Add 12 LoRA(s): cutifyier, krea-2-identity-edit, ars-midjourney-watercolor, impressionism…

.gitattributes

@@ -244,3 +244,74 @@ krea-2-detail-slider/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs
 krea-2-detail-slider/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
 krea-2-detail-slider/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
 krea-2-detail-slider/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+cutifyier/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+cutifyier/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+cutifyier/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+cutifyier/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+cutifyier/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+cutifyier/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea-2-identity-edit/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea-2-identity-edit/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea-2-identity-edit/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea-2-identity-edit/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea-2-identity-edit/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+krea-2-identity-edit/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ars-midjourney-watercolor/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ars-midjourney-watercolor/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ars-midjourney-watercolor/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ars-midjourney-watercolor/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ars-midjourney-watercolor/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+ars-midjourney-watercolor/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+impressionism/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+impressionism/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+impressionism/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+impressionism/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+impressionism/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+impressionism/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+carefull-style-krea-2-anima-il/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+carefull-style-krea-2-anima-il/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+carefull-style-krea-2-anima-il/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+carefull-style-krea-2-anima-il/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+carefull-style-krea-2-anima-il/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+carefull-style-krea-2-anima-il/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-age-slider-zit-krea-2/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-age-slider-zit-krea-2/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-age-slider-zit-krea-2/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-age-slider-zit-krea-2/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-age-slider-zit-krea-2/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+the-age-slider-zit-krea-2/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+frank-frazetta-style-oil-painting-krea2-flux/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+frank-frazetta-style-oil-painting-krea2-flux/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+frank-frazetta-style-oil-painting-krea2-flux/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+frank-frazetta-style-oil-painting-krea2-flux/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+frank-frazetta-style-oil-painting-krea2-flux/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+frank-frazetta-style-oil-painting-krea2-flux/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+omar-dogan-style-1-5-pony-illustrious-noob-zit-anima-krea2/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+omar-dogan-style-1-5-pony-illustrious-noob-zit-anima-krea2/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+omar-dogan-style-1-5-pony-illustrious-noob-zit-anima-krea2/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+omar-dogan-style-1-5-pony-illustrious-noob-zit-anima-krea2/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+omar-dogan-style-1-5-pony-illustrious-noob-zit-anima-krea2/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+omar-dogan-style-1-5-pony-illustrious-noob-zit-anima-krea2/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+60s-psychedelic-movie/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+60s-psychedelic-movie/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+60s-psychedelic-movie/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+60s-psychedelic-movie/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+60s-psychedelic-movie/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+60s-psychedelic-movie/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+70s-sci-fi-movie/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+70s-sci-fi-movie/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+70s-sci-fi-movie/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+70s-sci-fi-movie/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+70s-sci-fi-movie/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+70s-sci-fi-movie/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+body-weight-slider-skinny-fat-slider/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+body-weight-slider-skinny-fat-slider/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+body-weight-slider-skinny-fat-slider/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+body-weight-slider-skinny-fat-slider/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+body-weight-slider-skinny-fat-slider/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text
+body-weight-slider-skinny-fat-slider/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text
+anna-yamada-and-other-the-dangers-in-my-heart-characters/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text
+anna-yamada-and-other-the-dangers-in-my-heart-characters/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text
+anna-yamada-and-other-the-dangers-in-my-heart-characters/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text
+anna-yamada-and-other-the-dangers-in-my-heart-characters/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text
+anna-yamada-and-other-the-dangers-in-my-heart-characters/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text

60s-psychedelic-movie/60sPsyKrea2Raw.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd1d53eb64721ef82770c8364d9f27e9f56d56eaa88c263bc5f0781eaba9765b
+size 228587728

60s-psychedelic-movie/info.txt

@@ -0,0 +1,284 @@
+═══════════════════════════════════════════════════════════════════════════════
+  60S PSYCHEDELIC MOVIE
+═══════════════════════════════════════════════════════════════════════════════
+
+───────────────────────────────────────────────────────────────────────────────
+  IDENTITY
+───────────────────────────────────────────────────────────────────────────────
+  Base model        Krea 2
+  Generation mode   not declared by the author
+  Checkpoint        not determined — verify before use
+  Category          Visual style
+  Author            ArsMachina
+  Source            https://civitai.com/models/878199
+  Provenance        CivitAI · model 878199 · version 3129133
+
+───────────────────────────────────────────────────────────────────────────────
+  WEIGHTS
+───────────────────────────────────────────────────────────────────────────────
+  File              60sPsyKrea2Raw.safetensors
+  Version           Krea2
+  Size              218.0 MB
+  Format            SafeTensor / unknown precision
+  SHA256            fd1d53eb64721ef82770c8364d9f27e9f56d56eaa88c263bc5f0781eaba9765b
+  Published         2026-07-30
+
+───────────────────────────────────────────────────────────────────────────────
+  TRIGGER WORDS
+───────────────────────────────────────────────────────────────────────────────
+  › "ArsMovieStill, movie still from a 60s psychedelic movie,"
+
+───────────────────────────────────────────────────────────────────────────────
+  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               11,997
+  Rating                  100% positive (870/870)
+  Comments                0
+  Published               37 days ago
+  Download rate           25.6/day since release
+
+  Adoption rank           #46 of 150 in this collection  (Moderate)
+  Momentum rank           #86 of 150 by download rate  (Long tail · fresh release)
+
+  Adoption tier is the percentile of total downloads within this collection;
+  momentum is the percentile of downloads-per-day since release. Momentum is
+  ranked rather than measured against a fixed rate, because this base model
+  is itself new and every LoRA looks fast on an absolute scale. Both numbers
+  are CivitAI's, read on the scrape date at the foot of this file.
+
+───────────────────────────────────────────────────────────────────────────────
+  CHAINING / STACKING
+───────────────────────────────────────────────────────────────────────────────
+  Stacks with others      not stated
+
+  The author gave no chaining guidance. The rules below are the
+  general 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 — 11,997 downloads. [source: CivitAI stats]
+  + Positively received — 870 thumbs-up with no down-votes. [source:
+    CivitAI stats]
+  + Explicit trigger word(s) — ArsMovieStill, movie still from a 60s
+    psychedelic movie, — so the effect can be turned on and off from the
+    prompt. [source: model metadata]
+  + Compact at 218 MB — cheap to stack with other LoRAs. [source: file
+    size]
+
+───────────────────────────────────────────────────────────────────────────────
+  CONS & CAVEATS
+───────────────────────────────────────────────────────────────────────────────
+  - Author never states which generation mode it was trained for; test
+    against your own workflow before relying on it. [source: description]
+
+───────────────────────────────────────────────────────────────────────────────
+  COMPATIBILITY WARNING
+───────────────────────────────────────────────────────────────────────────────
+  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
+───────────────────────────────────────────────────────────────────────────────
+  This LoRA tries to take you into a 60s psychedelic movie
+
+  The trigger words are ArsMovieStill, movie still from a 60s psychedelic
+  movie
+  MJ7 Update coming soon: Reworking the dataset from scratch. Enhanced period
+  accuracy.
+
+  V2 Update: The training dataset has been expanded by 300%, significantly
+  increasing its scope.
+  This update enhances the range of scenes and characters, offering greater
+  flexibility.
+
+─────────────��─────────────────────────────────────────────────────────────────
+  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, 1024x1024)
+  02_image.jpeg  (image, 1024x1024)
+  03_image.jpeg  (image, 1024x1024)
+  04_image.jpeg  (image, 1664x2432)
+  05_image.jpeg  (image, 1728x2160)
+  06_image.jpeg  (image, 2160x2704)
+
+───────────────────────────────────────────────────────────────────────────────
+  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 00:13 UTC.
+  Stats and description are the author's; pros/cons are derived from the
+  evidence tagged beside each line.
+─────────────────────────────────────────────���─────────────────────────────────

60s-psychedelic-movie/metadata.json

@@ -0,0 +1,134 @@
+{
+  "name": "60s Psychedelic Movie",
+  "slug": "60s-psychedelic-movie",
+  "base_model": "Krea 2",
+  "media_type": "image",
+  "model_id": 878199,
+  "version_id": 3129133,
+  "version_name": "Krea2",
+  "source": "civitai",
+  "url": "https://civitai.com/models/878199",
+  "repo_id": null,
+  "file_path": null,
+  "author": "ArsMachina",
+  "category": {
+    "key": "style",
+    "label": "Visual style"
+  },
+  "modalities": [],
+  "primary_modality": "unspecified",
+  "checkpoint_family": "unknown",
+  "trigger_words": [
+    "ArsMovieStill, movie still from a 60s psychedelic movie,"
+  ],
+  "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 — 11,997 downloads. [source: CivitAI stats]",
+    "Positively received — 870 thumbs-up with no down-votes. [source: CivitAI stats]",
+    "Explicit trigger word(s) — ArsMovieStill, movie still from a 60s psychedelic movie, — so the effect can be turned on and off from the prompt. [source: model metadata]",
+    "Compact at 218 MB — cheap to stack with other LoRAs. [source: file size]"
+  ],
+  "cons": [
+    "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]"
+  ],
+  "file": {
+    "name": "60sPsyKrea2Raw.safetensors",
+    "size_bytes": 228587728,
+    "size_mb": 218.0,
+    "sha256": "fd1d53eb64721ef82770c8364d9f27e9f56d56eaa88c263bc5f0781eaba9765b"
+  },
+  "stats": {
+    "downloads": 11997,
+    "thumbs_up": 870,
+    "thumbs_down": 0
+  },
+  "popularity": {
+    "downloads": 11997,
+    "thumbs_up": 870,
+    "thumbs_down": 0,
+    "comments": 0,
+    "approval": "100% positive (870/870)",
+    "age_days": 36.6,
+    "downloads_per_day": 25.64,
+    "rank_by_downloads": 46,
+    "rank_by_momentum": 86,
+    "of_total": 150,
+    "adoption_tier": "Moderate",
+    "momentum": "Long tail · fresh release"
+  },
+  "selection_score": 22837.0,
+  "selection_reasons": [
+    "11,997 downloads",
+    "870 thumbs-up",
+    "documented triggers + usage notes",
+    "top-ranked for its category"
+  ],
+  "example_media": [
+    {
+      "file": "01_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1024",
+      "prompt": ""
+    },
+    {
+      "file": "02_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1024",
+      "prompt": ""
+    },
+    {
+      "file": "03_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1024",
+      "prompt": ""
+    },
+    {
+      "file": "04_image.jpeg",
+      "type": "image",
+      "resolution": "1664x2432",
+      "prompt": ""
+    },
+    {
+      "file": "05_image.jpeg",
+      "type": "image",
+      "resolution": "1728x2160",
+      "prompt": ""
+    },
+    {
+      "file": "06_image.jpeg",
+      "type": "image",
+      "resolution": "2160x2704",
+      "prompt": ""
+    }
+  ],
+  "tags": [
+    "movie",
+    "still",
+    "style"
+  ],
+  "scraped_at": "2026-09-05T00:13:30+00:00",
+  "path": "krea-2/60s-psychedelic-movie"
+}

70s-sci-fi-movie/70sSCiFiKrea2Raw.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d0cfed436f4dd838238779adc9e1da910021885a8ca402f7a4db169cb185c3c
+size 228587728

70s-sci-fi-movie/info.txt

@@ -0,0 +1,279 @@
+═══════════════════════════════════════════════════════════════════════════════
+  70S SCI-FI MOVIE
+═══════════════════════════════════════════════════════════════════════════════
+
+───────────────────────────────────────────────────────────────────────────────
+  IDENTITY
+───────────────────────────────────────────────────────────────────────────────
+  Base model        Krea 2
+  Generation mode   not declared by the author
+  Checkpoint        not determined — verify before use
+  Category          Visual style
+  Author            ArsMachina
+  Source            https://civitai.com/models/995002
+  Provenance        CivitAI · model 995002 · version 3129771
+
+───────────────────────────────────────────────────────────────────────────────
+  WEIGHTS
+───────────────────────────────────────────────────────────────────────────────
+  File              70sSCiFiKrea2Raw.safetensors
+  Version           Krea2
+  Size              218.0 MB
+  Format            SafeTensor / unknown precision
+  SHA256            8d0cfed436f4dd838238779adc9e1da910021885a8ca402f7a4db169cb185c3c
+  Published         2026-07-30
+
+───────────────────────────────────────────────────────────────────────────────
+  TRIGGER WORDS
+───────────────────────────────────────────────────────────────────────────────
+  › "ArsMovieStill, movie still from a 1970s technicolor Sci-Fi movie,"
+
+───────────────────────────────────────────────────────────────────────────────
+  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               12,163
+  Rating                  100% positive (833/833)
+  Comments                0
+  Published               37 days ago
+  Download rate           23.7/day since release
+
+  Adoption rank           #44 of 150 in this collection  (Moderate)
+  Momentum rank           #88 of 150 by download rate  (Long tail · fresh release)
+
+  Adoption tier is the percentile of total downloads within this collection;
+  momentum is the percentile of downloads-per-day since release. Momentum is
+  ranked rather than measured against a fixed rate, because this base model
+  is itself new and every LoRA looks fast on an absolute scale. Both numbers
+  are CivitAI's, read on the scrape date at the foot of this file.
+
+───────────────────────────────────────────────────────────────────────────────
+  CHAINING / STACKING
+───────────────────────────────────────────────────────────────────────────────
+  Stacks with others      not stated
+
+  The author gave no chaining guidance. The rules below are the
+  general 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 — 12,163 downloads. [source: CivitAI stats]
+  + Positively received — 833 thumbs-up with no down-votes. [source:
+    CivitAI stats]
+  + Explicit trigger word(s) — ArsMovieStill, movie still from a 1970s
+    technicolor Sci-Fi movie, — so the effect can be turned on and off from
+    the prompt. [source: model metadata]
+  + Compact at 218 MB — cheap to stack with other LoRAs. [source: file
+    size]
+
+───────────────────────────────────────────────────────────────────────────────
+  CONS & CAVEATS
+───────────────────────────────────────────────────────────────────────────────
+  - Author never states which generation mode it was trained for; test
+    against your own workflow before relying on it. [source: description]
+
+───────────────────────────────────────────────────────────────────────────────
+  COMPATIBILITY WARNING
+───────────────────────────────────────────────────────────────────────────────
+  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
+───────────────────────────────────────────────────────────────────────────────
+  MJ7 Update: A new version recreated using MidJourney V7.
+  It is more "true" to the era/movie style, but to achieve this I have almost
+  removed my "signiture look and feel" that is present in V1.
+  This LoRA aesthetic is aimed at the early 70s Sci-fi movies .
+  The trigger words are ArsMovieStill, movie still from a 1970s technicolor
+  Sci-Fi movie
+
+─────────────────────────────────���─────────────────────────────────────────────
+  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, 1024x1024)
+  02_image.jpeg  (image, 1024x1024)
+  03_image.jpeg  (image, 1080x1445)
+  04_image.jpeg  (image, 1024x1024)
+  05_image.jpeg  (image, 1024x1024)
+  06_image.jpeg  (image, 1024x1024)
+
+───────────────────────────────────────────────────────────────────────────────
+  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 00:13 UTC.
+  Stats and description are the author's; pros/cons are derived from the
+  evidence tagged beside each line.
+───────────────────────────────────────────────────────────────────────────────

70s-sci-fi-movie/metadata.json

@@ -0,0 +1,139 @@
+{
+  "name": "70s Sci-Fi Movie",
+  "slug": "70s-sci-fi-movie",
+  "base_model": "Krea 2",
+  "media_type": "image",
+  "model_id": 995002,
+  "version_id": 3129771,
+  "version_name": "Krea2",
+  "source": "civitai",
+  "url": "https://civitai.com/models/995002",
+  "repo_id": null,
+  "file_path": null,
+  "author": "ArsMachina",
+  "category": {
+    "key": "style",
+    "label": "Visual style"
+  },
+  "modalities": [],
+  "primary_modality": "unspecified",
+  "checkpoint_family": "unknown",
+  "trigger_words": [
+    "ArsMovieStill, movie still from a 1970s technicolor Sci-Fi movie,"
+  ],
+  "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 — 12,163 downloads. [source: CivitAI stats]",
+    "Positively received — 833 thumbs-up with no down-votes. [source: CivitAI stats]",
+    "Explicit trigger word(s) — ArsMovieStill, movie still from a 1970s technicolor Sci-Fi movie, — so the effect can be turned on and off from the prompt. [source: model metadata]",
+    "Compact at 218 MB — cheap to stack with other LoRAs. [source: file size]"
+  ],
+  "cons": [
+    "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]"
+  ],
+  "file": {
+    "name": "70sSCiFiKrea2Raw.safetensors",
+    "size_bytes": 228587728,
+    "size_mb": 218.0,
+    "sha256": "8d0cfed436f4dd838238779adc9e1da910021885a8ca402f7a4db169cb185c3c"
+  },
+  "stats": {
+    "downloads": 12163,
+    "thumbs_up": 833,
+    "thumbs_down": 0
+  },
+  "popularity": {
+    "downloads": 12163,
+    "thumbs_up": 833,
+    "thumbs_down": 0,
+    "comments": 0,
+    "approval": "100% positive (833/833)",
+    "age_days": 36.6,
+    "downloads_per_day": 23.7,
+    "rank_by_downloads": 44,
+    "rank_by_momentum": 88,
+    "of_total": 150,
+    "adoption_tier": "Moderate",
+    "momentum": "Long tail · fresh release"
+  },
+  "selection_score": 22559.0,
+  "selection_reasons": [
+    "12,163 downloads",
+    "833 thumbs-up",
+    "documented triggers + usage notes",
+    "top-ranked for its category"
+  ],
+  "example_media": [
+    {
+      "file": "01_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1024",
+      "prompt": ""
+    },
+    {
+      "file": "02_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1024",
+      "prompt": ""
+    },
+    {
+      "file": "03_image.jpeg",
+      "type": "image",
+      "resolution": "1080x1445",
+      "prompt": ""
+    },
+    {
+      "file": "04_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1024",
+      "prompt": ""
+    },
+    {
+      "file": "05_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1024",
+      "prompt": ""
+    },
+    {
+      "file": "06_image.jpeg",
+      "type": "image",
+      "resolution": "1024x1024",
+      "prompt": ""
+    }
+  ],
+  "tags": [
+    "midjourney",
+    "70s",
+    "movie",
+    "style",
+    "sci-fi",
+    "midjourney v6",
+    "arsmjstyle",
+    "still"
+  ],
+  "scraped_at": "2026-09-05T00:13:32+00:00",
+  "path": "krea-2/70s-sci-fi-movie"
+}

anna-yamada-and-other-the-dangers-in-my-heart-characters/Anna_Yamada__The_Dangers_in_My_Heart__-_Krea_2_epoch_10.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72583cbecec8316c8bec130f61b43ea577d6e33360e6623bb9015eb5f0db3d3e
+size 228588904

anna-yamada-and-other-the-dangers-in-my-heart-characters/info.txt

@@ -0,0 +1,350 @@
+═══════════════════════════════════════════════════════════════════════════════
+  ANNA YAMADA AND OTHER THE DANGERS IN MY HEART CHARACTERS (ILLUSTRIOUS, KREA
+  2, PONY)
+═══════════════════════════════════════════════════════════════════════════════
+
+───────────────────────────────────────────────────────────────────────────────
+  IDENTITY
+───────────────────────────────────────────────────────────────────────────────
+  Base model        Krea 2
+  Generation mode   not declared by the author
+  Checkpoint        not determined — verify before use
+  Category          Character / subject
+  Author            MetalchromeX
+  Source            https://civitai.com/models/630682
+  Provenance        CivitAI · model 630682 · version 3192556
+
+───────────────────────────────────────────────────────────────────────────────
+  WEIGHTS
+───────────────────────────────────────────────────────────────────────────────
+  File              Anna_Yamada__The_Dangers_in_My_Heart__-_Krea_2_epoch_10.safetensors
+  Version           Anna Yamada (Krea 2)
+  Size              218.0 MB
+  Format            SafeTensor / unknown precision
+  SHA256            72583cbecec8316c8bec130f61b43ea577d6e33360e6623bb9015eb5f0db3d3e
+  Published         2026-08-02
+
+───────────────────────────────────────────────────────────────────────────────
+  TRIGGER WORDS
+───────────────────────────────────────────────────────────────────────────────
+  › "&nn&y&m&d&, 1girl, solo, long hair, black hair, brown eyes, mole on
+    neck, multiple moles on neck, bangs, large breasts"
+  › "Anna 4 gym outfit"
+  › "Anna bear hoodie outfit"
+  › "Anna black turtleneck with skirt outfit"
+  › "Anna blue butter outfit"
+  › "Anna blue tanktop outfit"
+  › "Anna brown school uniform"
+  › "Anna cooking class outfit"
+  › "Anna ghost outfit"
+  › "Anna gym uniform outfit"
+  › "Anna gym uniform with blue shirt outfit"
+  › "Anna karate outfit"
+  › "Anna orange tanktop outfit"
+  › "Anna pink dress"
+  › "Anna pink dress outfit"
+  › "Anna rain hoodie outfit"
+  › "Anna red dress outfit"
+  › "Anna red dress turtleneck outfit"
+  › "Anna scarf outfit"
+  › "Anna teal tanktop outfit"
+  › "Anna TV show sailor outfit"
+  › "Anna twintail with choker outfit"
+  › "Anna white floral dress outfit"
+  › "Anna white hoodie with red hood outfit"
+  › "Anna white tanktop outfit, sports bra, glasses, cleavage, white sports
+    bra, off shoulder, jacket, black jacket"
+  › "Anna white turtleneck with plaid dress outfit"
+  › "Anna school uniform outfit, school uniform, skirt, black socks, socks,
+    shirt, cardigan, jacket, pleated skirt, white collar shirt, blue skirt"
+
+───────────────────────────────────────────────────────────────────────────────
+  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               11,881
+  Rating                  100% positive (781/781)
+  Comments                0
+  Published               33 days ago
+  Download rate           5.4/day since release
+
+  Adoption rank           #48 of 150 in this collection  (Moderate)
+  Momentum rank           #147 of 150 by download rate  (Long tail · fresh release)
+
+  Adoption tier is the percentile of total downloads within this collection;
+  momentum is the percentile of downloads-per-day since release. Momentum is
+  ranked rather than measured against a fixed rate, because this base model
+  is itself new and every LoRA looks fast on an absolute scale. Both numbers
+  are CivitAI's, read on the scrape date at the foot of this file.
+
+───────────────────────────────────────────────────────────────────────────────
+  CHAINING / STACKING
+───────────────────────────────────────────────────────────────────────────────
+  Stacks with others      not stated
+
+  The author gave no chaining guidance. The rules below are the
+  general 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 — 11,881 downloads. [source: CivitAI stats]
+  + Positively received — 781 thumbs-up with no down-votes. [source:
+    CivitAI stats]
+  + Explicit trigger word(s) — &nn&y&m&d&, 1girl, solo, long hair, black
+    hair, brown eyes, mole on neck, multiple moles on neck, bangs, large
+    breasts, Anna 4 gym outfit, Anna bear hoodie outfit, Anna black
+    turtleneck with skirt outfit, Anna blue butter outfit, Anna blue
+    tanktop outfit, Anna brown school uniform, Anna cooking class outfit,
+    Anna ghost outfit, Anna gym uniform outfit, Anna gym uniform with blue
+    shirt outfit, Anna karate outfit, Anna orange tanktop outfit, Anna pink
+    dress, Anna pink dress outfit, Anna rain hoodie outfit, Anna red dress
+    outfit, Anna red dress turtleneck outfit, Anna scarf outfit, Anna teal
+    tanktop outfit, Anna TV show sailor outfit, Anna twintail with choker
+    outfit, Anna white floral dress outfit, Anna white hoodie with red hood
+    outfit, Anna white tanktop outfit, sports bra, glasses, cleavage, white
+    sports bra, off shoulder, jacket, black jacket, Anna white turtleneck
+    with plaid dress outfit, Anna school uniform outfit, school uniform,
+    skirt, black socks, socks, shirt, cardigan, jacket, pleated skirt,
+    white collar shirt, blue skirt — so the effect can be turned on and off
+    from the prompt. [source: model metadata]
+  + Compact at 218 MB — cheap to stack with other LoRAs. [source: file
+    size]
+
+───────────────────────────────────────────────────────────────────────────────
+  CONS & CAVEATS
+───────────────────────────────────────────────────────────────────────────────
+  - Author never states which generation mode it was trained for; test
+    against your own workflow before relying on it. [source: description]
+
+───────────────────────────────────────────────────────────────────────────────
+  COMPATIBILITY WARNING
+──────────────────────────────────────────��────────────────────────────────────
+  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
+───────────────────────────────────────────────────────────────────────────────
+  ➼Like what I do? Consider throwing me a few dollars at https://ko-
+  fi.com/dandysombraes or commissioning me
+  https://www.fiverr.com/metalchromex/make-you-a-stable-diffusion-anime-
+  character-model so I can continue making cool crap for you all to enjoy for
+  free!
+  Update log: 8-2-26: I added a Krea 2 LoRA of Anna.
+  7-4-26: Happy 4th of July to my fellow Americans! I added a V3 Sanae Yamada
+  LoRA. V3 is similar to my V2 version, but has tweaked tags, and I removed
+  some 'eh' images from the dataset.
+  1-9-26: Added a V2 Anna Yamada LoRA. V2 has tweaked tags, I removed some
+  'eh' images from the dataset, and made other tweaks. V1 should outperform
+  the original version in most ways.
+  11-14-25: Added a Honoka Hara LoRA.
+  11-13-25: Added a Nico Kouda LoRA.
+  7-24-25: Added a version of Sanae with a lower step count (640 steps). This
+  might improve onsite generations.
+  6-25-25: I added a third version of Anna Yamada with a very low step count
+  (1,638 between 655 images). I also removed a small handful of 'eh' images.
+  This should, in theory, help with the on-site trainer and pairing with
+  other style LoRA.
+  6-7-25: Added a Chihiro Kobayashi Illustrious LoRA.
+  6-5-25: Added a Yurine Hanzawa Illustrious LoRA. Despite her limited
+  appearance toward the end of season 2, she turned out pretty well!
+  3-6-25: Added an improved 2nd version of my Sanae Yamada model with a
+  slightly better/expanded dataset and better training settings.
+  2-26-25: Added the new Kana Ichikawa Illustrious model and re-posted the
+  old Pony version of Kana; unpublished the original Pony listing.
+  2-16-25: Added a Moeko Sekine Illustrious LoRA.
+  2-14-25: Added a Moeko Sekine Pony LoRA and unpublished the old listing; an
+  Illustrious remake will be coming soon!
+  2-13-25: Renamed the page The Dangers in My Heart characters by
+  Metalchromex. Added the Sanae Yamada Illustrious LoRA and the Anna Yamada
+  V2 Illustrious LoRA (V2 makes V1 irrelevant) and unpublished the previous
+  versions.
+
+───────────────────────────────────────────────────────────────────────────────
+  EXAMPLE MEDIA
+───────────────────────────────────────────────────────────────────────────────
+  5 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, 1200x1920)
+  02_image.jpeg  (image, 1200x1920)
+  03_image.jpeg  (image, 1200x1920)
+  04_image.jpeg  (image, 2160x2864)
+  05_image.jpeg  (image, 2160x2864)
+
+────────────────────────────────────────────���──────────────────────────────────
+  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 00:13 UTC.
+  Stats and description are the author's; pros/cons are derived from the
+  evidence tagged beside each line.
+───────────────────────────────────────────────────────────────────────────────

anna-yamada-and-other-the-dangers-in-my-heart-characters/metadata.json

@@ -0,0 +1,156 @@
+{
+  "name": "Anna Yamada and other The Dangers in My Heart characters (Illustrious, Krea 2, Pony)",
+  "slug": "anna-yamada-and-other-the-dangers-in-my-heart-characters",
+  "base_model": "Krea 2",
+  "media_type": "image",
+  "model_id": 630682,
+  "version_id": 3192556,
+  "version_name": "Anna Yamada (Krea 2)",
+  "source": "civitai",
+  "url": "https://civitai.com/models/630682",
+  "repo_id": null,
+  "file_path": null,
+  "author": "MetalchromeX",
+  "category": {
+    "key": "character",
+    "label": "Character / subject"
+  },
+  "modalities": [],
+  "primary_modality": "unspecified",
+  "checkpoint_family": "unknown",
+  "trigger_words": [
+    "&nn&y&m&d&, 1girl, solo, long hair, black hair, brown eyes, mole on neck, multiple moles on neck, bangs, large breasts",
+    "Anna 4 gym outfit",
+    "Anna bear hoodie outfit",
+    "Anna black turtleneck with skirt outfit",
+    "Anna blue butter outfit",
+    "Anna blue tanktop outfit",
+    "Anna brown school uniform",
+    "Anna cooking class outfit",
+    "Anna ghost outfit",
+    "Anna gym uniform outfit",
+    "Anna gym uniform with blue shirt outfit",
+    "Anna karate outfit",
+    "Anna orange tanktop outfit",
+    "Anna pink dress",
+    "Anna pink dress outfit",
+    "Anna rain hoodie outfit",
+    "Anna red dress outfit",
+    "Anna red dress turtleneck outfit",
+    "Anna scarf outfit",
+    "Anna teal tanktop outfit",
+    "Anna TV show sailor outfit",
+    "Anna twintail with choker outfit",
+    "Anna white floral dress outfit",
+    "Anna white hoodie with red hood outfit",
+    "Anna white tanktop outfit, sports bra, glasses, cleavage, white sports bra, off shoulder, jacket, black jacket",
+    "Anna white turtleneck with plaid dress outfit",
+    "Anna school uniform outfit, school uniform, skirt, black socks, socks, shirt, cardigan, jacket, pleated skirt, white collar shirt, blue skirt"
+  ],
+  "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 — 11,881 downloads. [source: CivitAI stats]",
+    "Positively received — 781 thumbs-up with no down-votes. [source: CivitAI stats]",
+    "Explicit trigger word(s) — &nn&y&m&d&, 1girl, solo, long hair, black hair, brown eyes, mole on neck, multiple moles on neck, bangs, large breasts, Anna 4 gym outfit, Anna bear hoodie outfit, Anna black turtleneck with skirt outfit, Anna blue butter outfit, Anna blue tanktop outfit, Anna brown school uniform, Anna cooking class outfit, Anna ghost outfit, Anna gym uniform outfit, Anna gym uniform with blue shirt outfit, Anna karate outfit, Anna orange tanktop outfit, Anna pink dress, Anna pink dress outfit, Anna rain hoodie outfit, Anna red dress outfit, Anna red dress turtleneck outfit, Anna scarf outfit, Anna teal tanktop outfit, Anna TV show sailor outfit, Anna twintail with choker outfit, Anna white floral dress outfit, Anna white hoodie with red hood outfit, Anna white tanktop outfit, sports bra, glasses, cleavage, white sports bra, off shoulder, jacket, black jacket, Anna white turtleneck with plaid dress outfit, Anna school uniform outfit, school uniform, skirt, black socks, socks, shirt, cardigan, jacket, pleated skirt, white collar shirt, blue skirt — so the effect can be turned on and off from the prompt. [source: model metadata]",
+    "Compact at 218 MB — cheap to stack with other LoRAs. [source: file size]"
+  ],
+  "cons": [
+    "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]"
+  ],
+  "file": {
+    "name": "Anna_Yamada__The_Dangers_in_My_Heart__-_Krea_2_epoch_10.safetensors",
+    "size_bytes": 228588904,
+    "size_mb": 218.0,
+    "sha256": "72583cbecec8316c8bec130f61b43ea577d6e33360e6623bb9015eb5f0db3d3e"
+  },
+  "stats": {
+    "downloads": 11881,
+    "thumbs_up": 781,
+    "thumbs_down": 0
+  },
+  "popularity": {
+    "downloads": 11881,
+    "thumbs_up": 781,
+    "thumbs_down": 0,
+    "comments": 0,
+    "approval": "100% positive (781/781)",
+    "age_days": 33.1,
+    "downloads_per_day": 5.37,
+    "rank_by_downloads": 48,
+    "rank_by_momentum": 147,
+    "of_total": 150,
+    "adoption_tier": "Moderate",
+    "momentum": "Long tail · fresh release"
+  },
+  "selection_score": 21653.0,
+  "selection_reasons": [
+    "11,881 downloads",
+    "781 thumbs-up",
+    "documented triggers + usage notes",
+    "top-ranked for its category"
+  ],
+  "example_media": [
+    {
+      "file": "01_image.jpeg",
+      "type": "image",
+      "resolution": "1200x1920",
+      "prompt": ""
+    },
+    {
+      "file": "02_image.jpeg",
+      "type": "image",
+      "resolution": "1200x1920",
+      "prompt": ""
+    },
+    {
+      "file": "03_image.jpeg",
+      "type": "image",
+      "resolution": "1200x1920",
+      "prompt": ""
+    },
+    {
+      "file": "04_image.jpeg",
+      "type": "image",
+      "resolution": "2160x2864",
+      "prompt": ""
+    },
+    {
+      "file": "05_image.jpeg",
+      "type": "image",
+      "resolution": "2160x2864",
+      "prompt": ""
+    }
+  ],
+  "tags": [
+    "anna yamada",
+    "character",
+    "sanae yamada",
+    "moeko sekine",
+    "the dangers in my heart"
+  ],
+  "scraped_at": "2026-09-05T00:13:40+00:00",
+  "path": "krea-2/anna-yamada-and-other-the-dangers-in-my-heart-characters"
+}

ars-midjourney-watercolor/WaterColorV2Krea2Raw.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbcc2cc8cf265b18a8ed5af240bebfa1db7b2635bb21d40750d01e815544eccd
+size 228587736

ars-midjourney-watercolor/info.txt

@@ -0,0 +1,299 @@
+═══════════════════════════════════════════════════════════════════════════════
+  ARS MIDJOURNEY WATERCOLOR
+═══════════════════════════════════════════════════════════════════════════════
+
+───────────────────────────────────────────────────────────────────────────────
+  IDENTITY
+───────────────────────────────────────────────────────────────────────────────
+  Base model        Krea 2
+  Generation mode   not declared by the author
+  Checkpoint        not determined — verify before use
+  Category          Visual style
+  Author            ArsMachina
+  Source            https://civitai.com/models/551903
+  Provenance        CivitAI · model 551903 · version 3131742
+
+───────────────────────────────────────────────────────────────────────────────
+  WEIGHTS
+───────────────────────────────────────────────────────────────────────────────
+  File              WaterColorV2Krea2Raw.safetensors
+  Version           Krea2
+  Size              218.0 MB
+  Format            SafeTensor / unknown precision
+  SHA256            cbcc2cc8cf265b18a8ed5af240bebfa1db7b2635bb21d40750d01e815544eccd
+  Published         2026-07-30
+
+───────────────────────────────────────────────────────────────────────────────
+  TRIGGER WORDS
+───────────────────────────────────────────────────────────────────────────────
+  › "ArsMJStyle, Watercolor,"
+
+───────────────────────────────────────────────────────────────────────────────
+  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               13,719
+  Rating                  100% positive (931/931)
+  Comments                0
+  Published               36 days ago
+  Download rate           12.1/day since release
+
+  Adoption rank           #41 of 150 in this collection  (Moderate)
+  Momentum rank           #128 of 150 by download rate  (Long tail · fresh release)
+
+  Adoption tier is the percentile of total downloads within this collection;
+  momentum is the percentile of downloads-per-day since release. Momentum is
+  ranked rather than measured against a fixed rate, because this base model
+  is itself new and every LoRA looks fast on an absolute scale. Both numbers
+  are CivitAI's, read on the scrape date at the foot of this file.
+
+───────────────────────────────────────────────────────────────────────────────
+  CHAINING / STACKING
+───────────────────────────────────────────────────────────────────────────────
+  Stacks with others      not stated
+
+  The author gave no chaining guidance. The rules below are the
+  general 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 — 13,719 downloads. [source: CivitAI stats]
+  + Positively received — 931 thumbs-up with no down-votes. [source:
+    CivitAI stats]
+  + Explicit trigger word(s) — ArsMJStyle, Watercolor, — so the effect can
+    be turned on and off from the prompt. [source: model metadata]
+  + Compact at 218 MB — cheap to stack with other LoRAs. [source: file
+    size]
+
+───────────────────────────────────────────────────────────────────────────────
+  CONS & CAVEATS
+───────────────────────────────────────────────────────────────────────────────
+  - Author never states which generation mode it was trained for; test
+    against your own workflow before relying on it. [source: description]
+
+───────────────────────────────────────────────────────────────────────────────
+  COMPATIBILITY WARNING
+───────────────────────────────────────────────────────────────────────────────
+  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
+───────────────────────────────────────────────────────────────────────────────
+  MJ7 VERSION Update: The dataset was rebuilt entirely from scratch using
+  MidJourney V7.
+  The main difference from v1 and v2 is a shift towards more natural colors
+  and textures.
+
+  The Lora is trained on MidJourney Images that use my "Personalised Style" +
+  Watercolor What is new in v2:
+  - Significantly increased the training dataset
+
+  - Added support for a wider range of concepts
+
+  - Removed most of the biases that were present in v1 (closed eyes
+  everywhere for example)
+
+  - Added 3 different intensity of the colors ( intense , deep , soft colors
+  )
+
+  - My personal style is changed a little bit , so the images are more "crisp
+  and clear"
+
+  Works without the trigger words ArsMJStyle, Watercolor
+  Works great in the 0.2 - 1.5 range.
+  Start ~1.
+  As Watercolor is strong existing style and looks really nice as an abstract
+  painting, you can go pretty high with the strength and still get a nice
+  result.
+  I hope you will enjoy this LoRA as much as I did creating it!
+
+───────────────────────────────────────────────────────────────────────────────
+  EXAMPLE MEDIA
+───────────────────────────────────────────────────────────────────────────────
+  6 preview(s) in ./example_images/ — community results for
+  this LoRA. Videos keep a matching _poster.jpg still frame.
+
+  No prompts are recorded below: CivitAI's public API flags these posts as
+  having generation metadata but does not return it, so the prompts cannot be
+  scraped. Open the model page to read them.
+
+  01_image.jpeg  (image, 2376x3488)
+  02_image.jpeg  (image, 2376x3488)
+  03_image.jpeg  (image, 2376x3488)
+  04_image.jpeg  (image, 2376x3488)
+  05_image.jpeg  (image, 2376x3488)
+  06_image.jpeg  (image, 2376x3488)
+
+───────────────────────────────────────────────────────────────────────────────
+  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 00:13 UTC.
+  Stats and description are the author's; pros/cons are derived from the
+  evidence tagged beside each line.
+───────────────────────────────────────────────────────────────────────────────

ars-midjourney-watercolor/metadata.json

@@ -0,0 +1,136 @@
+{
+  "name": "Ars MidJourney Watercolor",
+  "slug": "ars-midjourney-watercolor",
+  "base_model": "Krea 2",
+  "media_type": "image",
+  "model_id": 551903,
+  "version_id": 3131742,
+  "version_name": "Krea2",
+  "source": "civitai",
+  "url": "https://civitai.com/models/551903",
+  "repo_id": null,
+  "file_path": null,
+  "author": "ArsMachina",
+  "category": {
+    "key": "style",
+    "label": "Visual style"
+  },
+  "modalities": [],
+  "primary_modality": "unspecified",
+  "checkpoint_family": "unknown",
+  "trigger_words": [
+    "ArsMJStyle, Watercolor,"
+  ],
+  "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 — 13,719 downloads. [source: CivitAI stats]",
+    "Positively received — 931 thumbs-up with no down-votes. [source: CivitAI stats]",
+    "Explicit trigger word(s) — ArsMJStyle, Watercolor, — so the effect can be turned on and off from the prompt. [source: model metadata]",
+    "Compact at 218 MB — cheap to stack with other LoRAs. [source: file size]"
+  ],
+  "cons": [
+    "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]"
+  ],
+  "file": {
+    "name": "WaterColorV2Krea2Raw.safetensors",
+    "size_bytes": 228587736,
+    "size_mb": 218.0,
+    "sha256": "cbcc2cc8cf265b18a8ed5af240bebfa1db7b2635bb21d40750d01e815544eccd"
+  },
+  "stats": {
+    "downloads": 13719,
+    "thumbs_up": 931,
+    "thumbs_down": 0
+  },
+  "popularity": {
+    "downloads": 13719,
+    "thumbs_up": 931,
+    "thumbs_down": 0,
+    "comments": 0,
+    "approval": "100% positive (931/931)",
+    "age_days": 36.5,
+    "downloads_per_day": 12.14,
+    "rank_by_downloads": 41,
+    "rank_by_momentum": 128,
+    "of_total": 150,
+    "adoption_tier": "Moderate",
+    "momentum": "Long tail · fresh release"
+  },
+  "selection_score": 25291.0,
+  "selection_reasons": [
+    "13,719 downloads",
+    "931 thumbs-up",
+    "documented triggers + usage notes",
+    "top-ranked for its category"
+  ],
+  "example_media": [
+    {
+      "file": "01_image.jpeg",
+      "type": "image",
+      "resolution": "2376x3488",
+      "prompt": ""
+    },
+    {
+      "file": "02_image.jpeg",
+      "type": "image",
+      "resolution": "2376x3488",
+      "prompt": ""
+    },
+    {
+      "file": "03_image.jpeg",
+      "type": "image",
+      "resolution": "2376x3488",
+      "prompt": ""
+    },
+    {
+      "file": "04_image.jpeg",
+      "type": "image",
+      "resolution": "2376x3488",
+      "prompt": ""
+    },
+    {
+      "file": "05_image.jpeg",
+      "type": "image",
+      "resolution": "2376x3488",
+      "prompt": ""
+    },
+    {
+      "file": "06_image.jpeg",
+      "type": "image",
+      "resolution": "2376x3488",
+      "prompt": ""
+    }
+  ],
+  "tags": [
+    "midjourney",
+    "arsmjstyle",
+    "midjourney v6",
+    "watercolor",
+    "style"
+  ],
+  "scraped_at": "2026-09-05T00:13:13+00:00",
+  "path": "krea-2/ars-midjourney-watercolor"
+}

body-weight-slider-skinny-fat-slider/Krea2Weight_v3.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19d9b17af57918e4168c77dd279c2aa0d87612eb55a0b516f7e7838e25e3a7bc
+size 107435704

body-weight-slider-skinny-fat-slider/info.txt

@@ -0,0 +1,288 @@
+═══════════════════════════════════════════════════════════════════════════════
+  BODY WEIGHT SLIDER. (SKINNY ===> FAT SLIDER)
+═══════════════════════════════════════════════════════════════════════════════
+
+───────────────────────────────────────────────────────────────────────────────
+  IDENTITY
+───────────────────────────────────────────────────────────────────────────────
+  Base model        Krea 2
+  Generation mode   not declared by the author
+  Checkpoint        not determined — verify before use
+  Category          Detail / anatomy fixer
+  Author            Maruhashi
+  Source            https://civitai.com/models/2513759
+  Provenance        CivitAI · model 2513759 · version 3129180
+
+───────────────────────────────────────────────────────────────────────────────
+  WEIGHTS
+───────────────────────────────────────────────────────────────────────────────
+  File              Krea2Weight_v3.safetensors
+  Version           Krea2
+  Size              102.5 MB
+  Format            SafeTensor / unknown precision
+  SHA256            19d9b17af57918e4168c77dd279c2aa0d87612eb55a0b516f7e7838e25e3a7bc
+  Published         2026-07-26
+
+───────────────────────────────────────────────────────────────────────────────
+  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               11,267
+  Rating                  100% positive (855/856)
+  Comments                13
+  Published               41 days ago
+  Download rate           85.9/day since release
+
+  Adoption rank           #50 of 150 in this collection  (Moderate)
+  Momentum rank           #33 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 — 11,267 downloads. [source: CivitAI stats]
+  + Positively received — 855 thumbs-up, 1 down. [source: CivitAI stats]
+  + Compact at 102 MB — cheap to stack with other LoRAs. [source: file
+    size]
+
+───────────────────────────────────────────────────────────────────────────────
+  CONS & CAVEATS
+───────────────────────────────────────────────────────────────────────────────
+  - No trigger word — the LoRA is always on once loaded, so strength is the
+    only control. [source: model metadata]
+  - 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
+───────────────────────────────────────────────────────────────────────────────
+  Recommended negative prompt: lowres, bad quality, worst quality, bad
+  anatomy, jpeg artifacts, signature, watermark, sketch
+
+  I also recommend using this upscaler:
+  https://openmodeldb.info/models/2x-AnimeSharpV4
+
+  Usable weight: from -3 to 3+, depending on the checkpoint and additional
+  LORA used
+
+  No trigger word is needed
+
+───────────────────────────────────────────────────────────────────────────────
+  VERSION NOTES
+───────────────────────────────────────────────────────────────────────────────
+  Works in wide range from -7 to 7, you can try higher weights depending on
+  checkpoint and additional lora used
+
+───────────────────────────────────────────────────────────────────────────────
+  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, 1664x2828)
+  02_image.jpeg  (image, 1440x1920)
+  03_image.jpeg  (image, 1440x1920)
+  04_image.jpeg  (image, 1440x1920)
+  05_image.jpeg  (image, 1440x1920)
+  06_image.jpeg  (image, 1344x1920)
+
+───────────────────────────────────────────────────────────────────────────────
+  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 00:13 UTC.
+  Stats and description are the author's; pros/cons are derived from the
+  evidence tagged beside each line.
+───────────────────────────────────────────────────────────────────────────────

body-weight-slider-skinny-fat-slider/metadata.json

@@ -0,0 +1,135 @@
+{
+  "name": "Body weight slider. (Skinny ===> Fat slider)",
+  "slug": "body-weight-slider-skinny-fat-slider",
+  "base_model": "Krea 2",
+  "media_type": "image",
+  "model_id": 2513759,
+  "version_id": 3129180,
+  "version_name": "Krea2",
+  "source": "civitai",
+  "url": "https://civitai.com/models/2513759",
+  "repo_id": null,
+  "file_path": null,
+  "author": "Maruhashi",
+  "category": {
+    "key": "detail",
+    "label": "Detail / anatomy fixer"
+  },
+  "modalities": [],
+  "primary_modality": "unspecified",
+  "checkpoint_family": "unknown",
+  "trigger_words": [],
+  "recommended_settings": {},
+  "chaining": {
+    "author": {
+      "stacks": "not stated",
+      "notes": "",
+      "avoid": [],
+      "requires": [],
+      "order": "",
+      "strength": "",
+      "stated": false
+    },
+    "general_rules": [
+      "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 — 11,267 downloads. [source: CivitAI stats]",
+    "Positively received — 855 thumbs-up, 1 down. [source: CivitAI stats]",
+    "Compact at 102 MB — cheap to stack with other LoRAs. [source: file size]"
+  ],
+  "cons": [
+    "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]",
+    "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]"
+  ],
+  "file": {
+    "name": "Krea2Weight_v3.safetensors",
+    "size_bytes": 107435704,
+    "size_mb": 102.46,
+    "sha256": "19d9b17af57918e4168c77dd279c2aa0d87612eb55a0b516f7e7838e25e3a7bc"
+  },
+  "stats": {
+    "downloads": 11267,
+    "thumbs_up": 855,
+    "thumbs_down": 1
+  },
+  "popularity": {
+    "downloads": 11267,
+    "thumbs_up": 855,
+    "thumbs_down": 1,
+    "comments": 13,
+    "approval": "100% positive (855/856)",
+    "age_days": 40.7,
+    "downloads_per_day": 85.92,
+    "rank_by_downloads": 50,
+    "rank_by_momentum": 33,
+    "of_total": 150,
+    "adoption_tier": "Moderate",
+    "momentum": "Trending · fresh release"
+  },
+  "selection_score": 21527.0,
+  "selection_reasons": [
+    "11,267 downloads",
+    "855 thumbs-up",
+    "top-ranked for its category"
+  ],
+  "example_media": [
+    {
+      "file": "01_image.jpeg",
+      "type": "image",
+      "resolution": "1664x2828",
+      "prompt": ""
+    },
+    {
+      "file": "02_image.jpeg",
+      "type": "image",
+      "resolution": "1440x1920",
+      "prompt": ""
+    },
+    {
+      "file": "03_image.jpeg",
+      "type": "image",
+      "resolution": "1440x1920",
+      "prompt": ""
+    },
+    {
+      "file": "04_image.jpeg",
+      "type": "image",
+      "resolution": "1440x1920",
+      "prompt": ""
+    },
+    {
+      "file": "05_image.jpeg",
+      "type": "image",
+      "resolution": "1440x1920",
+      "prompt": ""
+    },
+    {
+      "file": "06_image.jpeg",
+      "type": "image",
+      "resolution": "1344x1920",
+      "prompt": ""
+    }
+  ],
+  "tags": [
+    "slider",
+    "concept",
+    "illustrious",
+    "fat",
+    "weight slider",
+    "body weight slider",
+    "skinny"
+  ],
+  "scraped_at": "2026-09-05T00:13:37+00:00",
+  "path": "krea-2/body-weight-slider-skinny-fat-slider"
+}

carefull-style-krea-2-anima-il/carefull_style_krea_v1.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d7dd8770f2e0038c73d6d310935493c18b248ceeadf6174f409d00c74d603d7
+size 228588904