads": 361, + "thumbs_up": 48, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 361, + "thumbs_up": 48, + "thumbs_down": 0, + "comments": 1, + "approval": "100% positive (48/48)", + "age_days": 80.5, + "downloads_per_day": 4.48, + "rank_by_downloads": 192, + "rank_by_momentum": 127, + "of_total": 200, + "adoption_tier": "Niche", + "momentum": "Long tail" + }, + "selection_score": 70.9, + "selection_reasons": [ + "361 downloads", + "48 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "768x1152", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "768x1152", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "768x1152", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "768x1152", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "768x1152", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "768x1152", + "prompt": "" + } + ], + "tags": [ + "drawing", + "style", + "strokes", + "sketch", + "pencil" + ], + "scraped_at": "2026-09-05T03:55:52+00:00", + "path": "flux2-klein-9b/hand-drawn-sketch" +} diff --git a/instagram-reality/example_images/01_image.jpeg b/instagram-reality/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..f737111d02c568d0e1302729592892d34fd5afba --- /dev/null +++ b/instagram-reality/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:383844869bf846496c12278f1d09141120bc0564ec71bb5211c7cc4cae722d0f +size 246346 diff --git a/instagram-reality/example_images/02_image.jpeg b/instagram-reality/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..140423aebaf36ca2d044ac0aaec44e7e0eeb2633 --- /dev/null +++ b/instagram-reality/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2cbf00eca859479534f9a1d8c88eda98a7838ea6dca95b247c4cf30787f0be41 +size 1339395 diff --git a/instagram-reality/example_images/03_image.jpeg b/instagram-reality/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..75c7ed03375d7f00759a0fa79529aa5c3bff06e7 --- /dev/null +++ b/instagram-reality/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:75d5523bcddc8ccd15c7d2baec48d1724e6d42087f124f86bcb27cb72063f99a +size 273044 diff --git a/instagram-reality/example_images/04_image.jpeg b/instagram-reality/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..6bd7e8bff27158bc924146e8106da3182ef6a9a5 --- /dev/null +++ b/instagram-reality/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8e5741fdf7aa51bc3511e58d13ac7d44b631fd09fc028b3267def9fa8444de0 +size 1169810 diff --git a/instagram-reality/example_images/05_image.jpeg b/instagram-reality/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..3c416883062915485c161a582c1243abe483333f --- /dev/null +++ b/instagram-reality/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27717463110da3d6be770a35bee3b795676c76c00074364dc98eec51d5894088 +size 675560 diff --git a/instagram-reality/example_images/06_image.jpeg b/instagram-reality/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..ad26f23ecca3f52f8e61f92a1eac41ee8cfe07dd --- /dev/null +++ b/instagram-reality/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa3f114d8cc789f260ab04ca9252e47c7713509b5a76fe7ededb5c7c67876fc8 +size 642112 diff --git a/instagram-reality/info.txt b/instagram-reality/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..653a082e532b558b8c544df98a38ee9cb1c2712f --- /dev/null +++ b/instagram-reality/info.txt @@ -0,0 +1,237 @@ +═══════════════════════════════════════════════════════════════════════════════ + INSTAGRAM REALITY +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model FLUX.2 [klein] 9B + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Photorealism + Author nuclear_diffusion_ + Source https://civitai.com/models/2129457 + Provenance CivitAI · model 2129457 · version 2749882 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File klein_instagramreality_v2.safetensors + Version v2.0 (Klein 9B) + Size 168.0 MB + Format SafeTensor / unknown precision + SHA256 a314ac0d0f6b237243b94107c9750c0ee7ba50c32db6ca8f49c39dd9b6d10627 + Published 2026-03-06 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + › "instagram" + › "iphone" + › "amateur" + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps Distilled + FLUX.2-klein-9B: 4 (HF card example; it is step-distilled to exactly this, + and more steps do not refine further). Base FLUX.2-klein-base-9B: 50 in the + HF card example and in ComfyUI's benchmark; community practice runs lower, + with ~20 for production and 8-12 for fast iteration reported in the CivitAI + consistency-LoRA article., cfg Distilled: guidance_scale 1.0 (HF card). It + is guidance-distilled and its transformer config sets + guidance_embeds=false, so raising CFG degrades rather than improves. Base: + guidance_scale 4.0 (HF card); the CivitAI consistency-LoRA article + recommends 3.5-4.5 for most edits., sampler Not stated by BFL or by + docs.comfy.org, and the official flux2 GitHub repo documents no sampler, + scheduler or shift values either. Community consensus is euler with a + resolution-dependent shifted-Euler / flow-match schedule (ComfyUI's Flux2 + workflows use a scheduler that interpolates shift between base_shift and + max_shift by output resolution). Start from ComfyUI's downloadable Flux.2 + klein workflow template rather than a blog. See unverified., resolution + 1024x1024 in every BFL HF card example and in CivitAI's 9B training recipe + ("resolution": 1024). Non-square ratios work but no official list of + supported dimensions or a stated alignment multiple was found — see + unverified., lora_strength 1.0 is the diffusers/ComfyUI default and what + CivitAI applies to training previews (strength 1.0). Published per-LoRA + guidance varies and should be read from the LoRA, not assumed: the widely + used Klein 9B consistency LoRA's CivitAI article says start at 0.5 within a + 0.2-1.0 range (0.2-0.4 loose, 0.5-0.7 balanced, 0.8-1.0 maximum structural + stability) and warns that too high a strength stops edits from applying; + SOLRICKS' Realistic Detail LoRA card says 0.6-0.8 subtle, 0.8-1.0 stronger, + start 0.8; the vafipas663 distillation-delta LoRA card warns it "produces + static noise at strength > 0.5. And sometimes, you have to go as low as + 0.1.". + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 1,976 + Rating 100% positive (133/133) + Comments 1 + Published 182 days ago + Download rate 7.1/day since release + + Adoption rank #131 of 200 in this collection (Modest) + Momentum rank #84 of 200 by download rate (Steady) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general FLUX.2 [klein] 9B ones, not anything specific to this LoRA. + + General FLUX.2 [klein] 9B rules that apply here (see REFERENCE.txt): + + • Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model + output into model input. There is no CLIP side to route — klein LoRAs + are model-only because the Qwen3 text encoder is not trained (CivitAI's + recipe fixes trainTextEncoder: false). + • Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B + LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, + 5+20 vs 8+24 blocks), not a quality question. + • Mixing a "Flux.2 Klein 9B" LoRA and a "Flux.2 Klein 9B-base" LoRA in + one stack is fine — those strings describe the training checkpoint, not + incompatible tensor layouts. Set the sampler for the checkpoint + actually loaded. + • Run base-trained LoRAs on the distilled checkpoint when you want speed: + HF's klein LoRA guide reports the adapter loads there and typically + gives better results than on base, at 4 steps. + • A base-to-turbo / distillation-extraction LoRA is a whole-model delta, + not a content LoRA. Put it on the base checkpoint only, and treat its + strength as a steps/CFG dial rather than a style knob. Do not add it on + top of the already-distilled checkpoint. + • Read each LoRA's own published strength range instead of defaulting + everything to 1.0 — klein LoRAs in the wild ship very different ones + (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, + under 0.5 and sometimes 0.1 for the vafipas663 distillation delta). + • With a consistency or edit LoRA in the stack, excessive strength + suppresses the prompt: the consistency LoRA's article says that if + edits stop applying, lower it before changing anything else. + • Prefer runtime LoRA patching over baking a LoRA into the checkpoint: + ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein + models (4B or 9B) since a March 2026 update — issue #13637, still open. + • The author never states which weight family this targets. Two different + questions, two different answers. Within 9B (distilled vs base — the + two CivitAI strings): not a real incompatibility. Both load through the + same Flux2KleinPipeline and have the same repo layout; for the 4B pair, + where both transformer configs are public, the distil + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Well established — 1,976 downloads. [source: CivitAI stats] + + Positively received — 133 thumbs-up with no down-votes. [source: + CivitAI stats] + + Explicit trigger word(s) — instagram, iphone, amateur — so the effect + can be turned on and off from the prompt. [source: model metadata] + + Compact at 168 MB — cheap to stack with other LoRAs. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Two different questions, two different answers. Within 9B (distilled vs + base — the two CivitAI strings): not a real incompatibility. Both load + through the same Flux2KleinPipeline and have the same repo layout; for the + 4B pair, where both transformer configs are public, the distilled and base + configs are byte-for-byte identical in every shape field, and the 9B pair + is built the same way. HF's own klein LoRA guide states the workflow + directly: "Distilled models are step-compressed for fast inference; you + train against the base checkpoint and the adapter still loads on the + distilled model afterward", and "Applying the LoRA on the distilled model + typically gives better results than the base model, and it's faster" (that + post is written against 4B, but the family structure is the same). So the + two CivitAI strings record which checkpoint the author trained or previewed + against, not a hard compatibility wall. The real risk is a silent settings + mismatch: load a base-trained LoRA on the distilled checkpoint while + keeping the author's 20-50 step / CFG ~4 preview settings and the output + degrades badly (distilled wants 4 steps / guidance 1.0), or the reverse. + Nothing errors — it just looks wrong. Across sizes (9B vs 4B): a genuine + architectural mismatch, from the published transformer configs — different + hidden width (4096 vs 3072), different block counts (8+24 vs 5+20) and a + different text-conditioning dim (12288 vs 7680, following Qwen3-8B vs + Qwen3-4B). A 4B LoRA cannot patch a 9B model: expect missing keys and shape + errors, not degraded output. The exact ComfyUI error text is unverified. KV + variant: same 9B size and same 4-step distillation, but a distinct + checkpoint with its own pipeline; whether standard klein 9B LoRAs patch it + cleanly is not documented — see unverified. + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Photorealism style trained on Instagram images. Doesn't really need a + trigger word, but using descriptions like "amateur photo", "Instagram + photo" or "iPhone photo" might help. + Trained on 1000 high quality images and captioned using Gemini 3 Flash. + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 832x1216) + 02_image.jpeg (image, 768x1360) + 03_image.jpeg (image, 832x1216) + 04_image.jpeg (image, 832x1248) + 05_image.jpeg (image, 1024x1024) + 06_image.jpeg (image, 1280x1856) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + 1. Diffusion model -> ComfyUI/models/diffusion_models/. Distilled: + flux-2-klein-9b-fp8.safetensors. Base: flux-2-klein- + base-9b-fp8.safetensors. (BF16 single-file weights + flux-2-klein-9b.safetensors / flux-2-klein-base-9b.safetensors from the + gated HF repos also work if you have ~29GB; community GGUF builds exist + from unsloth and leejet for lower-VRAM setups.) + 2. Text encoder -> + ComfyUI/models/text_encoders/qwen_3_8b_fp8mixed.safetensors (from Comfy- + Org/vae-text-encorder-for-flux-klein-9b). This is the 9B family's encoder; + the 4B family uses qwen_3_4b.safetensors — do not mix them. + 3. VAE -> ComfyUI/models/vae/flux2-vae.safetensors (shared across klein + sizes). + 4. LoRA -> ComfyUI/models/loras/ as a plain .safetensors. + 5. Load with Load Diffusion Model (UNETLoader) for the transformer, + CLIPLoader for the Qwen3 text encoder, VAELoader for the VAE. + docs.comfy.org ships prebuilt Flux.2 klein graphs under Workflow Templates + for both the distilled and base branches — start from those rather than + wiring by hand, since neither the docs nor BFL's repo state + sampler/scheduler/shift in prose; those values live in the template. + 6. Insert LoraLoaderModelOnly between the diffusion-model loader and the + sampler. Model-only is correct for klein: the Qwen3 text encoder is not + trained by the standard klein trainers, so there is no text-encoder half to + patch. + 7. Set steps and guidance to match the checkpoint you actually loaded, not + the one the LoRA was trained on: 4 steps / guidance 1.0 on distilled, + ~20-50 steps / guidance ~4.0 on base. + 8. Note the 9B repos are gated on HuggingFace — you must accept the FLUX + Non-Commercial License on the repo page before any download, including + scripted ones. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 03:55 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/instagram-reality/klein_instagramreality_v2.safetensors b/instagram-reality/klein_instagramreality_v2.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..109441339985c497184bd4bdca1bc6b844ba0bce --- /dev/null +++ b/instagram-reality/klein_instagramreality_v2.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a314ac0d0f6b237243b94107c9750c0ee7ba50c32db6ca8f49c39dd9b6d10627 +size 176173344 diff --git a/instagram-reality/metadata.json b/instagram-reality/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..260c97c1019b4d5f36288c3792b99181bc7eb56e --- /dev/null +++ b/instagram-reality/metadata.json @@ -0,0 +1,135 @@ +{ + "name": "Instagram Reality", + "slug": "instagram-reality", + "base_model": "FLUX.2 [klein] 9B", + "media_type": "image", + "model_id": 2129457, + "version_id": 2749882, + "version_name": "v2.0 (Klein 9B)", + "source": "civitai", + "url": "https://civitai.com/models/2129457", + "repo_id": null, + "file_path": null, + "author": "nuclear_diffusion_", + "category": { + "key": "photoreal", + "label": "Photorealism" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [ + "instagram", + "iphone", + "amateur" + ], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model output into model input. There is no CLIP side to route — klein LoRAs are model-only because the Qwen3 text encoder is not trained (CivitAI's recipe fixes trainTextEncoder: false).", + "Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, 5+20 vs 8+24 blocks), not a quality question.", + "Mixing a \"Flux.2 Klein 9B\" LoRA and a \"Flux.2 Klein 9B-base\" LoRA in one stack is fine — those strings describe the training checkpoint, not incompatible tensor layouts. Set the sampler for the checkpoint actually loaded.", + "Run base-trained LoRAs on the distilled checkpoint when you want speed: HF's klein LoRA guide reports the adapter loads there and typically gives better results than on base, at 4 steps.", + "A base-to-turbo / distillation-extraction LoRA is a whole-model delta, not a content LoRA. Put it on the base checkpoint only, and treat its strength as a steps/CFG dial rather than a style knob. Do not add it on top of the already-distilled checkpoint.", + "Read each LoRA's own published strength range instead of defaulting everything to 1.0 — klein LoRAs in the wild ship very different ones (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, under 0.5 and sometimes 0.1 for the vafipas663 distillation delta).", + "With a consistency or edit LoRA in the stack, excessive strength suppresses the prompt: the consistency LoRA's article says that if edits stop applying, lower it before changing anything else.", + "Prefer runtime LoRA patching over baking a LoRA into the checkpoint: ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein models (4B or 9B) since a March 2026 update — issue #13637, still open.", + "The author never states which weight family this targets. Two different questions, two different answers. Within 9B (distilled vs base — the two CivitAI strings): not a real incompatibility. Both load through the same Flux2KleinPipeline and have the same repo layout; for the 4B pair, where both transformer configs are public, the distil" + ] + }, + "pros": [ + "Well established — 1,976 downloads. [source: CivitAI stats]", + "Positively received — 133 thumbs-up with no down-votes. [source: CivitAI stats]", + "Explicit trigger word(s) — instagram, iphone, amateur — so the effect can be turned on and off from the prompt. [source: model metadata]", + "Compact at 168 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]" + ], + "file": { + "name": "klein_instagramreality_v2.safetensors", + "size_bytes": 176173344, + "size_mb": 168.01, + "sha256": "a314ac0d0f6b237243b94107c9750c0ee7ba50c32db6ca8f49c39dd9b6d10627" + }, + "stats": { + "downloads": 1976, + "thumbs_up": 133, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 1976, + "thumbs_up": 133, + "thumbs_down": 0, + "comments": 1, + "approval": "100% positive (133/133)", + "age_days": 182.4, + "downloads_per_day": 7.07, + "rank_by_downloads": 131, + "rank_by_momentum": 84, + "of_total": 200, + "adoption_tier": "Modest", + "momentum": "Steady" + }, + "selection_score": 71.1, + "selection_reasons": [ + "1,976 downloads", + "133 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "768x1360", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "832x1248", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "1280x1856", + "prompt": "" + } + ], + "tags": [ + "photorealistic", + "instagram", + "realistic", + "style" + ], + "scraped_at": "2026-09-05T03:55:03+00:00", + "path": "flux2-klein-9b/instagram-reality" +} diff --git a/jacques-tati-movies-style/example_images/01_image.jpeg b/jacques-tati-movies-style/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..1558fbb0fed63a04e3753887d349083961c5d593 --- /dev/null +++ b/jacques-tati-movies-style/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c14734e07199778d40e66dac47b4d8383d0f9c9bc7a48c4395cf508601caa96c +size 191319 diff --git a/jacques-tati-movies-style/example_images/02_image.jpeg b/jacques-tati-movies-style/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..76b6ffdd81542403b2c44887481279c58a68120a --- /dev/null +++ b/jacques-tati-movies-style/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7eba04e1f1a767cdc4e10045a25fded2a5b5ca3c79371ca347b22fc11f0d283b +size 1241044 diff --git a/jacques-tati-movies-style/example_images/03_image.jpeg b/jacques-tati-movies-style/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..ab7e71fae81617d5c347f2eb628ea9b81d9b9b32 --- /dev/null +++ b/jacques-tati-movies-style/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da1984cb3d63ef1a7e454206b343baee86e14ebdabe58ff153a53eb5d1d9e7ce +size 1319208 diff --git a/jacques-tati-movies-style/example_images/04_image.jpeg b/jacques-tati-movies-style/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..09f4cd4748b4e01abc89f04b3df12cda6a971321 --- /dev/null +++ b/jacques-tati-movies-style/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58e9ddc8eabeb66d3c8a264fdb25015a9fb24dc6eab0f66f8562deddc34b277f +size 1336795 diff --git a/jacques-tati-movies-style/example_images/05_image.jpeg b/jacques-tati-movies-style/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..c5b2432e34bffb5e55cedafdae33a683f6179ef0 --- /dev/null +++ b/jacques-tati-movies-style/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95ca0161cc891122025af146527884b42ce95c2727a9dbbdb681b764264b4736 +size 1196826 diff --git a/jacques-tati-movies-style/example_images/06_image.jpeg b/jacques-tati-movies-style/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..273871f334eea97f9ae1a71992679444513cc2c3 --- /dev/null +++ b/jacques-tati-movies-style/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83311adcedfd86ed9fc43861aa41f4f0f56cbf8c0d87cb124de549d3ac4bf603 +size 1428410 diff --git a/jacques-tati-movies-style/info.txt b/jacques-tati-movies-style/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..864eacca61c7023908c1f89be2742ed46ceac1ab --- /dev/null +++ b/jacques-tati-movies-style/info.txt @@ -0,0 +1,229 @@ +═══════════════════════════════════════════════════════════════════════════════ + JACQUES TATI MOVIES STYLE +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model FLUX.2 [klein] 9B + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Visual style + Author Yofaraway + Source https://civitai.com/models/2065214 + Provenance CivitAI · model 2065214 · version 2931433 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File jacques_tati_klein9b.safetensors + Version Klein9B v1 + Size 79.0 MB + Format SafeTensor / unknown precision + SHA256 1b68a0ddc3cc106354d094f753e5769552986cfe5fec723e0330707ea6fa8f71 + Published 2026-05-14 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + › "Jacques Tati movie still" + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + Taken verbatim from the author's description: + + Sampler steps 8 + Resolution 768x1152 + CFG / guidance 1 + Sampler Euler + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 1,208 + Rating 100% positive (128/128) + Comments 0 + Published 113 days ago + Download rate 2.4/day since release + + Adoption rank #158 of 200 in this collection (Niche) + Momentum rank #192 of 200 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general FLUX.2 [klein] 9B ones, not anything specific to this LoRA. + + General FLUX.2 [klein] 9B rules that apply here (see REFERENCE.txt): + + • Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model + output into model input. There is no CLIP side to route — klein LoRAs + are model-only because the Qwen3 text encoder is not trained (CivitAI's + recipe fixes trainTextEncoder: false). + • Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B + LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, + 5+20 vs 8+24 blocks), not a quality question. + • Mixing a "Flux.2 Klein 9B" LoRA and a "Flux.2 Klein 9B-base" LoRA in + one stack is fine — those strings describe the training checkpoint, not + incompatible tensor layouts. Set the sampler for the checkpoint + actually loaded. + • Run base-trained LoRAs on the distilled checkpoint when you want speed: + HF's klein LoRA guide reports the adapter loads there and typically + gives better results than on base, at 4 steps. + • A base-to-turbo / distillation-extraction LoRA is a whole-model delta, + not a content LoRA. Put it on the base checkpoint only, and treat its + strength as a steps/CFG dial rather than a style knob. Do not add it on + top of the already-distilled checkpoint. + • Read each LoRA's own published strength range instead of defaulting + everything to 1.0 — klein LoRAs in the wild ship very different ones + (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, + under 0.5 and sometimes 0.1 for the vafipas663 distillation delta). + • With a consistency or edit LoRA in the stack, excessive strength + suppresses the prompt: the consistency LoRA's article says that if + edits stop applying, lower it before changing anything else. + • Prefer runtime LoRA patching over baking a LoRA into the checkpoint: + ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein + models (4B or 9B) since a March 2026 update — issue #13637, still open. + • The author never states which weight family this targets. Two different + questions, two different answers. Within 9B (distilled vs base — the + two CivitAI strings): not a real incompatibility. Both load through the + same Flux2KleinPipeline and have the same repo layout; for the 4B pair, + where both transformer configs are public, the distil + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Well established — 1,208 downloads. [source: CivitAI stats] + + Positively received — 128 thumbs-up with no down-votes. [source: + CivitAI stats] + + Explicit trigger word(s) — Jacques Tati movie still — so the effect can + be turned on and off from the prompt. [source: model metadata] + + Compact at 79 MB — cheap to stack with other LoRAs. [source: file size] + + Author published concrete recommended settings. [source: description] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Two different questions, two different answers. Within 9B (distilled vs + base — the two CivitAI strings): not a real incompatibility. Both load + through the same Flux2KleinPipeline and have the same repo layout; for the + 4B pair, where both transformer configs are public, the distilled and base + configs are byte-for-byte identical in every shape field, and the 9B pair + is built the same way. HF's own klein LoRA guide states the workflow + directly: "Distilled models are step-compressed for fast inference; you + train against the base checkpoint and the adapter still loads on the + distilled model afterward", and "Applying the LoRA on the distilled model + typically gives better results than the base model, and it's faster" (that + post is written against 4B, but the family structure is the same). So the + two CivitAI strings record which checkpoint the author trained or previewed + against, not a hard compatibility wall. The real risk is a silent settings + mismatch: load a base-trained LoRA on the distilled checkpoint while + keeping the author's 20-50 step / CFG ~4 preview settings and the output + degrades badly (distilled wants 4 steps / guidance 1.0), or the reverse. + Nothing errors — it just looks wrong. Across sizes (9B vs 4B): a genuine + architectural mismatch, from the published transformer configs — different + hidden width (4096 vs 3072), different block counts (8+24 vs 5+20) and a + different text-conditioning dim (12288 vs 7680, following Qwen3-8B vs + Qwen3-4B). A 4B LoRA cannot patch a 9B model: expect missing keys and shape + errors, not degraded output. The exact ComfyUI error text is unverified. KV + variant: same 9B size and same 4-step distillation, but a distinct + checkpoint with its own pipeline; whether standard klein 9B LoRAs patch it + cleanly is not documented — see unverified. + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Krea2 Showcase images : + - Turbo model + + - 768x1152 + + - Euler / simple / 8 steps / CFG 1 + + - Lora strength 1-1.5 + + Klein 9B Showcase images : + - Distilled model + + - 768x1152 + + - Euler / normal / 4 steps / CFG 1 + + - Lora strength of 1 + + Qwen I keep experimenting with qwen lora training, so here a lora trained + on screencaps from movies "Mon oncle" and "Playtime" directed by Jacques + Tati. The dataset was around 100 images, starting the caption with + "tatitati movie still of" + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 832x1216) + 02_image.jpeg (image, 768x1152) + 03_image.jpeg (image, 1152x768) + 04_image.jpeg (image, 768x1152) + 05_image.jpeg (image, 768x1152) + 06_image.jpeg (image, 1152x768) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + 1. Diffusion model -> ComfyUI/models/diffusion_models/. Distilled: + flux-2-klein-9b-fp8.safetensors. Base: flux-2-klein- + base-9b-fp8.safetensors. (BF16 single-file weights + flux-2-klein-9b.safetensors / flux-2-klein-base-9b.safetensors from the + gated HF repos also work if you have ~29GB; community GGUF builds exist + from unsloth and leejet for lower-VRAM setups.) + 2. Text encoder -> + ComfyUI/models/text_encoders/qwen_3_8b_fp8mixed.safetensors (from Comfy- + Org/vae-text-encorder-for-flux-klein-9b). This is the 9B family's encoder; + the 4B family uses qwen_3_4b.safetensors — do not mix them. + 3. VAE -> ComfyUI/models/vae/flux2-vae.safetensors (shared across klein + sizes). + 4. LoRA -> ComfyUI/models/loras/ as a plain .safetensors. + 5. Load with Load Diffusion Model (UNETLoader) for the transformer, + CLIPLoader for the Qwen3 text encoder, VAELoader for the VAE. + docs.comfy.org ships prebuilt Flux.2 klein graphs under Workflow Templates + for both the distilled and base branches — start from those rather than + wiring by hand, since neither the docs nor BFL's repo state + sampler/scheduler/shift in prose; those values live in the template. + 6. Insert LoraLoaderModelOnly between the diffusion-model loader and the + sampler. Model-only is correct for klein: the Qwen3 text encoder is not + trained by the standard klein trainers, so there is no text-encoder half to + patch. + 7. Set steps and guidance to match the checkpoint you actually loaded, not + the one the LoRA was trained on: 4 steps / guidance 1.0 on distilled, + ~20-50 steps / guidance ~4.0 on base. + 8. Note the 9B repos are gated on HuggingFace — you must accept the FLUX + Non-Commercial License on the repo page before any download, including + scripted ones. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 03:55 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/jacques-tati-movies-style/jacques_tati_klein9b.safetensors b/jacques-tati-movies-style/jacques_tati_klein9b.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bf0fcf050e712ca6a9f913b5ed254aa41f2db905 --- /dev/null +++ b/jacques-tati-movies-style/jacques_tati_klein9b.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b68a0ddc3cc106354d094f753e5769552986cfe5fec723e0330707ea6fa8f71 +size 82879472 diff --git a/jacques-tati-movies-style/metadata.json b/jacques-tati-movies-style/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..8c3e3222af6adad95eea167814a54d680c30c58c --- /dev/null +++ b/jacques-tati-movies-style/metadata.json @@ -0,0 +1,140 @@ +{ + "name": "Jacques Tati movies style", + "slug": "jacques-tati-movies-style", + "base_model": "FLUX.2 [klein] 9B", + "media_type": "image", + "model_id": 2065214, + "version_id": 2931433, + "version_name": "Klein9B v1", + "source": "civitai", + "url": "https://civitai.com/models/2065214", + "repo_id": null, + "file_path": null, + "author": "Yofaraway", + "category": { + "key": "style", + "label": "Visual style" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [ + "Jacques Tati movie still" + ], + "recommended_settings": { + "Sampler steps": "8", + "Resolution": "768x1152", + "CFG / guidance": "1", + "Sampler": "Euler" + }, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model output into model input. There is no CLIP side to route — klein LoRAs are model-only because the Qwen3 text encoder is not trained (CivitAI's recipe fixes trainTextEncoder: false).", + "Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, 5+20 vs 8+24 blocks), not a quality question.", + "Mixing a \"Flux.2 Klein 9B\" LoRA and a \"Flux.2 Klein 9B-base\" LoRA in one stack is fine — those strings describe the training checkpoint, not incompatible tensor layouts. Set the sampler for the checkpoint actually loaded.", + "Run base-trained LoRAs on the distilled checkpoint when you want speed: HF's klein LoRA guide reports the adapter loads there and typically gives better results than on base, at 4 steps.", + "A base-to-turbo / distillation-extraction LoRA is a whole-model delta, not a content LoRA. Put it on the base checkpoint only, and treat its strength as a steps/CFG dial rather than a style knob. Do not add it on top of the already-distilled checkpoint.", + "Read each LoRA's own published strength range instead of defaulting everything to 1.0 — klein LoRAs in the wild ship very different ones (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, under 0.5 and sometimes 0.1 for the vafipas663 distillation delta).", + "With a consistency or edit LoRA in the stack, excessive strength suppresses the prompt: the consistency LoRA's article says that if edits stop applying, lower it before changing anything else.", + "Prefer runtime LoRA patching over baking a LoRA into the checkpoint: ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein models (4B or 9B) since a March 2026 update — issue #13637, still open.", + "The author never states which weight family this targets. Two different questions, two different answers. Within 9B (distilled vs base — the two CivitAI strings): not a real incompatibility. Both load through the same Flux2KleinPipeline and have the same repo layout; for the 4B pair, where both transformer configs are public, the distil" + ] + }, + "pros": [ + "Well established — 1,208 downloads. [source: CivitAI stats]", + "Positively received — 128 thumbs-up with no down-votes. [source: CivitAI stats]", + "Explicit trigger word(s) — Jacques Tati movie still — so the effect can be turned on and off from the prompt. [source: model metadata]", + "Compact at 79 MB — cheap to stack with other LoRAs. [source: file size]", + "Author published concrete recommended settings. [source: description]" + ], + "cons": [ + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]" + ], + "file": { + "name": "jacques_tati_klein9b.safetensors", + "size_bytes": 82879472, + "size_mb": 79.04, + "sha256": "1b68a0ddc3cc106354d094f753e5769552986cfe5fec723e0330707ea6fa8f71" + }, + "stats": { + "downloads": 1208, + "thumbs_up": 128, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 1208, + "thumbs_up": 128, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (128/128)", + "age_days": 113.5, + "downloads_per_day": 2.44, + "rank_by_downloads": 158, + "rank_by_momentum": 192, + "of_total": 200, + "adoption_tier": "Niche", + "momentum": "Long tail" + }, + "selection_score": 70.9, + "selection_reasons": [ + "1,208 downloads", + "128 thumbs-up", + "documented triggers + usage notes", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "768x1152", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1152x768", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "768x1152", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "768x1152", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "1152x768", + "prompt": "" + } + ], + "tags": [ + "french cinema", + "style", + "movie", + "nouvelle vague" + ], + "scraped_at": "2026-09-05T03:55:09+00:00", + "path": "flux2-klein-9b/jacques-tati-movies-style" +} diff --git a/klein-realistic-comic-slider/example_images/01_image.jpeg b/klein-realistic-comic-slider/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..15ea247c4f74ecc9fdd3e5ea913dbcb6dd5d5d20 --- /dev/null +++ b/klein-realistic-comic-slider/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:266080e935ec5dffecbe530d5a355d2b63801a26db156b8eb52e475ca3cfe5cd +size 432028 diff --git a/klein-realistic-comic-slider/example_images/02_image.jpeg b/klein-realistic-comic-slider/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..a93fd5c8bbcc814de7487f59f317d374ae857128 --- /dev/null +++ b/klein-realistic-comic-slider/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:116e514134e77c19931db7eca444d46575247b5c513d1cc5975e6a82b09f0ed4 +size 458700 diff --git a/klein-realistic-comic-slider/example_images/03_image.jpeg b/klein-realistic-comic-slider/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..bed030f728782f7365687e6fac5ee7f60e2f944e --- /dev/null +++ b/klein-realistic-comic-slider/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a827749070e3d586dbf1c33181e8184ec53acbf73e99594e68de91bcdc511cce +size 421239 diff --git a/klein-realistic-comic-slider/example_images/04_image.jpeg b/klein-realistic-comic-slider/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..d441b405d46e5e6d7219f960ec60bc4a502f7406 --- /dev/null +++ b/klein-realistic-comic-slider/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76c3a7c1d92fb0099fe793bfaf6ff2b96839bc9c1e450bc1d4ee5c9c630b9ad6 +size 240172 diff --git a/klein-realistic-comic-slider/example_images/05_image.jpeg b/klein-realistic-comic-slider/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..4c9e4e5554ee8fa775c9178354e63bca7dbdcf4d --- /dev/null +++ b/klein-realistic-comic-slider/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:81260de2d9730cf9814cf4a931dfd831e86a6e4893e539b919155ca54b58bc76 +size 7049699 diff --git a/klein-realistic-comic-slider/example_images/06_poster.jpg b/klein-realistic-comic-slider/example_images/06_poster.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fb58f35f647d123e2a33ed35a0a78fc35f8d9700 --- /dev/null +++ b/klein-realistic-comic-slider/example_images/06_poster.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddbbe1afb9b0feaa54f155e9f72fb1df054502a264522b5f7ce4928c011b3dd4 +size 128437 diff --git a/klein-realistic-comic-slider/example_images/06_video.mp4 b/klein-realistic-comic-slider/example_images/06_video.mp4 new file mode 100644 index 0000000000000000000000000000000000000000..298e902b229a82229c9e0a375df11bc6f3d61bab --- /dev/null +++ b/klein-realistic-comic-slider/example_images/06_video.mp4 @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2d6275a4f92e77cbe5a9fa117dc34fc46d5ed719bc451cb6afc8e9283b7f4aa +size 466666 diff --git a/klein-realistic-comic-slider/info.txt b/klein-realistic-comic-slider/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..59c206f931470ac05a284de9571cdd5467b34748 --- /dev/null +++ b/klein-realistic-comic-slider/info.txt @@ -0,0 +1,239 @@ +═══════════════════════════════════════════════════════════════════════════════ + KLEIN REALISTIC / COMIC SLIDER +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model FLUX.2 [klein] 9B + Generation mode Text-to-image + Checkpoint FLUX.2-klein-9B (distilled) — CivitAI string "Flux.2 Klein 9B" + Category Visual style + Author l226 + Source https://civitai.com/models/2332534 + Provenance CivitAI · model 2332534 · version 2623808 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File klein_slider_comic.safetensors + Version Klein 9B + Size 19.8 MB + Format SafeTensor / unknown precision + SHA256 b790414cac216841cfcafccace9d4361da991267f0fdb62834facc98fb4b29b7 + Published 2026-01-23 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps Distilled + FLUX.2-klein-9B: 4 (HF card example; it is step-distilled to exactly this, + and more steps do not refine further). Base FLUX.2-klein-base-9B: 50 in the + HF card example and in ComfyUI's benchmark; community practice runs lower, + with ~20 for production and 8-12 for fast iteration reported in the CivitAI + consistency-LoRA article., cfg Distilled: guidance_scale 1.0 (HF card). It + is guidance-distilled and its transformer config sets + guidance_embeds=false, so raising CFG degrades rather than improves. Base: + guidance_scale 4.0 (HF card); the CivitAI consistency-LoRA article + recommends 3.5-4.5 for most edits., sampler Not stated by BFL or by + docs.comfy.org, and the official flux2 GitHub repo documents no sampler, + scheduler or shift values either. Community consensus is euler with a + resolution-dependent shifted-Euler / flow-match schedule (ComfyUI's Flux2 + workflows use a scheduler that interpolates shift between base_shift and + max_shift by output resolution). Start from ComfyUI's downloadable Flux.2 + klein workflow template rather than a blog. See unverified., resolution + 1024x1024 in every BFL HF card example and in CivitAI's 9B training recipe + ("resolution": 1024). Non-square ratios work but no official list of + supported dimensions or a stated alignment multiple was found — see + unverified., lora_strength 1.0 is the diffusers/ComfyUI default and what + CivitAI applies to training previews (strength 1.0). Published per-LoRA + guidance varies and should be read from the LoRA, not assumed: the widely + used Klein 9B consistency LoRA's CivitAI article says start at 0.5 within a + 0.2-1.0 range (0.2-0.4 loose, 0.5-0.7 balanced, 0.8-1.0 maximum structural + stability) and warns that too high a strength stops edits from applying; + SOLRICKS' Realistic Detail LoRA card says 0.6-0.8 subtle, 0.8-1.0 stronger, + start 0.8; the vafipas663 distillation-delta LoRA card warns it "produces + static noise at strength > 0.5. And sometimes, you have to go as low as + 0.1.". + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 2,401 + Rating 100% positive (112/112) + Comments 9 + Published 224 days ago + Download rate 10.7/day since release + + Adoption rank #117 of 200 in this collection (Modest) + Momentum rank #53 of 200 by download rate (Steady) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general FLUX.2 [klein] 9B ones, not anything specific to this LoRA. + + General FLUX.2 [klein] 9B rules that apply here (see REFERENCE.txt): + + • Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model + output into model input. There is no CLIP side to route — klein LoRAs + are model-only because the Qwen3 text encoder is not trained (CivitAI's + recipe fixes trainTextEncoder: false). + • Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B + LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, + 5+20 vs 8+24 blocks), not a quality question. + • Mixing a "Flux.2 Klein 9B" LoRA and a "Flux.2 Klein 9B-base" LoRA in + one stack is fine — those strings describe the training checkpoint, not + incompatible tensor layouts. Set the sampler for the checkpoint + actually loaded. + • Run base-trained LoRAs on the distilled checkpoint when you want speed: + HF's klein LoRA guide reports the adapter loads there and typically + gives better results than on base, at 4 steps. + • A base-to-turbo / distillation-extraction LoRA is a whole-model delta, + not a content LoRA. Put it on the base checkpoint only, and treat its + strength as a steps/CFG dial rather than a style knob. Do not add it on + top of the already-distilled checkpoint. + • Read each LoRA's own published strength range instead of defaulting + everything to 1.0 — klein LoRAs in the wild ship very different ones + (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, + under 0.5 and sometimes 0.1 for the vafipas663 distillation delta). + • With a consistency or edit LoRA in the stack, excessive strength + suppresses the prompt: the consistency LoRA's article says that if + edits stop applying, lower it before changing anything else. + • Prefer runtime LoRA patching over baking a LoRA into the checkpoint: + ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein + models (4B or 9B) since a March 2026 update — issue #13637, still open. + • Step-distilled to exactly 4 inference steps and guidance-distilled; HF + card example is num_inference_steps=4, guidance_scale=1.0. Its + transformer config has guidance_embeds=false. This is the intended + inference target: ComfyUI's blog benchmarks it at ~2s and 19.6GB VRAM + on an RTX 5090. Not offered as a training target by BFL's docs or + CivitAI's recipe. + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Well established — 2,401 downloads. [source: CivitAI stats] + + Positively received — 112 thumbs-up with no down-votes. [source: + CivitAI stats] + + Compact at 20 MB — cheap to stack with other LoRAs. [source: file size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Two different questions, two different answers. Within 9B (distilled vs + base — the two CivitAI strings): not a real incompatibility. Both load + through the same Flux2KleinPipeline and have the same repo layout; for the + 4B pair, where both transformer configs are public, the distilled and base + configs are byte-for-byte identical in every shape field, and the 9B pair + is built the same way. HF's own klein LoRA guide states the workflow + directly: "Distilled models are step-compressed for fast inference; you + train against the base checkpoint and the adapter still loads on the + distilled model afterward", and "Applying the LoRA on the distilled model + typically gives better results than the base model, and it's faster" (that + post is written against 4B, but the family structure is the same). So the + two CivitAI strings record which checkpoint the author trained or previewed + against, not a hard compatibility wall. The real risk is a silent settings + mismatch: load a base-trained LoRA on the distilled checkpoint while + keeping the author's 20-50 step / CFG ~4 preview settings and the output + degrades badly (distilled wants 4 steps / guidance 1.0), or the reverse. + Nothing errors — it just looks wrong. Across sizes (9B vs 4B): a genuine + architectural mismatch, from the published transformer configs — different + hidden width (4096 vs 3072), different block counts (8+24 vs 5+20) and a + different text-conditioning dim (12288 vs 7680, following Qwen3-8B vs + Qwen3-4B). A 4B LoRA cannot patch a 9B model: expect missing keys and shape + errors, not degraded output. The exact ComfyUI error text is unverified. KV + variant: same 9B size and same 4-step distillation, but a distinct + checkpoint with its own pipeline; whether standard klein 9B LoRAs patch it + cleanly is not documented — see unverified. + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + NEGATIVE REALISTIC / POSITIVE COMIC This LoRA can either make a realistic + scene into a comic style one, or make a comic image into a photorealistic + one. Works both in I2I and T2I. Note: This is a concept LoRA, not a style + LoRA. The model knows what a comic is, the LoRA just plays on these + weights. The training dataset does not contain any comic images. The sample + weights range from 0 to 8 (full effect). The last sample is a comic image + generated in T2I and made realistic in I2I with negative (-4) slider weight + (empty prompt). Hope you like it! + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 1216x832) + 02_image.jpeg (image, 1024x1024) + 03_image.jpeg (image, 1216x832) + 04_image.jpeg (image, 1216x832) + 05_image.jpeg (image, 2048x2048) + 06_video.mp4 (video, 768x768) + still: 06_poster.jpg + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + 1. Diffusion model -> ComfyUI/models/diffusion_models/. Distilled: + flux-2-klein-9b-fp8.safetensors. Base: flux-2-klein- + base-9b-fp8.safetensors. (BF16 single-file weights + flux-2-klein-9b.safetensors / flux-2-klein-base-9b.safetensors from the + gated HF repos also work if you have ~29GB; community GGUF builds exist + from unsloth and leejet for lower-VRAM setups.) + 2. Text encoder -> + ComfyUI/models/text_encoders/qwen_3_8b_fp8mixed.safetensors (from Comfy- + Org/vae-text-encorder-for-flux-klein-9b). This is the 9B family's encoder; + the 4B family uses qwen_3_4b.safetensors — do not mix them. + 3. VAE -> ComfyUI/models/vae/flux2-vae.safetensors (shared across klein + sizes). + 4. LoRA -> ComfyUI/models/loras/ as a plain .safetensors. + 5. Load with Load Diffusion Model (UNETLoader) for the transformer, + CLIPLoader for the Qwen3 text encoder, VAELoader for the VAE. + docs.comfy.org ships prebuilt Flux.2 klein graphs under Workflow Templates + for both the distilled and base branches — start from those rather than + wiring by hand, since neither the docs nor BFL's repo state + sampler/scheduler/shift in prose; those values live in the template. + 6. Insert LoraLoaderModelOnly between the diffusion-model loader and the + sampler. Model-only is correct for klein: the Qwen3 text encoder is not + trained by the standard klein trainers, so there is no text-encoder half to + patch. + 7. Set steps and guidance to match the checkpoint you actually loaded, not + the one the LoRA was trained on: 4 steps / guidance 1.0 on distilled, + ~20-50 steps / guidance ~4.0 on base. + 8. Note the 9B repos are gated on HuggingFace — you must accept the FLUX + Non-Commercial License on the repo page before any download, including + scripted ones. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 03:55 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/klein-realistic-comic-slider/klein_slider_comic.safetensors b/klein-realistic-comic-slider/klein_slider_comic.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1ab5ec3e6fc496424b703d945ca4d9ee3908f471 --- /dev/null +++ b/klein-realistic-comic-slider/klein_slider_comic.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b790414cac216841cfcafccace9d4361da991267f0fdb62834facc98fb4b29b7 +size 20738136 diff --git a/klein-realistic-comic-slider/metadata.json b/klein-realistic-comic-slider/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..5643d3214bfb523062fde190e305badb10a404c7 --- /dev/null +++ b/klein-realistic-comic-slider/metadata.json @@ -0,0 +1,133 @@ +{ + "name": "Klein Realistic / Comic Slider", + "slug": "klein-realistic-comic-slider", + "base_model": "FLUX.2 [klein] 9B", + "media_type": "image", + "model_id": 2332534, + "version_id": 2623808, + "version_name": "Klein 9B", + "source": "civitai", + "url": "https://civitai.com/models/2332534", + "repo_id": null, + "file_path": null, + "author": "l226", + "category": { + "key": "style", + "label": "Visual style" + }, + "modalities": [ + "t2i" + ], + "primary_modality": "t2i", + "checkpoint_family": "klein-9b-distilled", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model output into model input. There is no CLIP side to route — klein LoRAs are model-only because the Qwen3 text encoder is not trained (CivitAI's recipe fixes trainTextEncoder: false).", + "Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, 5+20 vs 8+24 blocks), not a quality question.", + "Mixing a \"Flux.2 Klein 9B\" LoRA and a \"Flux.2 Klein 9B-base\" LoRA in one stack is fine — those strings describe the training checkpoint, not incompatible tensor layouts. Set the sampler for the checkpoint actually loaded.", + "Run base-trained LoRAs on the distilled checkpoint when you want speed: HF's klein LoRA guide reports the adapter loads there and typically gives better results than on base, at 4 steps.", + "A base-to-turbo / distillation-extraction LoRA is a whole-model delta, not a content LoRA. Put it on the base checkpoint only, and treat its strength as a steps/CFG dial rather than a style knob. Do not add it on top of the already-distilled checkpoint.", + "Read each LoRA's own published strength range instead of defaulting everything to 1.0 — klein LoRAs in the wild ship very different ones (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, under 0.5 and sometimes 0.1 for the vafipas663 distillation delta).", + "With a consistency or edit LoRA in the stack, excessive strength suppresses the prompt: the consistency LoRA's article says that if edits stop applying, lower it before changing anything else.", + "Prefer runtime LoRA patching over baking a LoRA into the checkpoint: ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein models (4B or 9B) since a March 2026 update — issue #13637, still open.", + "Step-distilled to exactly 4 inference steps and guidance-distilled; HF card example is num_inference_steps=4, guidance_scale=1.0. Its transformer config has guidance_embeds=false. This is the intended inference target: ComfyUI's blog benchmarks it at ~2s and 19.6GB VRAM on an RTX 5090. Not offered as a training target by BFL's docs or CivitAI's recipe." + ] + }, + "pros": [ + "Well established — 2,401 downloads. [source: CivitAI stats]", + "Positively received — 112 thumbs-up with no down-votes. [source: CivitAI stats]", + "Compact at 20 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]" + ], + "file": { + "name": "klein_slider_comic.safetensors", + "size_bytes": 20738136, + "size_mb": 19.78, + "sha256": "b790414cac216841cfcafccace9d4361da991267f0fdb62834facc98fb4b29b7" + }, + "stats": { + "downloads": 2401, + "thumbs_up": 112, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 2401, + "thumbs_up": 112, + "thumbs_down": 0, + "comments": 9, + "approval": "100% positive (112/112)", + "age_days": 224.3, + "downloads_per_day": 10.7, + "rank_by_downloads": 117, + "rank_by_momentum": 53, + "of_total": 200, + "adoption_tier": "Modest", + "momentum": "Steady" + }, + "selection_score": 70.9, + "selection_reasons": [ + "2,401 downloads", + "112 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "1216x832", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1216x832", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "1216x832", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "2048x2048", + "prompt": "" + }, + { + "file": "06_video.mp4", + "type": "video", + "resolution": "768x768", + "prompt": "", + "poster": "06_poster.jpg" + } + ], + "tags": [ + "slider", + "concept", + "comic style", + "drawing" + ], + "scraped_at": "2026-09-05T03:55:45+00:00", + "path": "flux2-klein-9b/klein-realistic-comic-slider" +} diff --git a/the-body-sfw/The_Body_SFW_KK_Flux.2.klein.9B_000005000.safetensors b/the-body-sfw/The_Body_SFW_KK_Flux.2.klein.9B_000005000.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..089b6997ce08e6f74c499221c771ad291ff1e97b --- /dev/null +++ b/the-body-sfw/The_Body_SFW_KK_Flux.2.klein.9B_000005000.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6b9546e3e7fff5b290ab350b785a98421af28e060e3ea42ec171c8fca6b3b29 +size 82866792 diff --git a/the-body-sfw/example_images/01_image.jpeg b/the-body-sfw/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..8a7dd7eb4061fe574bcecca2fe2e69096f3f1c84 --- /dev/null +++ b/the-body-sfw/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6933b007d46b598cd37e14c4e82506c528b22cfbd3ce36924f2f93921e3e262f +size 313983 diff --git a/the-body-sfw/example_images/02_image.jpeg b/the-body-sfw/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..8f3e9c9ab20defeaa59177b5d730e3e1564249ab --- /dev/null +++ b/the-body-sfw/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e5efb5fbade761b2a57c631ba7263f92fd07886729d4faf3781f6de08facde81 +size 166370 diff --git a/the-body-sfw/example_images/03_image.jpeg b/the-body-sfw/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..f4cdac612621a14137e27aba87adf832fb27e6b8 --- /dev/null +++ b/the-body-sfw/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2b77b48b65500fc68cea7fd8be5fcf6a3c364ce79c5486d590091a3bb8ab52e +size 251068 diff --git a/the-body-sfw/example_images/04_image.jpeg b/the-body-sfw/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..9f8ec40c30df0fe66c86f7e23d0e6ee7f745f92c --- /dev/null +++ b/the-body-sfw/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b035f1d44b57f4c446b1a2234af4296f6f0a32e9a69c11d7e2512f5ea52c693f +size 173728 diff --git a/the-body-sfw/example_images/05_image.jpeg b/the-body-sfw/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..0dcf6df93f5a7bc4434c357a9bfb5fba8f030b78 --- /dev/null +++ b/the-body-sfw/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f0c81a8bc241ea1af84044bb565a3a01f7959148b43fb9db74f289826ef1110 +size 172601 diff --git a/the-body-sfw/example_images/06_image.jpeg b/the-body-sfw/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..16d3905af153a1d13da70c1ea0ecee2192ee6f5d --- /dev/null +++ b/the-body-sfw/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6933333586efdfe97a1a0a6ada7fbecc94c4dcb2cb77fe43a182a11c3b3e7c8c +size 220350 diff --git a/the-body-sfw/info.txt b/the-body-sfw/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..adbe9670f050e1b60ebe6de1873b9ca0e2a564da --- /dev/null +++ b/the-body-sfw/info.txt @@ -0,0 +1,237 @@ +═══════════════════════════════════════════════════════════════════════════════ + THE BODY SFW +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model FLUX.2 [klein] 9B + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Uncategorised + Author el_chupanibre + Source https://civitai.com/models/2565139 + Provenance CivitAI · model 2565139 · version 2882334 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File The_Body_SFW_KK_Flux.2.klein.9B_000005000.safetensors + Version KK - F2k9b + Size 79.0 MB + Format SafeTensor / unknown precision + SHA256 b6b9546e3e7fff5b290ab350b785a98421af28e060e3ea42ec171c8fca6b3b29 + Published 2026-04-22 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + › "woman" + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps Distilled + FLUX.2-klein-9B: 4 (HF card example; it is step-distilled to exactly this, + and more steps do not refine further). Base FLUX.2-klein-base-9B: 50 in the + HF card example and in ComfyUI's benchmark; community practice runs lower, + with ~20 for production and 8-12 for fast iteration reported in the CivitAI + consistency-LoRA article., cfg Distilled: guidance_scale 1.0 (HF card). It + is guidance-distilled and its transformer config sets + guidance_embeds=false, so raising CFG degrades rather than improves. Base: + guidance_scale 4.0 (HF card); the CivitAI consistency-LoRA article + recommends 3.5-4.5 for most edits., sampler Not stated by BFL or by + docs.comfy.org, and the official flux2 GitHub repo documents no sampler, + scheduler or shift values either. Community consensus is euler with a + resolution-dependent shifted-Euler / flow-match schedule (ComfyUI's Flux2 + workflows use a scheduler that interpolates shift between base_shift and + max_shift by output resolution). Start from ComfyUI's downloadable Flux.2 + klein workflow template rather than a blog. See unverified., resolution + 1024x1024 in every BFL HF card example and in CivitAI's 9B training recipe + ("resolution": 1024). Non-square ratios work but no official list of + supported dimensions or a stated alignment multiple was found — see + unverified., lora_strength 1.0 is the diffusers/ComfyUI default and what + CivitAI applies to training previews (strength 1.0). Published per-LoRA + guidance varies and should be read from the LoRA, not assumed: the widely + used Klein 9B consistency LoRA's CivitAI article says start at 0.5 within a + 0.2-1.0 range (0.2-0.4 loose, 0.5-0.7 balanced, 0.8-1.0 maximum structural + stability) and warns that too high a strength stops edits from applying; + SOLRICKS' Realistic Detail LoRA card says 0.6-0.8 subtle, 0.8-1.0 stronger, + start 0.8; the vafipas663 distillation-delta LoRA card warns it "produces + static noise at strength > 0.5. And sometimes, you have to go as low as + 0.1.". + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 1,622 + Rating 100% positive (94/94) + Comments 2 + Published 135 days ago + Download rate 4.6/day since release + + Adoption rank #139 of 200 in this collection (Modest) + Momentum rank #125 of 200 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general FLUX.2 [klein] 9B ones, not anything specific to this LoRA. + + General FLUX.2 [klein] 9B rules that apply here (see REFERENCE.txt): + + • Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model + output into model input. There is no CLIP side to route — klein LoRAs + are model-only because the Qwen3 text encoder is not trained (CivitAI's + recipe fixes trainTextEncoder: false). + • Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B + LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, + 5+20 vs 8+24 blocks), not a quality question. + • Mixing a "Flux.2 Klein 9B" LoRA and a "Flux.2 Klein 9B-base" LoRA in + one stack is fine — those strings describe the training checkpoint, not + incompatible tensor layouts. Set the sampler for the checkpoint + actually loaded. + • Run base-trained LoRAs on the distilled checkpoint when you want speed: + HF's klein LoRA guide reports the adapter loads there and typically + gives better results than on base, at 4 steps. + • A base-to-turbo / distillation-extraction LoRA is a whole-model delta, + not a content LoRA. Put it on the base checkpoint only, and treat its + strength as a steps/CFG dial rather than a style knob. Do not add it on + top of the already-distilled checkpoint. + • Read each LoRA's own published strength range instead of defaulting + everything to 1.0 — klein LoRAs in the wild ship very different ones + (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, + under 0.5 and sometimes 0.1 for the vafipas663 distillation delta). + • With a consistency or edit LoRA in the stack, excessive strength + suppresses the prompt: the consistency LoRA's article says that if + edits stop applying, lower it before changing anything else. + • Prefer runtime LoRA patching over baking a LoRA into the checkpoint: + ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein + models (4B or 9B) since a March 2026 update — issue #13637, still open. + • The author never states which weight family this targets. Two different + questions, two different answers. Within 9B (distilled vs base — the + two CivitAI strings): not a real incompatibility. Both load through the + same Flux2KleinPipeline and have the same repo layout; for the 4B pair, + where both transformer configs are public, the distil + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Well established — 1,622 downloads. [source: CivitAI stats] + + Positively received — 94 thumbs-up with no down-votes. [source: CivitAI + stats] + + Explicit trigger word(s) — woman — so the effect can be turned on and + off from the prompt. [source: model metadata] + + Compact at 79 MB — cheap to stack with other LoRAs. [source: file size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Author references NSFW output — review before using in shared work. + [source: description] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Thin documentation — expect trial and error. [source: description + length] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Two different questions, two different answers. Within 9B (distilled vs + base — the two CivitAI strings): not a real incompatibility. Both load + through the same Flux2KleinPipeline and have the same repo layout; for the + 4B pair, where both transformer configs are public, the distilled and base + configs are byte-for-byte identical in every shape field, and the 9B pair + is built the same way. HF's own klein LoRA guide states the workflow + directly: "Distilled models are step-compressed for fast inference; you + train against the base checkpoint and the adapter still loads on the + distilled model afterward", and "Applying the LoRA on the distilled model + typically gives better results than the base model, and it's faster" (that + post is written against 4B, but the family structure is the same). So the + two CivitAI strings record which checkpoint the author trained or previewed + against, not a hard compatibility wall. The real risk is a silent settings + mismatch: load a base-trained LoRA on the distilled checkpoint while + keeping the author's 20-50 step / CFG ~4 preview settings and the output + degrades badly (distilled wants 4 steps / guidance 1.0), or the reverse. + Nothing errors — it just looks wrong. Across sizes (9B vs 4B): a genuine + architectural mismatch, from the published transformer configs — different + hidden width (4096 vs 3072), different block counts (8+24 vs 5+20) and a + different text-conditioning dim (12288 vs 7680, following Qwen3-8B vs + Qwen3-4B). A 4B LoRA cannot patch a 9B model: expect missing keys and shape + errors, not degraded output. The exact ComfyUI error text is unverified. KV + variant: same 9B size and same 4-step distillation, but a distinct + checkpoint with its own pipeline; whether standard klein 9B LoRAs patch it + cleanly is not documented — see unverified. + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + a little side project + The LoRA did NOT learn any faces or any NSFW stuff + Only intention was to create SFWish big booty/hourglass body shots + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 832x1216) + 02_image.jpeg (image, 832x1216) + 03_image.jpeg (image, 1024x1024) + 04_image.jpeg (image, 832x1216) + 05_image.jpeg (image, 832x1216) + 06_image.jpeg (image, 1024x1024) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + 1. Diffusion model -> ComfyUI/models/diffusion_models/. Distilled: + flux-2-klein-9b-fp8.safetensors. Base: flux-2-klein- + base-9b-fp8.safetensors. (BF16 single-file weights + flux-2-klein-9b.safetensors / flux-2-klein-base-9b.safetensors from the + gated HF repos also work if you have ~29GB; community GGUF builds exist + from unsloth and leejet for lower-VRAM setups.) + 2. Text encoder -> + ComfyUI/models/text_encoders/qwen_3_8b_fp8mixed.safetensors (from Comfy- + Org/vae-text-encorder-for-flux-klein-9b). This is the 9B family's encoder; + the 4B family uses qwen_3_4b.safetensors — do not mix them. + 3. VAE -> ComfyUI/models/vae/flux2-vae.safetensors (shared across klein + sizes). + 4. LoRA -> ComfyUI/models/loras/ as a plain .safetensors. + 5. Load with Load Diffusion Model (UNETLoader) for the transformer, + CLIPLoader for the Qwen3 text encoder, VAELoader for the VAE. + docs.comfy.org ships prebuilt Flux.2 klein graphs under Workflow Templates + for both the distilled and base branches — start from those rather than + wiring by hand, since neither the docs nor BFL's repo state + sampler/scheduler/shift in prose; those values live in the template. + 6. Insert LoraLoaderModelOnly between the diffusion-model loader and the + sampler. Model-only is correct for klein: the Qwen3 text encoder is not + trained by the standard klein trainers, so there is no text-encoder half to + patch. + 7. Set steps and guidance to match the checkpoint you actually loaded, not + the one the LoRA was trained on: 4 steps / guidance 1.0 on distilled, + ~20-50 steps / guidance ~4.0 on base. + 8. Note the 9B repos are gated on HuggingFace — you must accept the FLUX + Non-Commercial License on the repo page before any download, including + scripted ones. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 03:55 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/the-body-sfw/metadata.json b/the-body-sfw/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..37e1d26f9ef0e67732cb860f66e1d3d0bea371ae --- /dev/null +++ b/the-body-sfw/metadata.json @@ -0,0 +1,132 @@ +{ + "name": "The Body SFW", + "slug": "the-body-sfw", + "base_model": "FLUX.2 [klein] 9B", + "media_type": "image", + "model_id": 2565139, + "version_id": 2882334, + "version_name": "KK - F2k9b", + "source": "civitai", + "url": "https://civitai.com/models/2565139", + "repo_id": null, + "file_path": null, + "author": "el_chupanibre", + "category": { + "key": "other", + "label": "Uncategorised" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [ + "woman" + ], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain LoRAs by stringing LoraLoaderModelOnly nodes in series, model output into model input. There is no CLIP side to route — klein LoRAs are model-only because the Qwen3 text encoder is not trained (CivitAI's recipe fixes trainTextEncoder: false).", + "Only stack LoRAs of the same parameter size. 9B stacks with 9B; a 4B LoRA in a 9B stack is an architectural mismatch (3072 vs 4096 hidden, 5+20 vs 8+24 blocks), not a quality question.", + "Mixing a \"Flux.2 Klein 9B\" LoRA and a \"Flux.2 Klein 9B-base\" LoRA in one stack is fine — those strings describe the training checkpoint, not incompatible tensor layouts. Set the sampler for the checkpoint actually loaded.", + "Run base-trained LoRAs on the distilled checkpoint when you want speed: HF's klein LoRA guide reports the adapter loads there and typically gives better results than on base, at 4 steps.", + "A base-to-turbo / distillation-extraction LoRA is a whole-model delta, not a content LoRA. Put it on the base checkpoint only, and treat its strength as a steps/CFG dial rather than a style knob. Do not add it on top of the already-distilled checkpoint.", + "Read each LoRA's own published strength range instead of defaulting everything to 1.0 — klein LoRAs in the wild ship very different ones (0.5 start for the consistency LoRA, 0.8 for SOLRICKS' detail LoRA, under 0.5 and sometimes 0.1 for the vafipas663 distillation delta).", + "With a consistency or edit LoRA in the stack, excessive strength suppresses the prompt: the consistency LoRA's article says that if edits stop applying, lower it before changing anything else.", + "Prefer runtime LoRA patching over baking a LoRA into the checkpoint: ComfyUI's Model Merge / Model Save nodes have not detected Flux2 Klein models (4B or 9B) since a March 2026 update — issue #13637, still open.", + "The author never states which weight family this targets. Two different questions, two different answers. Within 9B (distilled vs base — the two CivitAI strings): not a real incompatibility. Both load through the same Flux2KleinPipeline and have the same repo layout; for the 4B pair, where both transformer configs are public, the distil" + ] + }, + "pros": [ + "Well established — 1,622 downloads. [source: CivitAI stats]", + "Positively received — 94 thumbs-up with no down-votes. [source: CivitAI stats]", + "Explicit trigger word(s) — woman — so the effect can be turned on and off from the prompt. [source: model metadata]", + "Compact at 79 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "Author references NSFW output — review before using in shared work. [source: description]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Thin documentation — expect trial and error. [source: description length]" + ], + "file": { + "name": "The_Body_SFW_KK_Flux.2.klein.9B_000005000.safetensors", + "size_bytes": 82866792, + "size_mb": 79.03, + "sha256": "b6b9546e3e7fff5b290ab350b785a98421af28e060e3ea42ec171c8fca6b3b29" + }, + "stats": { + "downloads": 1622, + "thumbs_up": 94, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 1622, + "thumbs_up": 94, + "thumbs_down": 0, + "comments": 2, + "approval": "100% positive (94/94)", + "age_days": 135.4, + "downloads_per_day": 4.62, + "rank_by_downloads": 139, + "rank_by_momentum": 125, + "of_total": 200, + "adoption_tier": "Modest", + "momentum": "Long tail" + }, + "selection_score": 70.8, + "selection_reasons": [ + "1,622 downloads", + "94 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "832x1216", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + } + ], + "tags": [ + "concept" + ], + "scraped_at": "2026-09-05T03:55:55+00:00", + "path": "flux2-klein-9b/the-body-sfw" +}