Add 12 LoRA(s): slavekini-aka-slave-leia-outfit-clothing-krea2-anima-il, bartstyle-blue-archive-art-style-lora-anime-flat-color, 2000s-analog-core, granblue-fantasy-style…

.gitattributes

@@ -101,3 +101,74 @@ hyperdetailed-colored-pencil/example_images/03_image.jpeg filter=lfs diff=lfs me
 hyperdetailed-colored-pencil/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
 hyperdetailed-colored-pencil/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
 hyperdetailed-colored-pencil/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+slavekini-aka-slave-leia-outfit-clothing-krea2-anima-il/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+slavekini-aka-slave-leia-outfit-clothing-krea2-anima-il/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+slavekini-aka-slave-leia-outfit-clothing-krea2-anima-il/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+slavekini-aka-slave-leia-outfit-clothing-krea2-anima-il/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+slavekini-aka-slave-leia-outfit-clothing-krea2-anima-il/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+slavekini-aka-slave-leia-outfit-clothing-krea2-anima-il/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+bartstyle-blue-archive-art-style-lora-anime-flat-color/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+bartstyle-blue-archive-art-style-lora-anime-flat-color/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+bartstyle-blue-archive-art-style-lora-anime-flat-color/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+bartstyle-blue-archive-art-style-lora-anime-flat-color/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+bartstyle-blue-archive-art-style-lora-anime-flat-color/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+bartstyle-blue-archive-art-style-lora-anime-flat-color/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+2000s-analog-core/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+2000s-analog-core/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+2000s-analog-core/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+2000s-analog-core/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+2000s-analog-core/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+2000s-analog-core/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+granblue-fantasy-style/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+granblue-fantasy-style/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+granblue-fantasy-style/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+granblue-fantasy-style/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+granblue-fantasy-style/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+granblue-fantasy-style/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+famegrid-krea-2-z-image-qwen-realism-lora/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+famegrid-krea-2-z-image-qwen-realism-lora/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+famegrid-krea-2-z-image-qwen-realism-lora/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+famegrid-krea-2-z-image-qwen-realism-lora/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+famegrid-krea-2-z-image-qwen-realism-lora/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+famegrid-krea-2-z-image-qwen-realism-lora/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-space-marines-warhammer-40k-flux-pony-illustrious-anima/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-space-marines-warhammer-40k-flux-pony-illustrious-anima/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-space-marines-warhammer-40k-flux-pony-illustrious-anima/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-space-marines-warhammer-40k-flux-pony-illustrious-anima/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-space-marines-warhammer-40k-flux-pony-illustrious-anima/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-space-marines-warhammer-40k-flux-pony-illustrious-anima/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea2-textfusion-refusal-reduction-lora/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea2-textfusion-refusal-reduction-lora/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea2-textfusion-refusal-reduction-lora/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea2-textfusion-refusal-reduction-lora/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea2-textfusion-refusal-reduction-lora/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea2-textfusion-refusal-reduction-lora/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+background-detail-enhancer/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+background-detail-enhancer/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+background-detail-enhancer/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+background-detail-enhancer/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+background-detail-enhancer/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+background-detail-enhancer/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+aesthetic-amateur-photo/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+aesthetic-amateur-photo/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+aesthetic-amateur-photo/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+aesthetic-amateur-photo/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+aesthetic-amateur-photo/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+aesthetic-amateur-photo/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ultrareal-krea2-klein9b/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ultrareal-krea2-klein9b/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ultrareal-krea2-klein9b/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ultrareal-krea2-klein9b/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ultrareal-krea2-klein9b/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ultrareal-krea2-klein9b/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+new-fantasy-core-ill-flux-pony-sdxl-zit-detailer/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+new-fantasy-core-ill-flux-pony-sdxl-zit-detailer/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+new-fantasy-core-ill-flux-pony-sdxl-zit-detailer/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+new-fantasy-core-ill-flux-pony-sdxl-zit-detailer/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+new-fantasy-core-ill-flux-pony-sdxl-zit-detailer/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+new-fantasy-core-ill-flux-pony-sdxl-zit-detailer/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+satoshi-urushihara-style/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+satoshi-urushihara-style/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+satoshi-urushihara-style/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+satoshi-urushihara-style/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+satoshi-urushihara-style/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text

2000s-analog-core/info.txt

@@ -0,0 +1,311 @@
+═══════════════════════════════════════════════════════════════════════════════
+  2000S ANALOG CORE
+═══════════════════════════════════════════════════════════════════════════════
+
+───────────────────────────────────────────────────────────────────────────────
+  IDENTITY
+───────────────────────────────────────────────────────────────────────────────
+  Base model        Krea 2
+  Generation mode   not declared by the author
+  Checkpoint        not determined — verify before use
+  Category          Visual style
+  Author            Danrisi
+  Source            https://civitai.com/models/1134895
+  Provenance        CivitAI · model 1134895 · version 3103914
+
+───────────────────────────────────────────────────────────────────────────────
+  WEIGHTS
+───────────────────────────────────────────────────────────────────────────────
+  File              vhs_krea2.safetensors
+  Version           v1.0 Krea2
+  Size              109.0 MB
+  Format            SafeTensor / unknown precision
+  SHA256            cef79e53f88941e29ea7211543f624981207e32f15abef80c56fea3d0676e533
+  Published         2026-07-06
+
+───────────────────────────────────────────────────────────────────────────────
+  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
+───────────────────────────────────────────────────────────────────────────────
+  Taken verbatim from the author's description:
+
+  LoRA strength       0.9
+  Sampler steps       40
+  Resolution          704×704
+
+───────────────────────────────────────────────────────────────────────────────
+  ADOPTION & POPULARITY
+───────────────────────────────────────────────────────────────────────────────
+  Downloads               30,523
+  Rating                  100% positive (2334/2336)
+  Comments                0
+  Published               60 days ago
+  Download rate           89.9/day since release
+
+  Adoption rank           #18 of 150 in this collection  (High)
+  Momentum rank           #30 of 150 by download rate  (Trending)
+
+  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 Krea 2 ones, not anything specific to this LoRA.
+
+  General Krea 2 rules that apply here (see REFERENCE.txt):
+
+  • The most important rule is not about chaining but about which
+    checkpoint you chain onto: the official README says in bold "TRAIN on
+    Raw and RUN on Turbo", and "We highly recommend using RAW for training
+    LoRAs and applying them on Turbo for inference." LoRAs trained on Raw
+    are stated to work well on Turbo.
+  • Chain with LoraLoaderModelOnly nodes in series (model -> model). Krea 2
+    LoRAs are DiT-only, so there is no CLIP side to chain and the text-
+    encoder branch stays untouched however many LoRAs you add.
+  • Krea 2 LoRAs stack in practice and the community treats multi-LoRA as
+    routine - style + character, style + slider, realism + character.
+    Several CivitAI Krea 2 authors state their LoRAs stack cleanly; one
+    writes "I recently stacked over 20 of my own LoRAs on a single
+    generation and the output still came out great."
+  • Reduce per-LoRA strength when stacking. CivitAI Krea 2 authors converge
+    on dropping from ~1.0 to roughly 0.6-0.9 once more than one adapter is
+    loaded. The CivitAI 12GB training guide also notes rank affects
+    composability: "rank 8 produced a lighter style influence that was
+    easier to stack with other LoRAs, rank 16 provided a useful balance
+    between concept strength and composability, and rank 32 was better
+    suited to a precise object or concept intended to take priority over
+    other stacked LoRAs."
+  • Accelerator LoRAs are position-sensitive in intent, not in node order:
+    a Raw-to-Turbo delta adapter belongs on the Raw checkpoint, a
+    sub-8-step LoRA belongs on Turbo (or on Raw already carrying a Raw-to-
+    Turbo adapter). Chaining Raw -> Raw-to-Turbo adapter -> 4-step LoRA is
+    explicitly recommended by the 4-step LoRA's author.
+  • Whatever you stack, the schedule must stay Turbo's: cfg 1.0 in ComfyUI
+    (guidance 0.0 in diffusers) and mu/shift 1.15. The 4-step LoRA's card
+    says "Keep mu = 1.15. The training targets are anchored to that grid; a
+    different shift moves the [sampler off them]"; the same reasoning
+    applies to any Turbo-trained adapter in the chain.
+  • Reference/edit LoRAs are designed to be stacked under ordinary subject
+    and style LoRAs - the Identity Edit card states "Composes with your
+    LoRAs: character/body/style LoRAs stack on top and steer the prior" and
+    recommends a stacked subject LoRA when unusual subjects drift toward
+    the base prior.
+  • Krea 2 LoRAs cannot be chained with LoRAs for any other base model.
+    Many CivitAI pages carry multi-model titles like "(Krea-2 + ZIT)" or
+    "[Flux | ZIB | Krea2]" - those are separate files for separate
+    architectures shipped under one model page, not a combined adapter.
+  • To combine two full Krea 2 checkpoints rather than adapters, ComfyUI
+    ships a dedicated ModelMergeKrea2 node (added in v0.27.0) with per-
+    section weights for first., tmlp., txtmlp., tproj.,
+    txtfusion.projector, txtfusion.layerwise_blocks.0-1,
+    txtfusion.refiner_blocks.0-1, blocks.0-27 and last.
+  • The author never states which weight family this targets. Silent
+    degradation, never a hard error, between Raw and Turbo. The two share
+    one architecture - ai-toolkit's config is commented "The reference
+    'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)"
+    - so every LoRA tensor shape matches on both and ComfyUI/diffuser
+
+───────────────────────────────────────────────────────────────────────────────
+  PROS
+───────────────────────────────────────────────────────────────────────────────
+  + Widely adopted — 30,523 downloads. [source: CivitAI stats]
+  + Positively received — 2,334 thumbs-up, 2 down. [source: CivitAI stats]
+  + Compact at 109 MB — cheap to stack with other LoRAs. [source: file
+    size]
+  + Author published concrete recommended settings. [source: description]
+
+───────────────────────────────────────────────────────────────────────────────
+  CONS & CAVEATS
+───────────────────────────────────────────────────────────────────────────────
+  - 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
+───────────────────────────────────────────────────────────────────────────────
+  Silent degradation, never a hard error, between Raw and Turbo. The two
+  share one architecture - ai-toolkit's config is commented "The reference
+  'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)" - so
+  every LoRA tensor shape matches on both and ComfyUI/diffusers will load an
+  adapter against either without complaint. The lvladikov 4-step LoRA card
+  states this for the reverse direction: "Every layer it targets also exists
+  in Krea 2 Raw, so it will load there without complaint - but that is a side
+  effect of the shared architecture, not a supported mode", and reports
+  mixed, subject-dependent results with malformed subjects (duplicated heads,
+  fused limbs) at low step counts, needing 14+ steps with Raw's CFG on before
+  output is coherent. The intended direction (train on Raw, run on Turbo) is
+  officially blessed. The more common practical failure is a settings
+  mismatch rather than a weight mismatch: running a Turbo-targeted LoRA at
+  Raw's 52 steps + CFG 3.5, or leaving CFG enabled on Turbo, degrades output
+  with no warning. Loading a Krea 2 LoRA against a different model family
+  entirely (Flux, Qwen-Image, Z-Image) is also not an error - ComfyUI logs
+  "lora key not loaded: {key}" per key and generates as if no LoRA were
+  applied.
+
+───────────────────────────────────────────────────────────────────────────────
+  AUTHOR'S DESCRIPTION
+───────────────────────────────────────────────────────────────────────────────
+  Z-Image I used at strength 0.9 with res_2s_ode + beta. Workflow you can
+  parse from my images Qwen Edition: Here everything works perfectly,
+  timestamps, text - excellent. Images with women i generate with my new lora
+  https://civitai.com/models/1862761?modelVersionId=2338791 V3 Update:
+  Enhanced Timestamps
+  - Improved realism for on-screen display times and date stamps.
+
+  - Face Fixes
+  - Reduced the repetition of identical faces across generated frames,
+  enhancing variety.
+
+  - Expanded Dataset
+  - Now trained on VHS, VHS-C, and Hi8 footage to broaden the analog
+  aesthetic.
+
+  Usage & Tips
+  - Low-Res VHS Look:
+  - Use prompts with phrases like “amateur quality” and “low resolution” to
+  push your render toward that classic, fuzzy VHS vibe.
+
+  - Target ~0.5 megapixels (e.g. 704×704, 608 x 816) if you want reinforce
+  extreme lo-fi aesthetic.
+
+  - Higher-Res Hi8 Style:
+  - To achieve the “Hi8” look (cleaner but still analog), render at ~1
+  megapixel (e.g. 896 x 1152 or 1024×1024).
+
+  - You can go up to ~2 megapixels if you want more detail without losing the
+  retro feel.
+
+  - Recommended Trigger:
+  - The only trigger word you truly need for this model is “v8s” .
+
+  - Avoid mixing too many triggers, as that can introduce confusion and stray
+  from the classic 2000s analog style. V2 Update:
+
+  - Removed digital camera images from the dataset and added more footage
+  from camcorders.
+
+  - Fixed the issue where the same woman appeared in all images.
+
+  - Tried improving the sexy overlay text on footage, but it only generates
+  well sometimes.
+
+  Can’t say I’m fully happy with this version, but it’s not bad. Already
+  planning a new update with more features soon! Also thinking about updating
+  the Mavica LoRA.
+  This keeps your style while improving clarity and readability. Let me know
+  if you want any tweaks P.S: but what i really like in this lora, that it's
+  for some reason generate almost always perfect hands Step back into the
+  early 2000s, where Myspace was king, profile pictures were taken with a
+  digital potato, and everything felt just a little bit off (in the best
+  way). This LoRA captures that lo-fi, analog aesthetic of forgotten VHS
+  tapes and grainy digital cameras (yeah, i mix them in one LoRa). Perfect
+  for those who miss the charm of pixelated edges and questionable lighting
+  choices. No, it won't resurrect your old Myspace profile - sadly - but it
+  will bring back all those nostalgic vibes. Warning: May cause a sudden urge
+  to listen to emo playlists P.S.S: only v8s is needed in trigger words,
+  everything else is optional to enhance effect Usage as usual: dpmpp2m +
+  beta + 40 steps + 2.5-3 guidance I'm using with UltraReal checkpoint , but
+  works good with default flux.dev too P.S: sometimes can be some timestamps
+  or other features from vhs tapes (will be fixed in next versions). I just
+  wanted to test some techniques on this LoRa and result is pretty solid for
+  29 images in dataset. Also noticed one bug, that female faces often looks
+  similar, will fix in next release. But LoRas wit characters works good and
+  fix this problem
+
+───────────────────────────────────────────────────────────────────────────────
+  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, 1472x1024)
+  02_image.jpeg  (image, 1472x1024)
+  03_image.jpeg  (image, 1472x1024)
+  04_image.jpeg  (image, 1024x1472)
+  05_image.jpeg  (image, 1408x1088)
+  06_image.jpeg  (image, 1408x1088)
+
+───────────────────────────────────────────────────────────────────────────────
+  HOW TO USE (COMFYUI)
+───────────────────────────────────────────────────────────────────────────────
+  1. Update ComfyUI to at least v0.26.0 (released 2026-06-23), which added
+  local Krea2 support. For reference-image / style-reference / edit LoRAs you
+  need at least v0.29.0 (2026-07-29), which added regular and timestep-zero
+  reference images for Krea 2. ModelMergeKrea2 arrived in v0.27.0
+  (2026-06-30).
+
+  2. Put the base stack in place (files from huggingface.co/Comfy-Org/Krea-2,
+  which is not gated):
+     - ComfyUI/models/diffusion_models/krea2_turbo_fp8_scaled.safetensors
+  (documented recommended default; the repo also holds krea2_turbo_bf16,
+  _int8_convrot, _mxfp8, _nvfp4 and krea2_raw_bf16 / _fp8_scaled /
+  _int8_convrot)
+     - ComfyUI/models/text_encoders/qwen3vl_4b_fp8_scaled.safetensors (bf16
+  also available)
+     - ComfyUI/models/vae/qwen_image_vae.safetensors
+
+  3. Put the LoRA .safetensors in ComfyUI/models/loras/ - flat, no per-model
+  subfolder required.
+
+  4. Build or load the graph. The official image_krea2_turbo_t2i.json
+  template is: UNETLoader (krea2_turbo_fp8_scaled.safetensors) ->
+  LoraLoaderModelOnly -> KSampler; CLIPLoader
+  (qwen3vl_4b_fp8_scaled.safetensors, type "krea2") -> text encode ->
+  KSampler positive, with ConditioningZeroOut into negative; VAELoader
+  (qwen_image_vae.safetensors) -> VAEDecode -> SaveImage; EmptyLatentImage
+  1024x1024 driven by ResolutionSelector. An int8 sibling template
+  (image_krea2_turbo_t2i_int8.json) also ships.
+
+  5. Load the LoRA with LoraLoaderModelOnly, not LoraLoader - both official
+  templates do. Krea 2 LoRAs patch the DiT only; note that the DiT's own
+  text_fusion / txtfusion blocks are a normal LoRA target (the official style
+  LoRAs patch them), but the Qwen3-VL text encoder itself is not. Set
+  strength 1.0 for Krea's official style LoRAs (the shipped template uses
+  0.8) and add the LoRA's trigger word to the prompt.
+
+  6. Sampler settings: steps 8, cfg 1, sampler euler, scheduler simple,
+  denoise 1.0. Do not add a ModelSampling node for plain t2i - ComfyUI
+  already applies shift 1.15 from Krea2.sampling_settings.
+
+  7. For style reference, use the dedicated template: UNETLoader
+  krea2_turbo_int8_convrot.safetensors, LoraLoaderModelOnly
+  krea2_style_reference.safetensors at 1.0, ModelSamplingFlux (1.15, 0.5),
+  TextEncodeQwenImageEditPlus fed by LoadImage,
+  FluxKontextMultiReferenceLatentMethod set to index_timestep_zero, and
+  SamplerCustomAdvanced with KSamplerSelect euler + BasicScheduler simple/8 +
+  CFGGuider cfg 1.
+
+  8. For instruction editing (Krea 2 Identity Edit), stock nodes are not
+  enough: install the ComfyUI-Krea2Edit node pack, which ships its own
+  workflows, and run Turbo at 8-12 steps / cfg 1 (or Raw at 20 steps / cfg 3
+  for removals) with LoRA strength 1.0.
+
+  9. Low-VRAM alternative: install city96's ComfyUI-GGUF and use a community
+  GGUF build (realrebelai/KREA-2_GGUFs, quantized from Turbo;
+  molbal/krea2-gguf, quantized from Raw) with the Unet Loader (GGUF) node in
+  place of UNETLoader. There is no official GGUF release. LoRA loading is
+  unchanged.
+
+───────────────────────────────────────────────────────────────────────────────
+  Scraped by lorakit on 2026-09-05 00:11 UTC.
+  Stats and description are the author's; pros/cons are derived from the
+  evidence tagged beside each line.
+───────────────────────────────────────────────────────────────────────────────

2000s-analog-core/metadata.json

@@ -0,0 +1,141 @@
+{
+  "name": "2000s Analog Core",
+  "slug": "2000s-analog-core",
+  "base_model": "Krea 2",
+  "media_type": "image",
+  "model_id": 1134895,
+  "version_id": 3103914,
+  "version_name": "v1.0 Krea2",
+  "source": "civitai",
+  "url": "https://civitai.com/models/1134895",
+  "repo_id": null,
+  "file_path": null,
+  "author": "Danrisi",
+  "category": {
+    "key": "style",
+    "label": "Visual style"
+  },
+  "modalities": [],
+  "primary_modality": "unspecified",
+  "checkpoint_family": "unknown",
+  "trigger_words": [],
+  "recommended_settings": {
+    "LoRA strength": "0.9",
+    "Sampler steps": "40",
+    "Resolution": "704×704"
+  },
+  "chaining": {
+    "author": {
+      "stacks": "not stated",
+      "notes": "",
+      "avoid": [],
+      "requires": [],
+      "order": "",
+      "strength": "",
+      "stated": false
+    },
+    "general_rules": [
+      "The most important rule is not about chaining but about which checkpoint you chain onto: the official README says in bold \"TRAIN on Raw and RUN on Turbo\", and \"We highly recommend using RAW for training LoRAs and applying them on Turbo for inference.\" LoRAs trained on Raw are stated to work well on Turbo.",
+      "Chain with LoraLoaderModelOnly nodes in series (model -> model). Krea 2 LoRAs are DiT-only, so there is no CLIP side to chain and the text-encoder branch stays untouched however many LoRAs you add.",
+      "Krea 2 LoRAs stack in practice and the community treats multi-LoRA as routine - style + character, style + slider, realism + character. Several CivitAI Krea 2 authors state their LoRAs stack cleanly; one writes \"I recently stacked over 20 of my own LoRAs on a single generation and the output still came out great.\"",
+      "Reduce per-LoRA strength when stacking. CivitAI Krea 2 authors converge on dropping from ~1.0 to roughly 0.6-0.9 once more than one adapter is loaded. The CivitAI 12GB training guide also notes rank affects composability: \"rank 8 produced a lighter style influence that was easier to stack with other LoRAs, rank 16 provided a useful balance between concept strength and composability, and rank 32 was better suited to a precise object or concept intended to take priority over other stacked LoRAs.\"",
+      "Accelerator LoRAs are position-sensitive in intent, not in node order: a Raw-to-Turbo delta adapter belongs on the Raw checkpoint, a sub-8-step LoRA belongs on Turbo (or on Raw already carrying a Raw-to-Turbo adapter). Chaining Raw -> Raw-to-Turbo adapter -> 4-step LoRA is explicitly recommended by the 4-step LoRA's author.",
+      "Whatever you stack, the schedule must stay Turbo's: cfg 1.0 in ComfyUI (guidance 0.0 in diffusers) and mu/shift 1.15. The 4-step LoRA's card says \"Keep mu = 1.15. The training targets are anchored to that grid; a different shift moves the [sampler off them]\"; the same reasoning applies to any Turbo-trained adapter in the chain.",
+      "Reference/edit LoRAs are designed to be stacked under ordinary subject and style LoRAs - the Identity Edit card states \"Composes with your LoRAs: character/body/style LoRAs stack on top and steer the prior\" and recommends a stacked subject LoRA when unusual subjects drift toward the base prior.",
+      "Krea 2 LoRAs cannot be chained with LoRAs for any other base model. Many CivitAI pages carry multi-model titles like \"(Krea-2 + ZIT)\" or \"[Flux | ZIB | Krea2]\" - those are separate files for separate architectures shipped under one model page, not a combined adapter.",
+      "To combine two full Krea 2 checkpoints rather than adapters, ComfyUI ships a dedicated ModelMergeKrea2 node (added in v0.27.0) with per-section weights for first., tmlp., txtmlp., tproj., txtfusion.projector, txtfusion.layerwise_blocks.0-1, txtfusion.refiner_blocks.0-1, blocks.0-27 and last.",
+      "The author never states which weight family this targets. Silent degradation, never a hard error, between Raw and Turbo. The two share one architecture - ai-toolkit's config is commented \"The reference 'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)\" - so every LoRA tensor shape matches on both and ComfyUI/diffuser"
+    ]
+  },
+  "pros": [
+    "Widely adopted — 30,523 downloads. [source: CivitAI stats]",
+    "Positively received — 2,334 thumbs-up, 2 down. [source: CivitAI stats]",
+    "Compact at 109 MB — cheap to stack with other LoRAs. [source: file size]",
+    "Author published concrete recommended settings. [source: description]"
+  ],
+  "cons": [
+    "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": "vhs_krea2.safetensors",
+    "size_bytes": 114324720,
+    "size_mb": 109.03,
+    "sha256": "cef79e53f88941e29ea7211543f624981207e32f15abef80c56fea3d0676e533"
+  },
+  "stats": {
+    "downloads": 30523,
+    "thumbs_up": 2334,
+    "thumbs_down": 2
+  },
+  "popularity": {
+    "downloads": 30523,
+    "thumbs_up": 2334,
+    "thumbs_down": 2,
+    "comments": 0,
+    "approval": "100% positive (2334/2336)",
+    "age_days": 60.5,
+    "downloads_per_day": 89.92,
+    "rank_by_downloads": 18,
+    "rank_by_momentum": 30,
+    "of_total": 150,
+    "adoption_tier": "High",
+    "momentum": "Trending"
+  },
+  "selection_score": 58531.0,
+  "selection_reasons": [
+    "30,523 downloads",
+    "2,334 thumbs-up",
+    "top-ranked for its category"
+  ],
+  "example_media": [
+    {
+      "file": "01_image.jpeg",
+      "type": "image",
+      "resolution": "1472x1024",
+      "prompt": ""
+    },
+    {
+      "file": "02_image.jpeg",
+      "type": "image",
+      "resolution": "1472x1024",
+      "prompt": ""
+    },
+    {
+      "file": "03_image.jpeg",
+      "type": "image",
+      "resolution": "1472x1024",
+      "prompt": ""
+    },
+    {
+      "file": "04_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1472",
+      "prompt": ""
+    },
+    {
+      "file": "05_image.jpeg",
+      "type": "image",
+      "resolution": "1408x1088",
+      "prompt": ""
+    },
+    {
+      "file": "06_image.jpeg",
+      "type": "image",
+      "resolution": "1408x1088",
+      "prompt": ""
+    }
+  ],
+  "tags": [
+    "vhs",
+    "2000s",
+    "crt",
+    "old",
+    "90s",
+    "tape",
+    "style",
+    "vintage"
+  ],
+  "scraped_at": "2026-09-05T00:11:54+00:00",
+  "path": "krea-2/2000s-analog-core"
+}

2000s-analog-core/vhs_krea2.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cef79e53f88941e29ea7211543f624981207e32f15abef80c56fea3d0676e533
+size 114324720

aesthetic-amateur-photo/Krea_Amateur_V4.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ad3b5092c99b4a1696df4ab07407775309da6c21d36c9a506f6b5ff153552ec
+size 228588912

aesthetic-amateur-photo/info.txt

@@ -0,0 +1,333 @@
+═══════════════════════════════════════════════════════════════════════════════
+  AESTHETIC AMATEUR PHOTO
+═══════════════════════════════════════════════════════════════════════════════
+
+───────────────────────────────────────────────────────────────────────────────
+  IDENTITY
+───────────────────────────────────────────────────────────────────────────────
+  Base model        Krea 2
+  Generation mode   not declared by the author
+  Checkpoint        not determined — verify before use
+  Category          Photorealism
+  Author            powerpuff
+  Source            https://civitai.com/models/689192
+  Provenance        CivitAI · model 689192 · version 3284331
+
+───────────────────────────────────────────────────────────────────────────────
+  WEIGHTS
+───────────────────────────────────────────────────────────────────────────────
+  File              Krea_Amateur_V4.safetensors
+  Version           Krea V4
+  Size              218.0 MB
+  Format            SafeTensor / unknown precision
+  SHA256            0ad3b5092c99b4a1696df4ab07407775309da6c21d36c9a506f6b5ff153552ec
+  Published         2026-08-31
+
+───────────────────────────────────────────────────────────────────────────────
+  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 Turbo: 8
+  (official README and HF card CLI example; the ComfyUI template ships 8).
+  Raw: 52 (official README). The krea-2 CLI's generic default is 28.
+  Community Turbo LoRAs commonly suggest 8-12 (e.g. the Identity Edit card
+  recommends Turbo at 8-12, or Raw at 20 for removals). A community 4-step
+  LoRA on Turbo makes 4 valid; its author calls 2 steps an out-of-spec
+  preview mode., cfg Turbo: CFG disabled - guidance_scale 0.0 in diffusers
+  and --cfg 0.0 in the krea-2 CLI, which is cfg 1.0 in ComfyUI's KSampler
+  (the official template ships cfg 1). These are the same thing; do not set
+  ComfyUI cfg to 0. Raw: CFG 3.5 (official README; CLI generic default 4.5).
+  Negative prompting on Turbo is effectively unused - the official templates
+  route ConditioningZeroOut into the negative slot., sampler ComfyUI:
+  KSampler with sampler euler, scheduler simple, denoise 1.0 (verified in
+  image_krea2_turbo_t2i.json: [seed, 'randomize', 8, 1, 'euler', 'simple',
+  1]). Timestep shift mu = 1.15 for Turbo; ComfyUI bakes this in via
+  Krea2.sampling_settings = {"multiplier": 1.0, "shift": 1.15}, so the plain
+  t2i template needs no ModelSampling node, while the style-reference
+  template adds ModelSamplingFlux at max_shift 1.15 / base_shift 0.5. The CLI
+  exposes --mu 1.15 for Turbo; Raw uses a resolution-derived mu interpolated
+  between --y1 0.5 and --y2 1.15 (ai-toolkit mirrors this as base_shift 0.5 /
+  max_shift 1.15 over image_seq_len 256-6400, exponential time shift).,
+  resolution 1024x1024 default. Turbo: 1K-2K (the official CLI example
+  generates 2048x2048; ComfyUI's ResolutionSelector defaults to 1 megapixel,
+  set 2.0 for 2K). Raw: "trained to generate upto 1k resolution".
+  Width/height are padded up to a multiple of 16., lora_strength 1.0 for
+  Krea's nine official style LoRAs (docs.comfy.org lists Recommended Strength
+  1.0 for every one), though the shipped t2i template's LoraLoaderModelOnly
+  is set to 0.8. The style-reference LoRA is used at 1.0 in its template. The
+  4-step distill LoRA documents 1.0 as its trained point, 1.0-1.5 as coherent
+  extrapolation and above ~1.5 as speckle artefacting. Community subject
+  LoRAs are typically recommended around 0.8-1.2 and dropped toward 0.6-0.9
+  when several are stacked..
+
+───────────────────────────────────────────────────────────────────────────────
+  ADOPTION & POPULARITY
+───────────────────────────────────────────────────────────────────────────────
+  Downloads               32,058
+  Rating                  100% positive (1413/1415)
+  Comments                0
+  Published               4 days ago
+  Download rate           326.4/day since release
+
+  Adoption rank           #16 of 150 in this collection  (High)
+  Momentum rank           #5 of 150 by download rate  (Hot · 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 Krea 2 ones, not anything specific to this LoRA.
+
+  General Krea 2 rules that apply here (see REFERENCE.txt):
+
+  • The most important rule is not about chaining but about which
+    checkpoint you chain onto: the official README says in bold "TRAIN on
+    Raw and RUN on Turbo", and "We highly recommend using RAW for training
+    LoRAs and applying them on Turbo for inference." LoRAs trained on Raw
+    are stated to work well on Turbo.
+  • Chain with LoraLoaderModelOnly nodes in series (model -> model). Krea 2
+    LoRAs are DiT-only, so there is no CLIP side to chain and the text-
+    encoder branch stays untouched however many LoRAs you add.
+  • Krea 2 LoRAs stack in practice and the community treats multi-LoRA as
+    routine - style + character, style + slider, realism + character.
+    Several CivitAI Krea 2 authors state their LoRAs stack cleanly; one
+    writes "I recently stacked over 20 of my own LoRAs on a single
+    generation and the output still came out great."
+  • Reduce per-LoRA strength when stacking. CivitAI Krea 2 authors converge
+    on dropping from ~1.0 to roughly 0.6-0.9 once more than one adapter is
+    loaded. The CivitAI 12GB training guide also notes rank affects
+    composability: "rank 8 produced a lighter style influence that was
+    easier to stack with other LoRAs, rank 16 provided a useful balance
+    between concept strength and composability, and rank 32 was better
+    suited to a precise object or concept intended to take priority over
+    other stacked LoRAs."
+  • Accelerator LoRAs are position-sensitive in intent, not in node order:
+    a Raw-to-Turbo delta adapter belongs on the Raw checkpoint, a
+    sub-8-step LoRA belongs on Turbo (or on Raw already carrying a Raw-to-
+    Turbo adapter). Chaining Raw -> Raw-to-Turbo adapter -> 4-step LoRA is
+    explicitly recommended by the 4-step LoRA's author.
+  • Whatever you stack, the schedule must stay Turbo's: cfg 1.0 in ComfyUI
+    (guidance 0.0 in diffusers) and mu/shift 1.15. The 4-step LoRA's card
+    says "Keep mu = 1.15. The training targets are anchored to that grid; a
+    different shift moves the [sampler off them]"; the same reasoning
+    applies to any Turbo-trained adapter in the chain.
+  • Reference/edit LoRAs are designed to be stacked under ordinary subject
+    and style LoRAs - the Identity Edit card states "Composes with your
+    LoRAs: character/body/style LoRAs stack on top and steer the prior" and
+    recommends a stacked subject LoRA when unusual subjects drift toward
+    the base prior.
+  • Krea 2 LoRAs cannot be chained with LoRAs for any other base model.
+    Many CivitAI pages carry multi-model titles like "(Krea-2 + ZIT)" or
+    "[Flux | ZIB | Krea2]" - those are separate files for separate
+    architectures shipped under one model page, not a combined adapter.
+  • To combine two full Krea 2 checkpoints rather than adapters, ComfyUI
+    ships a dedicated ModelMergeKrea2 node (added in v0.27.0) with per-
+    section weights for first., tmlp., txtmlp., tproj.,
+    txtfusion.projector, txtfusion.layerwise_blocks.0-1,
+    txtfusion.refiner_blocks.0-1, blocks.0-27 and last.
+  • The author never states which weight family this targets. Silent
+    degradation, never a hard error, between Raw and Turbo. The two share
+    one architecture - ai-toolkit's config is commented "The reference
+    'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)"
+    - so every LoRA tensor shape matches on both and ComfyUI/diffuser
+
+───────────────────────────────────────────────────────────────────────────────
+  PROS
+───────────────────────────────────────────────────────────────────────────────
+  + Widely adopted — 32,058 downloads. [source: CivitAI stats]
+  + Positively received — 1,413 thumbs-up, 2 down. [source: CivitAI stats]
+  + Compact at 218 MB — cheap to stack with other LoRAs. [source: file
+    size]
+
+───────────────────────────────────────────────────────────────────────────────
+  CONS & CAVEATS
+───────────────────────────────────────────────────────────────────────────────
+  - Author labels this experimental — results vary shot to shot. [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
+───────────────────────────────────────────────────────────────────────────────
+  Silent degradation, never a hard error, between Raw and Turbo. The two
+  share one architecture - ai-toolkit's config is commented "The reference
+  'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)" - so
+  every LoRA tensor shape matches on both and ComfyUI/diffusers will load an
+  adapter against either without complaint. The lvladikov 4-step LoRA card
+  states this for the reverse direction: "Every layer it targets also exists
+  in Krea 2 Raw, so it will load there without complaint - but that is a side
+  effect of the shared architecture, not a supported mode", and reports
+  mixed, subject-dependent results with malformed subjects (duplicated heads,
+  fused limbs) at low step counts, needing 14+ steps with Raw's CFG on before
+  output is coherent. The intended direction (train on Raw, run on Turbo) is
+  officially blessed. The more common practical failure is a settings
+  mismatch rather than a weight mismatch: running a Turbo-targeted LoRA at
+  Raw's 52 steps + CFG 3.5, or leaving CFG enabled on Turbo, degrades output
+  with no warning. Loading a Krea 2 LoRA against a different model family
+  entirely (Flux, Qwen-Image, Z-Image) is also not an error - ComfyUI logs
+  "lora key not loaded: {key}" per key and generates as if no LoRA were
+  applied.
+
+───────────────────────────────────────────────────────────────────────────────
+  AUTHOR'S DESCRIPTION
+───────────────────────────────────────────────────────────────────────────────
+  Another amateur photo LoRA aimed at making images feel more realistic and
+  closer to everyday smartphone or social media photos. Expect sharper
+  details, brighter colors, and a more casual, natural look overall.
+
+  Krea V4:
+  Higher quality dataset, more steps, better captions.
+
+  Krea V3:
+  Similar to V2 but the captions used a tag-style instead of full sentences.
+
+  Krea V2:
+  More steps, more images, completely new captions
+
+  Krea V1:
+  Initial release trained with basic captions.
+
+  Z-Image Exp 1:
+  Experimental version trained with chinese captions.
+  Only 2000 steps and a small dataset but produces pretty good images.
+  Trigger:
+  这张照片是用iPhone 13拍摄的。
+
+  Z-Image V2:
+  More than doubled the steps compared to V1
+  Stronger effect
+  Can be used with higher strength
+
+  Z-Image V1:
+  A new version for Z-Image trained on a small dataset.
+
+  V4 Beta 2:
+  Trained on more images.
+
+  V4 Beta 1:
+  I have started to create a completely new dataset with only high quality
+  images. For Beta 1 I used 111 images, though I plan to significantly
+  increase it for Beta 2. All images only have the caption "43stet1c", but it
+  isn't needed to mention it in your prompt.
+  This combination seems to produce the best and most consistent results so
+  far.
+  Increase the strength if the effect isn't strong enough.
+
+  V3:
+  Same dataset as V2 but completely recaptioned with Claude 3.5 Sonnet and
+  upscaled to 1024 with Gigapixel.
+
+  V2:
+  Trained on 124 images with no captions at all. Let me know if this works
+  better than V1. Results can be inconsistent, use HiRes fix and Beta
+  scheduler.
+
+  V1:
+  This was trained on a dataset of 61 "aesthetic" images captioned with
+  Claude 3.5 Sonnet.
+
+───────────────────────────────────────────────────────────────────────────────
+  VERSION NOTES
+───────────────────────────────────────────────────────────────────────────────
+  Higher quality dataset, more steps, better captions.
+
+───────────────────────────────────────────────────────────────────────────────
+  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, 1024x1536)
+  02_image.jpeg  (image, 1088x1472)
+  03_image.jpeg  (image, 1024x1536)
+  04_image.jpeg  (image, 1088x1472)
+  05_image.jpeg  (image, 1024x1536)
+  06_image.jpeg  (image, 1024x1536)
+
+───────────────────────────────────────────────────────────────────────────────
+  HOW TO USE (COMFYUI)
+───────────────────────────────────────────────────────────────────────────────
+  1. Update ComfyUI to at least v0.26.0 (released 2026-06-23), which added
+  local Krea2 support. For reference-image / style-reference / edit LoRAs you
+  need at least v0.29.0 (2026-07-29), which added regular and timestep-zero
+  reference images for Krea 2. ModelMergeKrea2 arrived in v0.27.0
+  (2026-06-30).
+
+  2. Put the base stack in place (files from huggingface.co/Comfy-Org/Krea-2,
+  which is not gated):
+     - ComfyUI/models/diffusion_models/krea2_turbo_fp8_scaled.safetensors
+  (documented recommended default; the repo also holds krea2_turbo_bf16,
+  _int8_convrot, _mxfp8, _nvfp4 and krea2_raw_bf16 / _fp8_scaled /
+  _int8_convrot)
+     - ComfyUI/models/text_encoders/qwen3vl_4b_fp8_scaled.safetensors (bf16
+  also available)
+     - ComfyUI/models/vae/qwen_image_vae.safetensors
+
+  3. Put the LoRA .safetensors in ComfyUI/models/loras/ - flat, no per-model
+  subfolder required.
+
+  4. Build or load the graph. The official image_krea2_turbo_t2i.json
+  template is: UNETLoader (krea2_turbo_fp8_scaled.safetensors) ->
+  LoraLoaderModelOnly -> KSampler; CLIPLoader
+  (qwen3vl_4b_fp8_scaled.safetensors, type "krea2") -> text encode ->
+  KSampler positive, with ConditioningZeroOut into negative; VAELoader
+  (qwen_image_vae.safetensors) -> VAEDecode -> SaveImage; EmptyLatentImage
+  1024x1024 driven by ResolutionSelector. An int8 sibling template
+  (image_krea2_turbo_t2i_int8.json) also ships.
+
+  5. Load the LoRA with LoraLoaderModelOnly, not LoraLoader - both official
+  templates do. Krea 2 LoRAs patch the DiT only; note that the DiT's own
+  text_fusion / txtfusion blocks are a normal LoRA target (the official style
+  LoRAs patch them), but the Qwen3-VL text encoder itself is not. Set
+  strength 1.0 for Krea's official style LoRAs (the shipped template uses
+  0.8) and add the LoRA's trigger word to the prompt.
+
+  6. Sampler settings: steps 8, cfg 1, sampler euler, scheduler simple,
+  denoise 1.0. Do not add a ModelSampling node for plain t2i - ComfyUI
+  already applies shift 1.15 from Krea2.sampling_settings.
+
+  7. For style reference, use the dedicated template: UNETLoader
+  krea2_turbo_int8_convrot.safetensors, LoraLoaderModelOnly
+  krea2_style_reference.safetensors at 1.0, ModelSamplingFlux (1.15, 0.5),
+  TextEncodeQwenImageEditPlus fed by LoadImage,
+  FluxKontextMultiReferenceLatentMethod set to index_timestep_zero, and
+  SamplerCustomAdvanced with KSamplerSelect euler + BasicScheduler simple/8 +
+  CFGGuider cfg 1.
+
+  8. For instruction editing (Krea 2 Identity Edit), stock nodes are not
+  enough: install the ComfyUI-Krea2Edit node pack, which ships its own
+  workflows, and run Turbo at 8-12 steps / cfg 1 (or Raw at 20 steps / cfg 3
+  for removals) with LoRA strength 1.0.
+
+  9. Low-VRAM alternative: install city96's ComfyUI-GGUF and use a community
+  GGUF build (realrebelai/KREA-2_GGUFs, quantized from Turbo;
+  molbal/krea2-gguf, quantized from Raw) with the Unet Loader (GGUF) node in
+  place of UNETLoader. There is no official GGUF release. LoRA loading is
+  unchanged.
+
+───────────────────────────────────────────────────────────────────────────────
+  Scraped by lorakit on 2026-09-05 00:12 UTC.
+  Stats and description are the author's; pros/cons are derived from the
+  evidence tagged beside each line.
+───────────────────────────────────────────────────────────────────────────────

aesthetic-amateur-photo/metadata.json

@@ -0,0 +1,131 @@
+{
+  "name": "Aesthetic Amateur Photo",
+  "slug": "aesthetic-amateur-photo",
+  "base_model": "Krea 2",
+  "media_type": "image",
+  "model_id": 689192,
+  "version_id": 3284331,
+  "version_name": "Krea V4",
+  "source": "civitai",
+  "url": "https://civitai.com/models/689192",
+  "repo_id": null,
+  "file_path": null,
+  "author": "powerpuff",
+  "category": {
+    "key": "photoreal",
+    "label": "Photorealism"
+  },
+  "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": [
+      "The most important rule is not about chaining but about which checkpoint you chain onto: the official README says in bold \"TRAIN on Raw and RUN on Turbo\", and \"We highly recommend using RAW for training LoRAs and applying them on Turbo for inference.\" LoRAs trained on Raw are stated to work well on Turbo.",
+      "Chain with LoraLoaderModelOnly nodes in series (model -> model). Krea 2 LoRAs are DiT-only, so there is no CLIP side to chain and the text-encoder branch stays untouched however many LoRAs you add.",
+      "Krea 2 LoRAs stack in practice and the community treats multi-LoRA as routine - style + character, style + slider, realism + character. Several CivitAI Krea 2 authors state their LoRAs stack cleanly; one writes \"I recently stacked over 20 of my own LoRAs on a single generation and the output still came out great.\"",
+      "Reduce per-LoRA strength when stacking. CivitAI Krea 2 authors converge on dropping from ~1.0 to roughly 0.6-0.9 once more than one adapter is loaded. The CivitAI 12GB training guide also notes rank affects composability: \"rank 8 produced a lighter style influence that was easier to stack with other LoRAs, rank 16 provided a useful balance between concept strength and composability, and rank 32 was better suited to a precise object or concept intended to take priority over other stacked LoRAs.\"",
+      "Accelerator LoRAs are position-sensitive in intent, not in node order: a Raw-to-Turbo delta adapter belongs on the Raw checkpoint, a sub-8-step LoRA belongs on Turbo (or on Raw already carrying a Raw-to-Turbo adapter). Chaining Raw -> Raw-to-Turbo adapter -> 4-step LoRA is explicitly recommended by the 4-step LoRA's author.",
+      "Whatever you stack, the schedule must stay Turbo's: cfg 1.0 in ComfyUI (guidance 0.0 in diffusers) and mu/shift 1.15. The 4-step LoRA's card says \"Keep mu = 1.15. The training targets are anchored to that grid; a different shift moves the [sampler off them]\"; the same reasoning applies to any Turbo-trained adapter in the chain.",
+      "Reference/edit LoRAs are designed to be stacked under ordinary subject and style LoRAs - the Identity Edit card states \"Composes with your LoRAs: character/body/style LoRAs stack on top and steer the prior\" and recommends a stacked subject LoRA when unusual subjects drift toward the base prior.",
+      "Krea 2 LoRAs cannot be chained with LoRAs for any other base model. Many CivitAI pages carry multi-model titles like \"(Krea-2 + ZIT)\" or \"[Flux | ZIB | Krea2]\" - those are separate files for separate architectures shipped under one model page, not a combined adapter.",
+      "To combine two full Krea 2 checkpoints rather than adapters, ComfyUI ships a dedicated ModelMergeKrea2 node (added in v0.27.0) with per-section weights for first., tmlp., txtmlp., tproj., txtfusion.projector, txtfusion.layerwise_blocks.0-1, txtfusion.refiner_blocks.0-1, blocks.0-27 and last.",
+      "The author never states which weight family this targets. Silent degradation, never a hard error, between Raw and Turbo. The two share one architecture - ai-toolkit's config is commented \"The reference 'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)\" - so every LoRA tensor shape matches on both and ComfyUI/diffuser"
+    ]
+  },
+  "pros": [
+    "Widely adopted — 32,058 downloads. [source: CivitAI stats]",
+    "Positively received — 1,413 thumbs-up, 2 down. [source: CivitAI stats]",
+    "Compact at 218 MB — cheap to stack with other LoRAs. [source: file size]"
+  ],
+  "cons": [
+    "Author labels this experimental — results vary shot to shot. [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": "Krea_Amateur_V4.safetensors",
+    "size_bytes": 228588912,
+    "size_mb": 218.0,
+    "sha256": "0ad3b5092c99b4a1696df4ab07407775309da6c21d36c9a506f6b5ff153552ec"
+  },
+  "stats": {
+    "downloads": 32058,
+    "thumbs_up": 1413,
+    "thumbs_down": 2
+  },
+  "popularity": {
+    "downloads": 32058,
+    "thumbs_up": 1413,
+    "thumbs_down": 2,
+    "comments": 0,
+    "approval": "100% positive (1413/1415)",
+    "age_days": 4.4,
+    "downloads_per_day": 326.41,
+    "rank_by_downloads": 16,
+    "rank_by_momentum": 5,
+    "of_total": 150,
+    "adoption_tier": "High",
+    "momentum": "Hot · fresh release"
+  },
+  "selection_score": 49014.0,
+  "selection_reasons": [
+    "32,058 downloads",
+    "1,413 thumbs-up",
+    "top-ranked for its category"
+  ],
+  "example_media": [
+    {
+      "file": "01_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1536",
+      "prompt": ""
+    },
+    {
+      "file": "02_image.jpeg",
+      "type": "image",
+      "resolution": "1088x1472",
+      "prompt": ""
+    },
+    {
+      "file": "03_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1536",
+      "prompt": ""
+    },
+    {
+      "file": "04_image.jpeg",
+      "type": "image",
+      "resolution": "1088x1472",
+      "prompt": ""
+    },
+    {
+      "file": "05_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1536",
+      "prompt": ""
+    },
+    {
+      "file": "06_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1536",
+      "prompt": ""
+    }
+  ],
+  "tags": [
+    "photorealistic",
+    "style"
+  ],
+  "scraped_at": "2026-09-05T00:12:07+00:00",
+  "path": "krea-2/aesthetic-amateur-photo"
+}

background-detail-enhancer/info.txt

@@ -0,0 +1,277 @@
+═══════════════════════════════════════════════════════════════════════════════
+  BACKGROUND DETAIL ENHANCER✨
+═══════════════════════════════════════════════════════════════════════════════
+
+───────────────────────────────────────────────────────────────────────────────
+  IDENTITY
+───────────────────────────────────────────────────────────────────────────────
+  Base model        Krea 2
+  Generation mode   not declared by the author
+  Checkpoint        not determined — verify before use
+  Category          Detail / anatomy fixer
+  Author            ZyloO
+  Source            https://civitai.com/models/633524
+  Provenance        CivitAI · model 633524 · version 3211112
+
+───────────────────────────────────────────────────────────────────────────────
+  WEIGHTS
+───────────────────────────────────────────────────────────────────────────────
+  File              zy_BackgroundDetail_K2.safetensors
+  Version           Krea 2
+  Size              218.0 MB
+  Format            SafeTensor / unknown precision
+  SHA256            190a8835e96bcae59853cf7025ee0dad3a5963f240b945cb26e03798d12ccbba
+  Published         2026-08-09
+
+───────────────────────────────────────────────────────────────────────────────
+  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
+───────────────────────────────────────────────────────────────────────────────
+  Taken verbatim from the author's description:
+
+  LoRA strength       0.1–0.75
+
+───────────────────────────────────────────────────────────────────────────────
+  ADOPTION & POPULARITY
+───────────────────────────────────────────────────────────────────────────────
+  Downloads               23,711
+  Rating                  100% positive (2830/2831)
+  Comments                12
+  Published               26 days ago
+  Download rate           70.2/day since release
+
+  Adoption rank           #24 of 150 in this collection  (High)
+  Momentum rank           #39 of 150 by download rate  (Steady · 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 Krea 2 ones, not anything specific to this LoRA.
+
+  General Krea 2 rules that apply here (see REFERENCE.txt):
+
+  • The most important rule is not about chaining but about which
+    checkpoint you chain onto: the official README says in bold "TRAIN on
+    Raw and RUN on Turbo", and "We highly recommend using RAW for training
+    LoRAs and applying them on Turbo for inference." LoRAs trained on Raw
+    are stated to work well on Turbo.
+  • Chain with LoraLoaderModelOnly nodes in series (model -> model). Krea 2
+    LoRAs are DiT-only, so there is no CLIP side to chain and the text-
+    encoder branch stays untouched however many LoRAs you add.
+  • Krea 2 LoRAs stack in practice and the community treats multi-LoRA as
+    routine - style + character, style + slider, realism + character.
+    Several CivitAI Krea 2 authors state their LoRAs stack cleanly; one
+    writes "I recently stacked over 20 of my own LoRAs on a single
+    generation and the output still came out great."
+  • Reduce per-LoRA strength when stacking. CivitAI Krea 2 authors converge
+    on dropping from ~1.0 to roughly 0.6-0.9 once more than one adapter is
+    loaded. The CivitAI 12GB training guide also notes rank affects
+    composability: "rank 8 produced a lighter style influence that was
+    easier to stack with other LoRAs, rank 16 provided a useful balance
+    between concept strength and composability, and rank 32 was better
+    suited to a precise object or concept intended to take priority over
+    other stacked LoRAs."
+  • Accelerator LoRAs are position-sensitive in intent, not in node order:
+    a Raw-to-Turbo delta adapter belongs on the Raw checkpoint, a
+    sub-8-step LoRA belongs on Turbo (or on Raw already carrying a Raw-to-
+    Turbo adapter). Chaining Raw -> Raw-to-Turbo adapter -> 4-step LoRA is
+    explicitly recommended by the 4-step LoRA's author.
+  • Whatever you stack, the schedule must stay Turbo's: cfg 1.0 in ComfyUI
+    (guidance 0.0 in diffusers) and mu/shift 1.15. The 4-step LoRA's card
+    says "Keep mu = 1.15. The training targets are anchored to that grid; a
+    different shift moves the [sampler off them]"; the same reasoning
+    applies to any Turbo-trained adapter in the chain.
+  • Reference/edit LoRAs are designed to be stacked under ordinary subject
+    and style LoRAs - the Identity Edit card states "Composes with your
+    LoRAs: character/body/style LoRAs stack on top and steer the prior" and
+    recommends a stacked subject LoRA when unusual subjects drift toward
+    the base prior.
+  • Krea 2 LoRAs cannot be chained with LoRAs for any other base model.
+    Many CivitAI pages carry multi-model titles like "(Krea-2 + ZIT)" or
+    "[Flux | ZIB | Krea2]" - those are separate files for separate
+    architectures shipped under one model page, not a combined adapter.
+  • To combine two full Krea 2 checkpoints rather than adapters, ComfyUI
+    ships a dedicated ModelMergeKrea2 node (added in v0.27.0) with per-
+    section weights for first., tmlp., txtmlp., tproj.,
+    txtfusion.projector, txtfusion.layerwise_blocks.0-1,
+    txtfusion.refiner_blocks.0-1, blocks.0-27 and last.
+  • The author never states which weight family this targets. Silent
+    degradation, never a hard error, between Raw and Turbo. The two share
+    one architecture - ai-toolkit's config is commented "The reference
+    'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)"
+    - so every LoRA tensor shape matches on both and ComfyUI/diffuser
+
+───────────────────────────────────────────────────────────────────────────────
+  PROS
+───────────────────────────────────────────────────────────────────────────────
+  + Widely adopted — 23,711 downloads. [source: CivitAI stats]
+  + Positively received — 2,830 thumbs-up, 1 down. [source: CivitAI stats]
+  + Compact at 218 MB — cheap to stack with other LoRAs. [source: file
+    size]
+  + Author published concrete recommended settings. [source: description]
+
+───────────────────────────────────────────────────────────────────────────────
+  CONS & CAVEATS
+───────────────────────────────────────────────────────────────────────────────
+  - 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
+───────────────────────────────────────────────────────────────────────────────
+  Silent degradation, never a hard error, between Raw and Turbo. The two
+  share one architecture - ai-toolkit's config is commented "The reference
+  'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)" - so
+  every LoRA tensor shape matches on both and ComfyUI/diffusers will load an
+  adapter against either without complaint. The lvladikov 4-step LoRA card
+  states this for the reverse direction: "Every layer it targets also exists
+  in Krea 2 Raw, so it will load there without complaint - but that is a side
+  effect of the shared architecture, not a supported mode", and reports
+  mixed, subject-dependent results with malformed subjects (duplicated heads,
+  fused limbs) at low step counts, needing 14+ steps with Raw's CFG on before
+  output is coherent. The intended direction (train on Raw, run on Turbo) is
+  officially blessed. The more common practical failure is a settings
+  mismatch rather than a weight mismatch: running a Turbo-targeted LoRA at
+  Raw's 52 steps + CFG 3.5, or leaving CFG enabled on Turbo, degrades output
+  with no warning. Loading a Krea 2 LoRA against a different model family
+  entirely (Flux, Qwen-Image, Z-Image) is also not an error - ComfyUI logs
+  "lora key not loaded: {key}" per key and generates as if no LoRA were
+  applied.
+
+───────────────────────────────────────────────────────────────────────────────
+  AUTHOR'S DESCRIPTION
+───────────────────────────────────────────────────────────────────────────────
+  ✨ REFERRAL CODE : DMD9XGMN ℹ️ Model Overview
+  - This LoRA enhances the background of the image, adding more detail and
+  clutter to it.
+
+  - Image Comparison
+
+  🏠 Recommended Parameters Pony Settings: - Strength : 0.1-0.75 .
+
+  - Clip Skip : 2
+
+  Additional Information: - Useful Prompt Words:
+  - Clutter
+
+  - Background
+
+  - Detailed background
+
+  - Elaborate
+
+  - Rich
+
+  - Dynamic
+
+  - Immersive
+
+  - Intricate
+
+  🏋️‍♀️ Training LoRas - Read the following article for tips and my training
+  preset
+
+  🌞 Requests - Do you want a specific LoRA? You can request a commission on
+  my Ko-Fi
+
+  ⚡️ Buzz for the Best Images ⚡️
+
+───────────────────────────────────────────────────────────────────────────────
+  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, 768x1024)
+  02_image.jpeg  (image, 768x1024)
+  03_image.jpeg  (image, 768x1376)
+  04_image.jpeg  (image, 768x1024)
+  05_image.jpeg  (image, 768x1024)
+  06_image.jpeg  (image, 768x1024)
+
+───────────────────────────────────────────────────────────────────────────────
+  HOW TO USE (COMFYUI)
+───────────────────────────────────────────────────────────────────────────────
+  1. Update ComfyUI to at least v0.26.0 (released 2026-06-23), which added
+  local Krea2 support. For reference-image / style-reference / edit LoRAs you
+  need at least v0.29.0 (2026-07-29), which added regular and timestep-zero
+  reference images for Krea 2. ModelMergeKrea2 arrived in v0.27.0
+  (2026-06-30).
+
+  2. Put the base stack in place (files from huggingface.co/Comfy-Org/Krea-2,
+  which is not gated):
+     - ComfyUI/models/diffusion_models/krea2_turbo_fp8_scaled.safetensors
+  (documented recommended default; the repo also holds krea2_turbo_bf16,
+  _int8_convrot, _mxfp8, _nvfp4 and krea2_raw_bf16 / _fp8_scaled /
+  _int8_convrot)
+     - ComfyUI/models/text_encoders/qwen3vl_4b_fp8_scaled.safetensors (bf16
+  also available)
+     - ComfyUI/models/vae/qwen_image_vae.safetensors
+
+  3. Put the LoRA .safetensors in ComfyUI/models/loras/ - flat, no per-model
+  subfolder required.
+
+  4. Build or load the graph. The official image_krea2_turbo_t2i.json
+  template is: UNETLoader (krea2_turbo_fp8_scaled.safetensors) ->
+  LoraLoaderModelOnly -> KSampler; CLIPLoader
+  (qwen3vl_4b_fp8_scaled.safetensors, type "krea2") -> text encode ->
+  KSampler positive, with ConditioningZeroOut into negative; VAELoader
+  (qwen_image_vae.safetensors) -> VAEDecode -> SaveImage; EmptyLatentImage
+  1024x1024 driven by ResolutionSelector. An int8 sibling template
+  (image_krea2_turbo_t2i_int8.json) also ships.
+
+  5. Load the LoRA with LoraLoaderModelOnly, not LoraLoader - both official
+  templates do. Krea 2 LoRAs patch the DiT only; note that the DiT's own
+  text_fusion / txtfusion blocks are a normal LoRA target (the official style
+  LoRAs patch them), but the Qwen3-VL text encoder itself is not. Set
+  strength 1.0 for Krea's official style LoRAs (the shipped template uses
+  0.8) and add the LoRA's trigger word to the prompt.
+
+  6. Sampler settings: steps 8, cfg 1, sampler euler, scheduler simple,
+  denoise 1.0. Do not add a ModelSampling node for plain t2i - ComfyUI
+  already applies shift 1.15 from Krea2.sampling_settings.
+
+  7. For style reference, use the dedicated template: UNETLoader
+  krea2_turbo_int8_convrot.safetensors, LoraLoaderModelOnly
+  krea2_style_reference.safetensors at 1.0, ModelSamplingFlux (1.15, 0.5),
+  TextEncodeQwenImageEditPlus fed by LoadImage,
+  FluxKontextMultiReferenceLatentMethod set to index_timestep_zero, and
+  SamplerCustomAdvanced with KSamplerSelect euler + BasicScheduler simple/8 +
+  CFGGuider cfg 1.
+
+  8. For instruction editing (Krea 2 Identity Edit), stock nodes are not
+  enough: install the ComfyUI-Krea2Edit node pack, which ships its own
+  workflows, and run Turbo at 8-12 steps / cfg 1 (or Raw at 20 steps / cfg 3
+  for removals) with LoRA strength 1.0.
+
+  9. Low-VRAM alternative: install city96's ComfyUI-GGUF and use a community
+  GGUF build (realrebelai/KREA-2_GGUFs, quantized from Turbo;
+  molbal/krea2-gguf, quantized from Raw) with the Unet Loader (GGUF) node in
+  place of UNETLoader. There is no official GGUF release. LoRA loading is
+  unchanged.
+
+───────────────────────────────────────────────────────────────────────────────
+  Scraped by lorakit on 2026-09-05 00:12 UTC.
+  Stats and description are the author's; pros/cons are derived from the
+  evidence tagged beside each line.
+───────────────────────────────────────────────────────────────────────────────

background-detail-enhancer/metadata.json

@@ -0,0 +1,139 @@
+{
+  "name": "Background Detail Enhancer✨",
+  "slug": "background-detail-enhancer",
+  "base_model": "Krea 2",
+  "media_type": "image",
+  "model_id": 633524,
+  "version_id": 3211112,
+  "version_name": "Krea 2",
+  "source": "civitai",
+  "url": "https://civitai.com/models/633524",
+  "repo_id": null,
+  "file_path": null,
+  "author": "ZyloO",
+  "category": {
+    "key": "detail",
+    "label": "Detail / anatomy fixer"
+  },
+  "modalities": [],
+  "primary_modality": "unspecified",
+  "checkpoint_family": "unknown",
+  "trigger_words": [],
+  "recommended_settings": {
+    "LoRA strength": "0.1–0.75"
+  },
+  "chaining": {
+    "author": {
+      "stacks": "not stated",
+      "notes": "",
+      "avoid": [],
+      "requires": [],
+      "order": "",
+      "strength": "",
+      "stated": false
+    },
+    "general_rules": [
+      "The most important rule is not about chaining but about which checkpoint you chain onto: the official README says in bold \"TRAIN on Raw and RUN on Turbo\", and \"We highly recommend using RAW for training LoRAs and applying them on Turbo for inference.\" LoRAs trained on Raw are stated to work well on Turbo.",
+      "Chain with LoraLoaderModelOnly nodes in series (model -> model). Krea 2 LoRAs are DiT-only, so there is no CLIP side to chain and the text-encoder branch stays untouched however many LoRAs you add.",
+      "Krea 2 LoRAs stack in practice and the community treats multi-LoRA as routine - style + character, style + slider, realism + character. Several CivitAI Krea 2 authors state their LoRAs stack cleanly; one writes \"I recently stacked over 20 of my own LoRAs on a single generation and the output still came out great.\"",
+      "Reduce per-LoRA strength when stacking. CivitAI Krea 2 authors converge on dropping from ~1.0 to roughly 0.6-0.9 once more than one adapter is loaded. The CivitAI 12GB training guide also notes rank affects composability: \"rank 8 produced a lighter style influence that was easier to stack with other LoRAs, rank 16 provided a useful balance between concept strength and composability, and rank 32 was better suited to a precise object or concept intended to take priority over other stacked LoRAs.\"",
+      "Accelerator LoRAs are position-sensitive in intent, not in node order: a Raw-to-Turbo delta adapter belongs on the Raw checkpoint, a sub-8-step LoRA belongs on Turbo (or on Raw already carrying a Raw-to-Turbo adapter). Chaining Raw -> Raw-to-Turbo adapter -> 4-step LoRA is explicitly recommended by the 4-step LoRA's author.",
+      "Whatever you stack, the schedule must stay Turbo's: cfg 1.0 in ComfyUI (guidance 0.0 in diffusers) and mu/shift 1.15. The 4-step LoRA's card says \"Keep mu = 1.15. The training targets are anchored to that grid; a different shift moves the [sampler off them]\"; the same reasoning applies to any Turbo-trained adapter in the chain.",
+      "Reference/edit LoRAs are designed to be stacked under ordinary subject and style LoRAs - the Identity Edit card states \"Composes with your LoRAs: character/body/style LoRAs stack on top and steer the prior\" and recommends a stacked subject LoRA when unusual subjects drift toward the base prior.",
+      "Krea 2 LoRAs cannot be chained with LoRAs for any other base model. Many CivitAI pages carry multi-model titles like \"(Krea-2 + ZIT)\" or \"[Flux | ZIB | Krea2]\" - those are separate files for separate architectures shipped under one model page, not a combined adapter.",
+      "To combine two full Krea 2 checkpoints rather than adapters, ComfyUI ships a dedicated ModelMergeKrea2 node (added in v0.27.0) with per-section weights for first., tmlp., txtmlp., tproj., txtfusion.projector, txtfusion.layerwise_blocks.0-1, txtfusion.refiner_blocks.0-1, blocks.0-27 and last.",
+      "The author never states which weight family this targets. Silent degradation, never a hard error, between Raw and Turbo. The two share one architecture - ai-toolkit's config is commented \"The reference 'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)\" - so every LoRA tensor shape matches on both and ComfyUI/diffuser"
+    ]
+  },
+  "pros": [
+    "Widely adopted — 23,711 downloads. [source: CivitAI stats]",
+    "Positively received — 2,830 thumbs-up, 1 down. [source: CivitAI stats]",
+    "Compact at 218 MB — cheap to stack with other LoRAs. [source: file size]",
+    "Author published concrete recommended settings. [source: description]"
+  ],
+  "cons": [
+    "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": "zy_BackgroundDetail_K2.safetensors",
+    "size_bytes": 228588904,
+    "size_mb": 218.0,
+    "sha256": "190a8835e96bcae59853cf7025ee0dad3a5963f240b945cb26e03798d12ccbba"
+  },
+  "stats": {
+    "downloads": 23711,
+    "thumbs_up": 2830,
+    "thumbs_down": 1
+  },
+  "popularity": {
+    "downloads": 23711,
+    "thumbs_up": 2830,
+    "thumbs_down": 1,
+    "comments": 12,
+    "approval": "100% positive (2830/2831)",
+    "age_days": 26.5,
+    "downloads_per_day": 70.23,
+    "rank_by_downloads": 24,
+    "rank_by_momentum": 39,
+    "of_total": 150,
+    "adoption_tier": "High",
+    "momentum": "Steady · fresh release"
+  },
+  "selection_score": 57671.0,
+  "selection_reasons": [
+    "23,711 downloads",
+    "2,830 thumbs-up",
+    "top-ranked for its category"
+  ],
+  "example_media": [
+    {
+      "file": "01_image.jpeg",
+      "type": "image",
+      "resolution": "768x1024",
+      "prompt": ""
+    },
+    {
+      "file": "02_image.jpeg",
+      "type": "image",
+      "resolution": "768x1024",
+      "prompt": ""
+    },
+    {
+      "file": "03_image.jpeg",
+      "type": "image",
+      "resolution": "768x1376",
+      "prompt": ""
+    },
+    {
+      "file": "04_image.jpeg",
+      "type": "image",
+      "resolution": "768x1024",
+      "prompt": ""
+    },
+    {
+      "file": "05_image.jpeg",
+      "type": "image",
+      "resolution": "768x1024",
+      "prompt": ""
+    },
+    {
+      "file": "06_image.jpeg",
+      "type": "image",
+      "resolution": "768x1024",
+      "prompt": ""
+    }
+  ],
+  "tags": [
+    "scenery",
+    "backgrounds",
+    "enhancer",
+    "detailer",
+    "detail",
+    "concept",
+    "clutter",
+    "environment"
+  ],
+  "scraped_at": "2026-09-05T00:12:04+00:00",
+  "path": "krea-2/background-detail-enhancer"
+}

background-detail-enhancer/zy_BackgroundDetail_K2.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:190a8835e96bcae59853cf7025ee0dad3a5963f240b945cb26e03798d12ccbba
+size 228588904

bartstyle-blue-archive-art-style-lora-anime-flat-color/BArtstyle-KREA2-V2-1024.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c19dc382d92002bfc53ece7df5556acc288579af9f0719752059dfae7fe9f70
+size 938539208

bartstyle-blue-archive-art-style-lora-anime-flat-color/info.txt

@@ -0,0 +1,493 @@
+═══════════════════════════════════════════════════════════════════════════════
+  BARTSTYLE | BLUE ARCHIVE ART STYLE LORA | ANIME FLAT COLOR | 蔚蓝档案画风模型 |
+  ブルーアーカイブ画風モデル
+═══════════════════════════════════════════════════════════════════════════════
+
+───────────────────────────────────────────────────────────────────────────────
+  IDENTITY
+───────────────────────────────────────────────────────────────────────────────
+  Base model        Krea 2
+  Generation mode   not declared by the author
+  Checkpoint        not determined — verify before use
+  Category          Visual style
+  Author            4BEraser
+  Source            https://civitai.com/models/29215
+  Provenance        CivitAI · model 29215 · version 3220090
+
+───────────────────────────────────────────────────────────────────────────────
+  WEIGHTS
+───────────────────────────────────────────────────────────────────────────────
+  File              BArtstyle-KREA2-V2-1024.safetensors
+  Version           Krea2-V2
+  Size              895.1 MB
+  Format            SafeTensor / unknown precision
+  SHA256            9c19dc382d92002bfc53ece7df5556acc288579af9f0719752059dfae7fe9f70
+  Published         2026-08-11
+
+───────────────────────────────────────────────────────────────────────────────
+  TRIGGER WORDS
+───────────────────────────────────────────────────────────────────────────────
+  › "azure youth anime art style"
+
+───────────────────────────────────────────────────────────────────────────────
+  RECOMMENDED SETTINGS
+───────────────────────────────────────────────────────────────────────────────
+  Taken verbatim from the author's description:
+
+  LoRA strength       0.6–0.7
+  Resolution          1024x768
+  Sampler             DPM++ 2M
+
+───────────────────────────────────────────────────────────────────────────────
+  ADOPTION & POPULARITY
+───────────────────────────────────────────────────────────────────────────────
+  Downloads               28,534
+  Rating                  100% positive (2683/2684)
+  Comments                0
+  Published               24 days ago
+  Download rate           82.3/day since release
+
+  Adoption rank           #21 of 150 in this collection  (High)
+  Momentum rank           #34 of 150 by download rate  (Trending · 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 Krea 2 ones, not anything specific to this LoRA.
+
+  General Krea 2 rules that apply here (see REFERENCE.txt):
+
+  • The most important rule is not about chaining but about which
+    checkpoint you chain onto: the official README says in bold "TRAIN on
+    Raw and RUN on Turbo", and "We highly recommend using RAW for training
+    LoRAs and applying them on Turbo for inference." LoRAs trained on Raw
+    are stated to work well on Turbo.
+  • Chain with LoraLoaderModelOnly nodes in series (model -> model). Krea 2
+    LoRAs are DiT-only, so there is no CLIP side to chain and the text-
+    encoder branch stays untouched however many LoRAs you add.
+  • Krea 2 LoRAs stack in practice and the community treats multi-LoRA as
+    routine - style + character, style + slider, realism + character.
+    Several CivitAI Krea 2 authors state their LoRAs stack cleanly; one
+    writes "I recently stacked over 20 of my own LoRAs on a single
+    generation and the output still came out great."
+  • Reduce per-LoRA strength when stacking. CivitAI Krea 2 authors converge
+    on dropping from ~1.0 to roughly 0.6-0.9 once more than one adapter is
+    loaded. The CivitAI 12GB training guide also notes rank affects
+    composability: "rank 8 produced a lighter style influence that was
+    easier to stack with other LoRAs, rank 16 provided a useful balance
+    between concept strength and composability, and rank 32 was better
+    suited to a precise object or concept intended to take priority over
+    other stacked LoRAs."
+  • Accelerator LoRAs are position-sensitive in intent, not in node order:
+    a Raw-to-Turbo delta adapter belongs on the Raw checkpoint, a
+    sub-8-step LoRA belongs on Turbo (or on Raw already carrying a Raw-to-
+    Turbo adapter). Chaining Raw -> Raw-to-Turbo adapter -> 4-step LoRA is
+    explicitly recommended by the 4-step LoRA's author.
+  • Whatever you stack, the schedule must stay Turbo's: cfg 1.0 in ComfyUI
+    (guidance 0.0 in diffusers) and mu/shift 1.15. The 4-step LoRA's card
+    says "Keep mu = 1.15. The training targets are anchored to that grid; a
+    different shift moves the [sampler off them]"; the same reasoning
+    applies to any Turbo-trained adapter in the chain.
+  • Reference/edit LoRAs are designed to be stacked under ordinary subject
+    and style LoRAs - the Identity Edit card states "Composes with your
+    LoRAs: character/body/style LoRAs stack on top and steer the prior" and
+    recommends a stacked subject LoRA when unusual subjects drift toward
+    the base prior.
+  • Krea 2 LoRAs cannot be chained with LoRAs for any other base model.
+    Many CivitAI pages carry multi-model titles like "(Krea-2 + ZIT)" or
+    "[Flux | ZIB | Krea2]" - those are separate files for separate
+    architectures shipped under one model page, not a combined adapter.
+  • To combine two full Krea 2 checkpoints rather than adapters, ComfyUI
+    ships a dedicated ModelMergeKrea2 node (added in v0.27.0) with per-
+    section weights for first., tmlp., txtmlp., tproj.,
+    txtfusion.projector, txtfusion.layerwise_blocks.0-1,
+    txtfusion.refiner_blocks.0-1, blocks.0-27 and last.
+  • The author never states which weight family this targets. Silent
+    degradation, never a hard error, between Raw and Turbo. The two share
+    one architecture - ai-toolkit's config is commented "The reference
+    'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)"
+    - so every LoRA tensor shape matches on both and ComfyUI/diffuser
+
+───────────────────────────────────────────────────────────────────────────────
+  PROS
+───────────────────────────────────────────────────────────────────────────────
+  + Widely adopted — 28,534 downloads. [source: CivitAI stats]
+  + Positively received — 2,683 thumbs-up, 1 down. [source: CivitAI stats]
+  + Explicit trigger word(s) — azure youth anime art style — so the effect
+    can be turned on and off from the prompt. [source: model metadata]
+  + Author published concrete recommended settings. [source: description]
+
+───────────────────────────────────────────────────────────────────────────────
+  CONS & CAVEATS
+───────────────────────────────────────────────────────────────────────────────
+  - Author says the LoRA may be overtrained — start at low strength.
+    [source: description]
+  - Needs a reduced strength; the default 1.0 is likely too strong.
+    [source: description]
+  - 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]
+  - Large at 895 MB — slower to load and heavier to stack. [source: file
+    size]
+
+───────────────────────────────────────────────────────────────────────────────
+  COMPATIBILITY WARNING
+───────────────────────────────────────────────────────────────────────────────
+  Silent degradation, never a hard error, between Raw and Turbo. The two
+  share one architecture - ai-toolkit's config is commented "The reference
+  'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)" - so
+  every LoRA tensor shape matches on both and ComfyUI/diffusers will load an
+  adapter against either without complaint. The lvladikov 4-step LoRA card
+  states this for the reverse direction: "Every layer it targets also exists
+  in Krea 2 Raw, so it will load there without complaint - but that is a side
+  effect of the shared architecture, not a supported mode", and reports
+  mixed, subject-dependent results with malformed subjects (duplicated heads,
+  fused limbs) at low step counts, needing 14+ steps with Raw's CFG on before
+  output is coherent. The intended direction (train on Raw, run on Turbo) is
+  officially blessed. The more common practical failure is a settings
+  mismatch rather than a weight mismatch: running a Turbo-targeted LoRA at
+  Raw's 52 steps + CFG 3.5, or leaving CFG enabled on Turbo, degrades output
+  with no warning. Loading a Krea 2 LoRA against a different model family
+  entirely (Flux, Qwen-Image, Z-Image) is also not an error - ComfyUI logs
+  "lora key not loaded: {key}" per key and generates as if no LoRA were
+  applied.
+
+───────────────────────────────────────────────────────────────────────────────
+  AUTHOR'S DESCRIPTION
+───────────────────────────────────────────────────────────────────────────────
+  KREA 2 V2 Update: Dataset overhaul, trained 10 epoch each on 512/768/1024.
+  Dim set to 64 for test.
+  The halo should be easy to remove by reduce LoRA weight. The best balance
+  is 0.8 in my test.
+  Trigger word has changed like official Krea 2 loras:
+  azure youth anime art style But you should be able to use it without
+  trigger word, cuz it's pretty weak.
+  It can be used as a high-contrast or flat color style module.
+  Use proper prompt like Krea 2 prompting guide . Still, the model do not
+  contain any character information. You still need character lora if you
+  want a typical character.
+  ====================
+  KREA 2 V1 Update: I've back for a little test, trained on my BAXL dataset
+  with NL captions. Trained on my own 4090.
+  Weight 1 is fine, Trigger word is
+  BArtstyle style USE WITH NATURAL LANGUAGE. DANBOORU TAG IS NOT WORKING.
+  If it doesn't work well, try remove words like "anime, cartoon" in the
+  prompt. It may slightly affect the style.
+  Halo may undesirably come out in certain prompts. If happens, try a
+  different prompt or reduce lora weight, or just try another seed will help.
+  This model do not contain any character information. You still need
+  character lora to make it work properly, this is limited by Krea 2's base
+  model.
+  I will test it after I train my first character lora for Krea2 and update
+  info here.
+
+  Info below has expired. Time flies huh? It has been 2 years since I train
+  model.
+  ===============================================
+
+  I've wasted too much time on this model... Probably this will be the final
+  update, I may focus on SDXL when 4090 be affordable to me, or just fixing
+  other LoRA I made if 4090 is still cost way too much in next half year.
+  The original decision was using dreambooth for update, but I'm using 3060
+  so the time effort is not good. So, still LoRA.
+  REv3 Change Log： - Base model now have 3 versions, which have different
+  feature:
+  - CF3 - Have a good balance between replica art style and user-friendly.
+  - You should use it on CounterFeitV3 (and those CF3's mixed model, e.g.
+  PrimeMix/MixPro)
+
+  - DO NOT USE IT WITH Anything SERIES. I tried and it looks awful.
+
+  - Silicon29 - Trained on Silicon29-Dark, it sacrificed a bit BA's clean
+  line art, but it has a more BA-ish vivid color+god ray than CF3 version.
+  - I strongly recommend this version (at least you should try it out!)
+
+  - It has a better variety with non-base model when still kept the base
+  model's characteristic.
+
+  - Pink theme is its weakness. If you want to generate the BA Finale's blue-
+  pink gradient scenery, use CF3/ANY version instead.
+
+  -
+
+  - ANY - Trained on ANYv5, harder to use than CF3 or Silicon29, but it seems
+  to be the most BA one. Only suggest when you using AnythingV5 or those
+  model based on it.
+
+  - Dataset rehauled
+  - Manually downscale + denoise before send it to bucket to avoid aliasing.
+
+  - Trained resolution is 1024x768 . So it's better generate horizonal image
+  than vertical one. Suggest generate resolution :
+  - 432x768 (Vertical, x2.5 for 1080P)
+
+  - 960x544 (Horizonal, x2 for 1080P)
+
+  - Concept change
+  - now the typical BA style (blue sky+ modern city / ruined classroom +
+  water) has its own concept.
+  - trigger word: blue sky
+
+  - blue theme now can generate nighttime.
+
+  - Preivew (test on Silicon29 ver.)
+  -
+
+  - Block Weight Training
+  - Use a more conservative set. So these LoRAs will affect more anatomy but
+  not sacrificed too much effect.
+
+  - While using REv3, your LoRA weight can up to 1, s ince the anatomy is
+  basically hopeless XD
+  - joke aside, v9b has sacrificed too much color for meh anatomy while STILL
+  CAN NOT use at a higher weight.
+
+  - REv3's anatomy is better than v7 I guess.
+
+  - I've tested some sampler.
+  - For conclusion: Restart ≈ DPM++ 2M Karras > DPM++ 2S a Karras > any
+  sampler with SDE
+  - UniPC is basically a downgrade version of Restart.
+
+  - DPM++ 2S a Karras is oversaturated on typical model and weight. not bad
+  but just not BA style, you should use a lower weight on this sampler.
+
+  - Restart & DPM++ 2M Karras will cause blue or green spots or something
+  like jpeg artifact in typical situation. It can not be fixed cuz it's base
+  model's problem - Although CounterFeit V3 provide a relatively good art,
+  when still focus on prompts, the model itself has been overtrained (I
+  believe that's why it's very picky on VAEs.)
+
+  - ↑ that's why you should use CFv2.5's VAE while using this lora models.
+
+  ==================
+  V9b Change Log： - Base model changed to CounterFeit V3.
+
+  - Use Block Weight Training. Anatomy should be way better than V7.
+
+  - BAPV no longer used as trigger word. You can still use color scheme to
+  trigger typical color.
+
+  - For the best result, you should choose DPM++ 2S a Karras sampler & use
+  4x-ultrasharp as upscaler.
+
+  Known Issue： - Color shifting has been noticeably sacrificed for better
+  anatomy. Check the last preivew for more info. V9b will affect less on the
+  whole image's composition and charactor posing. I might release a version
+  without Block Weight Training, which has better colors, but you should know
+  how to properly use LoRA Block Weight.
+
+  - It tends to give a high-contrast two-tone shadow even on nighttime.
+  Consider combine with night concept if you need to generate dark scene.
+
+  - Bad Composition still occurs if you are using a relatively high weight
+  (over 0.7), consider use LoRA Block Weight (LBW) with OUTALL setting to
+  reduce the affect. If you use LBW, higher weight up to 1 is Okay.
+
+  V7: Since the training data is based on Blue Archive's PV & CG screenshot,
+  this LoRA do not very capable with NSFW content.
+  V9 might be better. But I haven't tried it out. Feel free to test it by
+  yourself!
+  Weight 0.6-0.7 is good enough. About 0.4-0.8 is working zone. Higher will
+  have a better color scheme but worse anatomy.
+  ==================
+  Support color scheme in grid: (V7 sample)
+
+  Still, it only works perfectly without any charactor LoRA. You can get
+  similar result with charactor LoRA, but you need to be patient. V9b has
+  been MUCH better, but still you need to roll dices.
+
+  The method below is only for older LoHA version. ===============
+  Compared to Bluemix (the 4g ckpt model), this LoHA is more capable with
+  character LoRA, it will tend to be keeping character's facial
+  characteristic and outfit, only change the line and coloring. (like a color
+  filter rather than totally different art style).
+  If you like the character be fully switch to BA art style, generating OC or
+  don't care about the LoRA character's outfit, I suggest you to use BlueMix
+  instead.
+
+  Suggestion:
+  - use nature language to generate base environment (e.g. A girl is standing
+  alongside the river. )
+
+  - weight 0.8-1, (higher value for better result but worse anatomy, vice
+  versa.)
+
+  - base resolution: w or h lower than 768x768, over will cause the image
+  broken more often.
+
+  - base model: suggest anyv3/v5, not recommend any oil painting/pastel model
+  like AOM series or counterfeit. It CAN use, but is only a color filter.
+
+  - environment prompts (still work!):
+  - Blue Archive-ish theme (Millennium?): tree, cloud, road, city, day, blue
+  sky, outdoors
+
+  - Japanese theme (Hyakkiyako?): cloud, day, outdoors, cherry blossoms,
+
+  - Nightsky: night, night sky, starry sky,
+
+  - You can try other tags generated by taggers of Blue Archive CGs.
+
+  ===
+  I'm learning how to make a LoHA/artstyle model. This is my first step
+  result, my PC isn't very good for testing so I release it for more
+  feedbacks from your side, so this is basically an Alpha test, the model do
+  work but not very user-friendly.
+  If you want to get a good result, you should know how to use lora block
+  weight.
+  If you don't know how to use it but still have interest on learning some
+  new tricks, then read the following guide. My goal is make a artstyle LoHA
+  that is capable to most of chara LoRAs.
+  So I trained it on Animefull-pruned, works perfectly on that one and
+  Anything V3, will work on AnythingV4.5 without character LoRA, but almost
+  won't work on Counterfeit or AOM3. It will more likely a color filter than
+  an art style on those models.
+  Maybe I should train another version based on AnythingV4.5. If you have any
+  suggestion about art style training, let me know. I will be very appreciate
+  of that.
+
+  It can work without tags, but sometimes will make weird color.
+  Here are some tag groups for more stable results:
+  - Blue Archive-ish theme (Millennium?): tree, cloud, road, city, day, blue
+  sky, outdoors
+
+  - Japanese theme (Hyakkiyako?): pink theme, cloud, day, outdoors, cherry
+  blossoms,
+
+  - Nightsky: blue theme, night, night sky, starry sky, cityscape,
+
+  - You can try other tags generated by taggers of Blue Archive CGs.
+
+  You 'd better use it with LoRA block weight if you want to input a
+  character lora. You can get it as webui plugin. Otherwise you should lower
+  this model's weight to below 0.6, which will work bit poorly but still
+  work, or it will make the image color very awful.
+  ↓If you get something like this, lower the artstyle model weight, or use
+  the method below.
+
+  When you use it with LoRA block weight, there are 2 methods:
+  - A: By default, you should remove your charactor lora's artstyle by this:
+  <YourCharaModelName:1:1,1,1,1,1,0,0,0,1,1,1,1,1,1,1,0,0>
+  don't forget to add :ALL to this model or it may won't be working.
+  e.g. <BArstylev1c:1:ALL>
+
+  - B: If you noticed that your character's appearance has been obviously
+  affected by this model, keep only the color of the artstyle model instead
+  like this:
+  <BArstylev1c:1:1,0,0,0,0,0,0,0,0,0,0,0.8,1,1,1,1,1>
+  In this situation, you can keep Charactor model in full.
+  e.g. <YourCharaModelName:1:ALL>
+
+  - Basically, method A will works more like the original artstyle than
+  method B.
+
+  - It seems that LoHA model have more block weights, I don't have much time
+  to test since my PC will easily crash on it, if you have some information
+  about it, please tell me. But so far using LoRA block weight is fine.
+
+  - ↓Still, if you get everything right but still got something like this,
+  try to use the second way instead.
+
+  Method A when the charactor is affected.
+
+  Method B
+
+───────────────────────────────────────────────────────────────────────────────
+  VERSION NOTES
+───────────────────────────────────────────────────────────────────────────────
+  Dataset overhaul
+
+───────────────────────────────────────────────────────────────────────────────
+  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, 1288x1936)
+  02_image.jpeg  (image, 1536x2728)
+  03_image.jpeg  (image, 1536x2728)
+  04_image.jpeg  (image, 1024x1024)
+  05_image.jpeg  (image, 1536x2728)
+  06_image.jpeg  (image, 1288x1936)
+
+───────────────────────────────────────────────────────────────────────────────
+  HOW TO USE (COMFYUI)
+───────────────────────────────────────────────────────────────���───────────────
+  1. Update ComfyUI to at least v0.26.0 (released 2026-06-23), which added
+  local Krea2 support. For reference-image / style-reference / edit LoRAs you
+  need at least v0.29.0 (2026-07-29), which added regular and timestep-zero
+  reference images for Krea 2. ModelMergeKrea2 arrived in v0.27.0
+  (2026-06-30).
+
+  2. Put the base stack in place (files from huggingface.co/Comfy-Org/Krea-2,
+  which is not gated):
+     - ComfyUI/models/diffusion_models/krea2_turbo_fp8_scaled.safetensors
+  (documented recommended default; the repo also holds krea2_turbo_bf16,
+  _int8_convrot, _mxfp8, _nvfp4 and krea2_raw_bf16 / _fp8_scaled /
+  _int8_convrot)
+     - ComfyUI/models/text_encoders/qwen3vl_4b_fp8_scaled.safetensors (bf16
+  also available)
+     - ComfyUI/models/vae/qwen_image_vae.safetensors
+
+  3. Put the LoRA .safetensors in ComfyUI/models/loras/ - flat, no per-model
+  subfolder required.
+
+  4. Build or load the graph. The official image_krea2_turbo_t2i.json
+  template is: UNETLoader (krea2_turbo_fp8_scaled.safetensors) ->
+  LoraLoaderModelOnly -> KSampler; CLIPLoader
+  (qwen3vl_4b_fp8_scaled.safetensors, type "krea2") -> text encode ->
+  KSampler positive, with ConditioningZeroOut into negative; VAELoader
+  (qwen_image_vae.safetensors) -> VAEDecode -> SaveImage; EmptyLatentImage
+  1024x1024 driven by ResolutionSelector. An int8 sibling template
+  (image_krea2_turbo_t2i_int8.json) also ships.
+
+  5. Load the LoRA with LoraLoaderModelOnly, not LoraLoader - both official
+  templates do. Krea 2 LoRAs patch the DiT only; note that the DiT's own
+  text_fusion / txtfusion blocks are a normal LoRA target (the official style
+  LoRAs patch them), but the Qwen3-VL text encoder itself is not. Set
+  strength 1.0 for Krea's official style LoRAs (the shipped template uses
+  0.8) and add the LoRA's trigger word to the prompt.
+
+  6. Sampler settings: steps 8, cfg 1, sampler euler, scheduler simple,
+  denoise 1.0. Do not add a ModelSampling node for plain t2i - ComfyUI
+  already applies shift 1.15 from Krea2.sampling_settings.
+
+  7. For style reference, use the dedicated template: UNETLoader
+  krea2_turbo_int8_convrot.safetensors, LoraLoaderModelOnly
+  krea2_style_reference.safetensors at 1.0, ModelSamplingFlux (1.15, 0.5),
+  TextEncodeQwenImageEditPlus fed by LoadImage,
+  FluxKontextMultiReferenceLatentMethod set to index_timestep_zero, and
+  SamplerCustomAdvanced with KSamplerSelect euler + BasicScheduler simple/8 +
+  CFGGuider cfg 1.
+
+  8. For instruction editing (Krea 2 Identity Edit), stock nodes are not
+  enough: install the ComfyUI-Krea2Edit node pack, which ships its own
+  workflows, and run Turbo at 8-12 steps / cfg 1 (or Raw at 20 steps / cfg 3
+  for removals) with LoRA strength 1.0.
+
+  9. Low-VRAM alternative: install city96's ComfyUI-GGUF and use a community
+  GGUF build (realrebelai/KREA-2_GGUFs, quantized from Turbo;
+  molbal/krea2-gguf, quantized from Raw) with the Unet Loader (GGUF) node in
+  place of UNETLoader. There is no official GGUF release. LoRA loading is
+  unchanged.
+
+───────────────────────────────────────────────────────────────────────────────
+  Scraped by lorakit on 2026-09-05 00:11 UTC.
+  Stats and description are the author's; pros/cons are derived from the
+  evidence tagged beside each line.
+───────────────────────────────────────────────────────────────────────────────

bartstyle-blue-archive-art-style-lora-anime-flat-color/metadata.json

@@ -0,0 +1,144 @@
+{
+  "name": "BArtstyle | Blue Archive Art Style LoRA | Anime Flat Color | 蔚蓝档案画风模型 | ブルーアーカイブ画風モデル",
+  "slug": "bartstyle-blue-archive-art-style-lora-anime-flat-color",
+  "base_model": "Krea 2",
+  "media_type": "image",
+  "model_id": 29215,
+  "version_id": 3220090,
+  "version_name": "Krea2-V2",
+  "source": "civitai",
+  "url": "https://civitai.com/models/29215",
+  "repo_id": null,
+  "file_path": null,
+  "author": "4BEraser",
+  "category": {
+    "key": "style",
+    "label": "Visual style"
+  },
+  "modalities": [],
+  "primary_modality": "unspecified",
+  "checkpoint_family": "unknown",
+  "trigger_words": [
+    "azure youth anime art style"
+  ],
+  "recommended_settings": {
+    "LoRA strength": "0.6–0.7",
+    "Resolution": "1024x768",
+    "Sampler": "DPM++ 2M"
+  },
+  "chaining": {
+    "author": {
+      "stacks": "not stated",
+      "notes": "",
+      "avoid": [],
+      "requires": [],
+      "order": "",
+      "strength": "",
+      "stated": false
+    },
+    "general_rules": [
+      "The most important rule is not about chaining but about which checkpoint you chain onto: the official README says in bold \"TRAIN on Raw and RUN on Turbo\", and \"We highly recommend using RAW for training LoRAs and applying them on Turbo for inference.\" LoRAs trained on Raw are stated to work well on Turbo.",
+      "Chain with LoraLoaderModelOnly nodes in series (model -> model). Krea 2 LoRAs are DiT-only, so there is no CLIP side to chain and the text-encoder branch stays untouched however many LoRAs you add.",
+      "Krea 2 LoRAs stack in practice and the community treats multi-LoRA as routine - style + character, style + slider, realism + character. Several CivitAI Krea 2 authors state their LoRAs stack cleanly; one writes \"I recently stacked over 20 of my own LoRAs on a single generation and the output still came out great.\"",
+      "Reduce per-LoRA strength when stacking. CivitAI Krea 2 authors converge on dropping from ~1.0 to roughly 0.6-0.9 once more than one adapter is loaded. The CivitAI 12GB training guide also notes rank affects composability: \"rank 8 produced a lighter style influence that was easier to stack with other LoRAs, rank 16 provided a useful balance between concept strength and composability, and rank 32 was better suited to a precise object or concept intended to take priority over other stacked LoRAs.\"",
+      "Accelerator LoRAs are position-sensitive in intent, not in node order: a Raw-to-Turbo delta adapter belongs on the Raw checkpoint, a sub-8-step LoRA belongs on Turbo (or on Raw already carrying a Raw-to-Turbo adapter). Chaining Raw -> Raw-to-Turbo adapter -> 4-step LoRA is explicitly recommended by the 4-step LoRA's author.",
+      "Whatever you stack, the schedule must stay Turbo's: cfg 1.0 in ComfyUI (guidance 0.0 in diffusers) and mu/shift 1.15. The 4-step LoRA's card says \"Keep mu = 1.15. The training targets are anchored to that grid; a different shift moves the [sampler off them]\"; the same reasoning applies to any Turbo-trained adapter in the chain.",
+      "Reference/edit LoRAs are designed to be stacked under ordinary subject and style LoRAs - the Identity Edit card states \"Composes with your LoRAs: character/body/style LoRAs stack on top and steer the prior\" and recommends a stacked subject LoRA when unusual subjects drift toward the base prior.",
+      "Krea 2 LoRAs cannot be chained with LoRAs for any other base model. Many CivitAI pages carry multi-model titles like \"(Krea-2 + ZIT)\" or \"[Flux | ZIB | Krea2]\" - those are separate files for separate architectures shipped under one model page, not a combined adapter.",
+      "To combine two full Krea 2 checkpoints rather than adapters, ComfyUI ships a dedicated ModelMergeKrea2 node (added in v0.27.0) with per-section weights for first., tmlp., txtmlp., tproj., txtfusion.projector, txtfusion.layerwise_blocks.0-1, txtfusion.refiner_blocks.0-1, blocks.0-27 and last.",
+      "The author never states which weight family this targets. Silent degradation, never a hard error, between Raw and Turbo. The two share one architecture - ai-toolkit's config is commented \"The reference 'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)\" - so every LoRA tensor shape matches on both and ComfyUI/diffuser"
+    ]
+  },
+  "pros": [
+    "Widely adopted — 28,534 downloads. [source: CivitAI stats]",
+    "Positively received — 2,683 thumbs-up, 1 down. [source: CivitAI stats]",
+    "Explicit trigger word(s) — azure youth anime art style — so the effect can be turned on and off from the prompt. [source: model metadata]",
+    "Author published concrete recommended settings. [source: description]"
+  ],
+  "cons": [
+    "Author says the LoRA may be overtrained — start at low strength. [source: description]",
+    "Needs a reduced strength; the default 1.0 is likely too strong. [source: description]",
+    "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]",
+    "Large at 895 MB — slower to load and heavier to stack. [source: file size]"
+  ],
+  "file": {
+    "name": "BArtstyle-KREA2-V2-1024.safetensors",
+    "size_bytes": 938539208,
+    "size_mb": 895.06,
+    "sha256": "9c19dc382d92002bfc53ece7df5556acc288579af9f0719752059dfae7fe9f70"
+  },
+  "stats": {
+    "downloads": 28534,
+    "thumbs_up": 2683,
+    "thumbs_down": 1
+  },
+  "popularity": {
+    "downloads": 28534,
+    "thumbs_up": 2683,
+    "thumbs_down": 1,
+    "comments": 0,
+    "approval": "100% positive (2683/2684)",
+    "age_days": 24.2,
+    "downloads_per_day": 82.34,
+    "rank_by_downloads": 21,
+    "rank_by_momentum": 34,
+    "of_total": 150,
+    "adoption_tier": "High",
+    "momentum": "Trending · fresh release"
+  },
+  "selection_score": 61130.0,
+  "selection_reasons": [
+    "28,534 downloads",
+    "2,683 thumbs-up",
+    "documented triggers + usage notes",
+    "top-ranked for its category"
+  ],
+  "example_media": [
+    {
+      "file": "01_image.jpeg",
+      "type": "image",
+      "resolution": "1288x1936",
+      "prompt": ""
+    },
+    {
+      "file": "02_image.jpeg",
+      "type": "image",
+      "resolution": "1536x2728",
+      "prompt": ""
+    },
+    {
+      "file": "03_image.jpeg",
+      "type": "image",
+      "resolution": "1536x2728",
+      "prompt": ""
+    },
+    {
+      "file": "04_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1024",
+      "prompt": ""
+    },
+    {
+      "file": "05_image.jpeg",
+      "type": "image",
+      "resolution": "1536x2728",
+      "prompt": ""
+    },
+    {
+      "file": "06_image.jpeg",
+      "type": "image",
+      "resolution": "1288x1936",
+      "prompt": ""
+    }
+  ],
+  "tags": [
+    "style",
+    "game art",
+    "artstyle",
+    "blue archive",
+    "anime"
+  ],
+  "scraped_at": "2026-09-05T00:11:48+00:00",
+  "path": "krea-2/bartstyle-blue-archive-art-style-lora-anime-flat-color"
+}

famegrid-krea-2-z-image-qwen-realism-lora/famegrid_spicy.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b4a37a9b9ec3885bf39242f2d4f691f186d959cbaaa30b0630e4631767b425e
+size 1562410408

famegrid-krea-2-z-image-qwen-realism-lora/info.txt

@@ -0,0 +1,396 @@
+═══════════════════════════════════════════════════════════════════════════════
+  FAMEGRID (KREA 2 / Z-IMAGE / QWEN) - REALISM LORA
+═══════════════════════════════════════════════════════════════════════════════
+
+───────────────────────────────────────────────────────────────────────────────
+  IDENTITY
+───────────────────────────────────────────────────────────────────────────────
+  Base model        Krea 2
+  Generation mode   Instruction / reference editing (community, not official)
+  Checkpoint        Krea 2 RAW (krea/Krea-2-Raw)
+  Category          Photorealism
+  Author            UltraMuse
+  Source            https://civitai.com/models/2088956
+  Provenance        CivitAI · model 2088956 · version 3278885
+
+───────────────────────────────────────────────────────────────────────────────
+  WEIGHTS
+───────────────────────────────────────────────────────────────────────────────
+  File              famegrid_spicy.safetensors
+  Version           Krea 2 Spice Preview FIX
+  Size              1,490.0 MB
+  Format            SafeTensor / unknown precision
+  SHA256            7b4a37a9b9ec3885bf39242f2d4f691f186d959cbaaa30b0630e4631767b425e
+  Published         2026-08-29
+
+───────────────────────────────────────────────────────────────────────────────
+  TRIGGER WORDS
+───────────────────────────────────────────────────────────────────────────────
+  › "famegrid"
+
+───────────────────────────────────────────────────────────────────────────────
+  RECOMMENDED SETTINGS
+───────────────────────────────────────────────────────────────────────────────
+  Taken verbatim from the author's description:
+
+  LoRA strength       0.3–0.6
+  Sampler steps       8–16
+  CFG / guidance      1.0
+
+───────────────────────────────────────────────────────────────────────────────
+  ADOPTION & POPULARITY
+───────────────────────────────────────────────────────────────────────────────
+  Downloads               39,677
+  Rating                  100% positive (1818/1826)
+  Comments                89
+  Published               6 days ago
+  Download rate           522.5/day since release
+
+  Adoption rank           #11 of 150 in this collection  (Top tier)
+  Momentum rank           #2 of 150 by download rate  (Hot · 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 Krea 2 ones, not anything specific to this LoRA.
+
+  General Krea 2 rules that apply here (see REFERENCE.txt):
+
+  • The most important rule is not about chaining but about which
+    checkpoint you chain onto: the official README says in bold "TRAIN on
+    Raw and RUN on Turbo", and "We highly recommend using RAW for training
+    LoRAs and applying them on Turbo for inference." LoRAs trained on Raw
+    are stated to work well on Turbo.
+  • Chain with LoraLoaderModelOnly nodes in series (model -> model). Krea 2
+    LoRAs are DiT-only, so there is no CLIP side to chain and the text-
+    encoder branch stays untouched however many LoRAs you add.
+  • Krea 2 LoRAs stack in practice and the community treats multi-LoRA as
+    routine - style + character, style + slider, realism + character.
+    Several CivitAI Krea 2 authors state their LoRAs stack cleanly; one
+    writes "I recently stacked over 20 of my own LoRAs on a single
+    generation and the output still came out great."
+  • Reduce per-LoRA strength when stacking. CivitAI Krea 2 authors converge
+    on dropping from ~1.0 to roughly 0.6-0.9 once more than one adapter is
+    loaded. The CivitAI 12GB training guide also notes rank affects
+    composability: "rank 8 produced a lighter style influence that was
+    easier to stack with other LoRAs, rank 16 provided a useful balance
+    between concept strength and composability, and rank 32 was better
+    suited to a precise object or concept intended to take priority over
+    other stacked LoRAs."
+  • Accelerator LoRAs are position-sensitive in intent, not in node order:
+    a Raw-to-Turbo delta adapter belongs on the Raw checkpoint, a
+    sub-8-step LoRA belongs on Turbo (or on Raw already carrying a Raw-to-
+    Turbo adapter). Chaining Raw -> Raw-to-Turbo adapter -> 4-step LoRA is
+    explicitly recommended by the 4-step LoRA's author.
+  • Whatever you stack, the schedule must stay Turbo's: cfg 1.0 in ComfyUI
+    (guidance 0.0 in diffusers) and mu/shift 1.15. The 4-step LoRA's card
+    says "Keep mu = 1.15. The training targets are anchored to that grid; a
+    different shift moves the [sampler off them]"; the same reasoning
+    applies to any Turbo-trained adapter in the chain.
+  • Reference/edit LoRAs are designed to be stacked under ordinary subject
+    and style LoRAs - the Identity Edit card states "Composes with your
+    LoRAs: character/body/style LoRAs stack on top and steer the prior" and
+    recommends a stacked subject LoRA when unusual subjects drift toward
+    the base prior.
+  • Krea 2 LoRAs cannot be chained with LoRAs for any other base model.
+    Many CivitAI pages carry multi-model titles like "(Krea-2 + ZIT)" or
+    "[Flux | ZIB | Krea2]" - those are separate files for separate
+    architectures shipped under one model page, not a combined adapter.
+  • To combine two full Krea 2 checkpoints rather than adapters, ComfyUI
+    ships a dedicated ModelMergeKrea2 node (added in v0.27.0) with per-
+    section weights for first., tmlp., txtmlp., tproj.,
+    txtfusion.projector, txtfusion.layerwise_blocks.0-1,
+    txtfusion.refiner_blocks.0-1, blocks.0-27 and last.
+  • Undistilled base checkpoint, no step or CFG distillation and no post-
+    training. The card says "its not recommended for inference use. Its a
+    good base for finetuning or post-training". Official settings: 52
+    steps, CFG 3.5, trained up to 1K. This is where Krea tells you to train
+    LoRAs.
+
+───────────────────────────────────────────────────────────────────────────────
+  PROS
+───────────────────────────────────────────────────────────────────────────────
+  + Widely adopted — 39,677 downloads. [source: CivitAI stats]
+  + Positively received — 1,818 thumbs-up, 8 down. [source: CivitAI stats]
+  + Explicit trigger word(s) — famegrid — so the effect can be turned on
+    and off from the prompt. [source: model metadata]
+  + Author published concrete recommended settings. [source: description]
+
+───────────────────────────────────────────────────────────────────────────────
+  CONS & CAVEATS
+───────────────────────────────────────────────────────────────────────────────
+  - Author labels this experimental — results vary shot to shot. [source:
+    description]
+  - Needs a reduced strength; the default 1.0 is likely too strong.
+    [source: description]
+  - Author references NSFW output — review before using in shared work.
+    [source: description]
+  - Large at 1490 MB — slower to load and heavier to stack. [source: file
+    size]
+
+───────────────────────────────────────────────────────────────────────────────
+  COMPATIBILITY WARNING
+───────────────────────────────────────────────────────────────────────────────
+  Silent degradation, never a hard error, between Raw and Turbo. The two
+  share one architecture - ai-toolkit's config is commented "The reference
+  'single_mmdit_large_wide' architecture (oss_raw / oss_turbo share it)" - so
+  every LoRA tensor shape matches on both and ComfyUI/diffusers will load an
+  adapter against either without complaint. The lvladikov 4-step LoRA card
+  states this for the reverse direction: "Every layer it targets also exists
+  in Krea 2 Raw, so it will load there without complaint - but that is a side
+  effect of the shared architecture, not a supported mode", and reports
+  mixed, subject-dependent results with malformed subjects (duplicated heads,
+  fused limbs) at low step counts, needing 14+ steps with Raw's CFG on before
+  output is coherent. The intended direction (train on Raw, run on Turbo) is
+  officially blessed. The more common practical failure is a settings
+  mismatch rather than a weight mismatch: running a Turbo-targeted LoRA at
+  Raw's 52 steps + CFG 3.5, or leaving CFG enabled on Turbo, degrades output
+  with no warning. Loading a Krea 2 LoRA against a different model family
+  entirely (Flux, Qwen-Image, Z-Image) is also not an error - ComfyUI logs
+  "lora key not loaded: {key}" per key and generates as if no LoRA were
+  applied.
+
+───────────────────────────────────────────────────────────────────────────────
+  AUTHOR'S DESCRIPTION
+───────────────────────────────────────────────────────────────────────────────
+  🚀 New Release — Krea 2 Spice Preview Preview Fix update: Apologies for the
+  screw-up; I uploaded the wrong checkpoint in the last release. Replace that
+  one with this here. WORKFLOW: https://pastebin.com/GKMm2Njs Still use the
+  original model page to upload to the gallery ----------
+  This preview model was trained on approximately 1/20th of the dataset
+  planned for the final release. It builds on FameGrid Natural’s realistic,
+  everyday photographic style, producing casual images that look like
+  authentic modern smartphone photos rather than polished or heavily staged
+  shoots. Important: If using with character lora keep strength 0.3-0.6. You
+  can also try turning down character lora as much as possible to keep
+  likeness and turn up the Famegrid lora more.
+  The preview currently has minimal NSFW support involving men and works with
+  the same workflow as FameGrid Natural V1.
+  Contribute to training of the full version. Cloud GPU training has become
+  too expensive for me to continue funding these models alone. The final
+  version will therefore be released as an exclusive model for contributors
+  only (as crowdfunding didn't work).
+  Contributions over $10 will receive the final model once training is
+  complete. After the funding threshold is reached, local access will be
+  available exclusively to qualifying contributors—there will be no other way
+  to obtain the model for local use.
+  The model will also be hosted on FameGrid.ai.
+  Secure your contributor access here:
+  https://ultramuse.art/train-lora/fundraiser
+  Very limited contributor spaces are available.
+  🔑 Trigger Word Famegrid Start your prompt with the trigger word.
+  Example:
+  Famegrid, casual smartphone photo of a woman sitting at a neighborhood
+  coffee shop ⚙️ Recommended LoRA Weight Recommended range: 0.3–1.0
+  - Use a lower weight for a subtle influence.
+
+  - Use a higher weight for a stronger FameGrid Natural aesthetic.
+
+  - Adjust within the recommended range to suit your prompt and workflow.
+
+  📥 Workflow Available on Discord The FameGrid Natural Krea 2 workflow is
+  available for free in the Free Resources channel of the UltraMuse Discord.
+  👉 Join the UltraMuse Discord and Download the Workflow 👉 You can also use
+  this model and train LoRAs at Famegrid.ai
+
+  🎬 Introducing Creator Studio The All-in-One AI Influencer Operating System
+  Creator Studio brings the entire AI influencer workflow into one connected
+  production studio.
+  Plan content, create consistent characters, generate and edit images,
+  produce videos and audio, write captions, publish posts, and review
+  performance—all from one creator-focused platform.
+  We tested more than 100 AI models and platforms, selected the strongest
+  options, and put them inside one useful wrapper with the tools creators
+  need to produce high-quality AI content.
+  Whether you manage one virtual creator or an entire roster, Creator Studio
+  helps turn scattered generations into a repeatable content-production
+  system.
+  Bring Your Own API or ComfyUI Workflow Creator Studio is not a closed
+  ecosystem. Add any API or import a custom ComfyUI workflow, give the agent
+  the documentation, and it can figure out the integration, build the
+  controls, and generate the UI for it.
+  All 27 Creator Studio Tools Create and Direct • Copilot • Image Generation
+  • Image Edit • AI Swap • Copy
+  Characters and Campaigns • Photoshoot Studio • Character Creation •
+  Character Manager
+  Video and Sound • Video Studio • Motion Control • Audio Studio • Lipsync
+  Studio
+  Finish and Enhance • Upscale Studio • Skin Enhancement • Post-Process
+  Workflows and Operations • ComfyUI Workflows • Provider Builder • Caption
+  Studio • Publish Studio • Social Analytics • LoRA Training
+  Automated Flows • Inspiration Dump • InstaClone • InstaReel • UGC Creator •
+  Hybrid Lipsync • Watermark
+  See It in Action https://ultramuse.art/creatorstudio/demo/
+  ━━━━━━━━━━━━━━━━━━━━ 🧬 Krea 2 & Z-Image LoRA Trainer Train your own custom
+  character and style LoRAs for Krea 2 and Z-Image .
+  The trainer helps you prepare your dataset, configure the training, monitor
+  its progress, and create reusable LoRAs for consistent AI influencers,
+  characters, styles, and creator content.
+  The Krea 2 and Z-Image LoRA Trainer is available through the UltraMuse
+  Discord server:
+  👉 [Access the LoRA Trainer Here](https://ultramuse.art/train-lora]
+  🚀 Newest Release — July 2026 🔥 FameGrid Krea 2 Standard v1 The Highest-
+  Quality FameGrid Yet FameGrid Krea 2 is the highest-quality FameGrid
+  release yet. It produces excellent social-media-style images with natural
+  posing, believable skin, polished lighting, strong compositions, and the
+  modern creator aesthetic FameGrid was built for.
+  This release is designed for AI influencers, lifestyle content, fashion and
+  beauty photography, creator campaigns, product promotions, and premium
+  social posts. It delivers cleaner, more convincing results while keeping
+  images visually engaging and ready for a feed.
+  What It Does • Creates premium social-media and creator-style imagery
+  • Produces natural poses and believable body language
+  • Improves skin, lighting, styling, and overall visual polish
+  • Excels at lifestyle, fashion, beauty, influencer, and promotional content
+  • Maintains a clean, modern aesthetic without overly complicated prompts
+  Trigger Word famegrid
+  Start your prompt with the trigger word. Example:
+  famegrid, candid social media photo of...
+  Recommended Starting Settings • Sampler: er_sde
+  • Guidance: 1
+  • Steps: 8–16
+  These are starting points. Adjust the LoRA weight, steps, and workflow to
+  suit the look you want.
+  Best Uses ✨ Social media posts • AI influencers and virtual creators •
+  Lifestyle photography • Fashion and beauty content • Product promotions •
+  Marketing campaigns • Character-driven content
+  Workflow Available on Discord The official FameGrid Krea 2 workflow is
+  available through the UltraMuse Discord, along with updates, tutorials,
+  support, and additional resources.
+  https://discord.gg/4zbGm5j6jW 🚀 Support FameGrid Krea 2 V2 (Enhanced NSFW)
+  Help fund the next version of FameGrid Krea 2 , including improved realism,
+  anatomy, posing, prompt adherence, social-media image quality, and expanded
+  NSFW support.
+  👉 [Support FameGrid Krea 2 V2]( https://ultramuse.art/train-
+  lora/fundraiser)
+  ━━━━━━━━━━━━━━━━━━━━
+  📚 Previous FameGrid Releases 🔥 FameGrid Spice V2 — June 2026 Spice V2
+  delivered a major jump in skin quality, posing, prompt following, body-type
+  variety, male generation quality, ethnicity representation, and artifact
+  reduction.
+  Recommended Spice V2 Settings • Sampler: sa_diag_implicit / qin_zhang_2s
+  • Scheduler: beta57
+  • Steps: 9 (20 with the ClownShark BetaSamplingScheduler)
+  • CFG: 1.0 • ETA: 0.50 • Denoise: 1.0
+  • Sampler Mode: resample • Bongmath: true
+  🚀 FameGrid Revolution The previous Z-Image flagship, compatible with
+  Z-Image Base and Turbo and built for strong prompt adherence, influencer
+  photography, and social-media aesthetics.
+  Available variants include FameGrid Revolution Bold, More Bold, ExtraReal,
+  and UltraReal.
+  Model and workflow:
+  https://civitai.com/models/2431502/famegrid-revolution-hybrid-workflow-zib-
+  zit
+  🎨 FameGrid Qwen Series The FameGrid Qwen releases introduced improved skin
+  realism and more believable subjects.
+  Trigger Words IGMODEL — influencer-style realism
+  rlskn — enhanced skin texture and natural imperfections
+  Recommended Qwen Settings • Model Flow: 2 • Sampler: res_2s • Scheduler:
+  simple
+  • CFG: 1–3 (1 recommended) • LoRA Weight: 1.0–2.0 (1.0 recommended)
+  ━━━━━━━━━━━━━━━━━━━━
+  💬 Community and Resources Get official workflows, updates, checkpoints,
+  tutorials, prompt resources, troubleshooting help, experimental releases,
+  community showcases, and early access in the UltraMuse Discord.
+  Discord: https://discord.gg/4zbGm5j6jW
+  UltraMuse: https://ultramuse.art
+  ━━━━━━━━━━━━━━━━━━━━
+  ❤️ About FameGrid FameGrid was built around one goal: to create modern
+  social-media-quality imagery that feels believable, natural, and visually
+  compelling.
+  The focus has always been natural posing, realistic skin, clean
+  compositions, modern lighting, strong prompt adherence, and professional
+  creator aesthetics.
+
+───────────────────────────────────────────────────────────────────────────────
+  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, 1248x1824)
+  02_image.jpeg  (image, 1248x1824)
+  03_image.jpeg  (image, 1200x1600)
+  04_image.jpeg  (image, 1200x1600)
+  05_image.jpeg  (image, 1248x1824)
+  06_image.jpeg  (image, 1248x1824)
+
+───────────────────────────────────────────────────────────────────────────────
+  HOW TO USE (COMFYUI)
+───────────────────────────────────────────────────────────────────────────────
+  1. Update ComfyUI to at least v0.26.0 (released 2026-06-23), which added
+  local Krea2 support. For reference-image / style-reference / edit LoRAs you
+  need at least v0.29.0 (2026-07-29), which added regular and timestep-zero
+  reference images for Krea 2. ModelMergeKrea2 arrived in v0.27.0
+  (2026-06-30).
+
+  2. Put the base stack in place (files from huggingface.co/Comfy-Org/Krea-2,
+  which is not gated):
+     - ComfyUI/models/diffusion_models/krea2_turbo_fp8_scaled.safetensors
+  (documented recommended default; the repo also holds krea2_turbo_bf16,
+  _int8_convrot, _mxfp8, _nvfp4 and krea2_raw_bf16 / _fp8_scaled /
+  _int8_convrot)
+     - ComfyUI/models/text_encoders/qwen3vl_4b_fp8_scaled.safetensors (bf16
+  also available)
+     - ComfyUI/models/vae/qwen_image_vae.safetensors
+
+  3. Put the LoRA .safetensors in ComfyUI/models/loras/ - flat, no per-model
+  subfolder required.
+
+  4. Build or load the graph. The official image_krea2_turbo_t2i.json
+  template is: UNETLoader (krea2_turbo_fp8_scaled.safetensors) ->
+  LoraLoaderModelOnly -> KSampler; CLIPLoader
+  (qwen3vl_4b_fp8_scaled.safetensors, type "krea2") -> text encode ->
+  KSampler positive, with ConditioningZeroOut into negative; VAELoader
+  (qwen_image_vae.safetensors) -> VAEDecode -> SaveImage; EmptyLatentImage
+  1024x1024 driven by ResolutionSelector. An int8 sibling template
+  (image_krea2_turbo_t2i_int8.json) also ships.
+
+  5. Load the LoRA with LoraLoaderModelOnly, not LoraLoader - both official
+  templates do. Krea 2 LoRAs patch the DiT only; note that the DiT's own
+  text_fusion / txtfusion blocks are a normal LoRA target (the official style
+  LoRAs patch them), but the Qwen3-VL text encoder itself is not. Set
+  strength 1.0 for Krea's official style LoRAs (the shipped template uses
+  0.8) and add the LoRA's trigger word to the prompt.
+
+  6. Sampler settings: steps 8, cfg 1, sampler euler, scheduler simple,
+  denoise 1.0. Do not add a ModelSampling node for plain t2i - ComfyUI
+  already applies shift 1.15 from Krea2.sampling_settings.
+
+  7. For style reference, use the dedicated template: UNETLoader
+  krea2_turbo_int8_convrot.safetensors, LoraLoaderModelOnly
+  krea2_style_reference.safetensors at 1.0, ModelSamplingFlux (1.15, 0.5),
+  TextEncodeQwenImageEditPlus fed by LoadImage,
+  FluxKontextMultiReferenceLatentMethod set to index_timestep_zero, and
+  SamplerCustomAdvanced with KSamplerSelect euler + BasicScheduler simple/8 +
+  CFGGuider cfg 1.
+
+  8. For instruction editing (Krea 2 Identity Edit), stock nodes are not
+  enough: install the ComfyUI-Krea2Edit node pack, which ships its own
+  workflows, and run Turbo at 8-12 steps / cfg 1 (or Raw at 20 steps / cfg 3
+  for removals) with LoRA strength 1.0.
+
+  9. Low-VRAM alternative: install city96's ComfyUI-GGUF and use a community
+  GGUF build (realrebelai/KREA-2_GGUFs, quantized from Turbo;
+  molbal/krea2-gguf, quantized from Raw) with the Unet Loader (GGUF) node in
+  place of UNETLoader. There is no official GGUF release. LoRA loading is
+  unchanged.
+
+───────────────────────────────────────────────────────────────────────────────
+  Scraped by lorakit on 2026-09-05 00:11 UTC.
+  Stats and description are the author's; pros/cons are derived from the
+  evidence tagged beside each line.
+───────────────────────────────────────────────────────────────────────────────

famegrid-krea-2-z-image-qwen-realism-lora/metadata.json

@@ -0,0 +1,152 @@
+{
+  "name": "Famegrid (Krea 2 / Z-Image / Qwen) - Realism LoRA",
+  "slug": "famegrid-krea-2-z-image-qwen-realism-lora",
+  "base_model": "Krea 2",
+  "media_type": "image",
+  "model_id": 2088956,
+  "version_id": 3278885,
+  "version_name": "Krea 2 Spice Preview FIX",
+  "source": "civitai",
+  "url": "https://civitai.com/models/2088956",
+  "repo_id": null,
+  "file_path": null,
+  "author": "UltraMuse",
+  "category": {
+    "key": "photoreal",
+    "label": "Photorealism"
+  },
+  "modalities": [
+    "edit"
+  ],
+  "primary_modality": "edit",
+  "checkpoint_family": "raw",
+  "trigger_words": [
+    "famegrid"
+  ],
+  "recommended_settings": {
+    "LoRA strength": "0.3–0.6",
+    "Sampler steps": "8–16",
+    "CFG / guidance": "1.0"
+  },
+  "chaining": {
+    "author": {
+      "stacks": "not stated",
+      "notes": "",
+      "avoid": [],
+      "requires": [],
+      "order": "",
+      "strength": "",
+      "stated": false
+    },
+    "general_rules": [
+      "The most important rule is not about chaining but about which checkpoint you chain onto: the official README says in bold \"TRAIN on Raw and RUN on Turbo\", and \"We highly recommend using RAW for training LoRAs and applying them on Turbo for inference.\" LoRAs trained on Raw are stated to work well on Turbo.",
+      "Chain with LoraLoaderModelOnly nodes in series (model -> model). Krea 2 LoRAs are DiT-only, so there is no CLIP side to chain and the text-encoder branch stays untouched however many LoRAs you add.",
+      "Krea 2 LoRAs stack in practice and the community treats multi-LoRA as routine - style + character, style + slider, realism + character. Several CivitAI Krea 2 authors state their LoRAs stack cleanly; one writes \"I recently stacked over 20 of my own LoRAs on a single generation and the output still came out great.\"",
+      "Reduce per-LoRA strength when stacking. CivitAI Krea 2 authors converge on dropping from ~1.0 to roughly 0.6-0.9 once more than one adapter is loaded. The CivitAI 12GB training guide also notes rank affects composability: \"rank 8 produced a lighter style influence that was easier to stack with other LoRAs, rank 16 provided a useful balance between concept strength and composability, and rank 32 was better suited to a precise object or concept intended to take priority over other stacked LoRAs.\"",
+      "Accelerator LoRAs are position-sensitive in intent, not in node order: a Raw-to-Turbo delta adapter belongs on the Raw checkpoint, a sub-8-step LoRA belongs on Turbo (or on Raw already carrying a Raw-to-Turbo adapter). Chaining Raw -> Raw-to-Turbo adapter -> 4-step LoRA is explicitly recommended by the 4-step LoRA's author.",
+      "Whatever you stack, the schedule must stay Turbo's: cfg 1.0 in ComfyUI (guidance 0.0 in diffusers) and mu/shift 1.15. The 4-step LoRA's card says \"Keep mu = 1.15. The training targets are anchored to that grid; a different shift moves the [sampler off them]\"; the same reasoning applies to any Turbo-trained adapter in the chain.",
+      "Reference/edit LoRAs are designed to be stacked under ordinary subject and style LoRAs - the Identity Edit card states \"Composes with your LoRAs: character/body/style LoRAs stack on top and steer the prior\" and recommends a stacked subject LoRA when unusual subjects drift toward the base prior.",
+      "Krea 2 LoRAs cannot be chained with LoRAs for any other base model. Many CivitAI pages carry multi-model titles like \"(Krea-2 + ZIT)\" or \"[Flux | ZIB | Krea2]\" - those are separate files for separate architectures shipped under one model page, not a combined adapter.",
+      "To combine two full Krea 2 checkpoints rather than adapters, ComfyUI ships a dedicated ModelMergeKrea2 node (added in v0.27.0) with per-section weights for first., tmlp., txtmlp., tproj., txtfusion.projector, txtfusion.layerwise_blocks.0-1, txtfusion.refiner_blocks.0-1, blocks.0-27 and last.",
+      "Undistilled base checkpoint, no step or CFG distillation and no post-training. The card says \"its not recommended for inference use. Its a good base for finetuning or post-training\". Official settings: 52 steps, CFG 3.5, trained up to 1K. This is where Krea tells you to train LoRAs."
+    ]
+  },
+  "pros": [
+    "Widely adopted — 39,677 downloads. [source: CivitAI stats]",
+    "Positively received — 1,818 thumbs-up, 8 down. [source: CivitAI stats]",
+    "Explicit trigger word(s) — famegrid — so the effect can be turned on and off from the prompt. [source: model metadata]",
+    "Author published concrete recommended settings. [source: description]"
+  ],
+  "cons": [
+    "Author labels this experimental — results vary shot to shot. [source: description]",
+    "Needs a reduced strength; the default 1.0 is likely too strong. [source: description]",
+    "Author references NSFW output — review before using in shared work. [source: description]",
+    "Large at 1490 MB — slower to load and heavier to stack. [source: file size]"
+  ],
+  "file": {
+    "name": "famegrid_spicy.safetensors",
+    "size_bytes": 1562410408,
+    "size_mb": 1490.03,
+    "sha256": "7b4a37a9b9ec3885bf39242f2d4f691f186d959cbaaa30b0630e4631767b425e"
+  },
+  "stats": {
+    "downloads": 39677,
+    "thumbs_up": 1818,
+    "thumbs_down": 8
+  },
+  "popularity": {
+    "downloads": 39677,
+    "thumbs_up": 1818,
+    "thumbs_down": 8,
+    "comments": 89,
+    "approval": "100% positive (1818/1826)",
+    "age_days": 6.0,
+    "downloads_per_day": 522.47,
+    "rank_by_downloads": 11,
+    "rank_by_momentum": 2,
+    "of_total": 150,
+    "adoption_tier": "Top tier",
+    "momentum": "Hot · fresh release"
+  },
+  "selection_score": 61893.0,
+  "selection_reasons": [
+    "39,677 downloads",
+    "1,818 thumbs-up",
+    "documented triggers + usage notes",
+    "top-ranked for its category"
+  ],
+  "example_media": [
+    {
+      "file": "01_image.jpeg",
+      "type": "image",
+      "resolution": "1248x1824",
+      "prompt": ""
+    },
+    {
+      "file": "02_image.jpeg",
+      "type": "image",
+      "resolution": "1248x1824",
+      "prompt": ""
+    },
+    {
+      "file": "03_image.jpeg",
+      "type": "image",
+      "resolution": "1200x1600",
+      "prompt": ""
+    },
+    {
+      "file": "04_image.jpeg",
+      "type": "image",
+      "resolution": "1200x1600",
+      "prompt": ""
+    },
+    {
+      "file": "05_image.jpeg",
+      "type": "image",
+      "resolution": "1248x1824",
+      "prompt": ""
+    },
+    {
+      "file": "06_image.jpeg",
+      "type": "image",
+      "resolution": "1248x1824",
+      "prompt": ""
+    }
+  ],
+  "tags": [
+    "women",
+    "character",
+    "instagram",
+    "realism",
+    "human",
+    "model",
+    "influencer",
+    "facebook",
+    "tik tok",
+    "social media",
+    "woman",
+    "sexy"
+  ],
+  "scraped_at": "2026-09-05T00:11:58+00:00",
+  "path": "krea-2/famegrid-krea-2-z-image-qwen-realism-lora"
+}