zeste-expressions/info.txt

═══════════════════════════════════════════════════════════════════════════════
  ZESTE EXPRESSIONS
═══════════════════════════════════════════════════════════════════════════════

───────────────────────────────────────────────────────────────────────────────
  IDENTITY
───────────────────────────────────────────────────────────────────────────────
  Base model        Krea 2
  Generation mode   not declared by the author
  Checkpoint        not determined — verify before use
  Category          Speed / step-reduction
  Author            zeste
  Source            https://civitai.com/models/2821450
  Provenance        CivitAI · model 2821450 · version 3205293

───────────────────────────────────────────────────────────────────────────────
  WEIGHTS
───────────────────────────────────────────────────────────────────────────────
  File              ZesteExpressionsV26K.safetensors
  Version           v2.0
  Size              218.0 MB
  Format            SafeTensor / unknown precision
  SHA256            9059a216568d84992ec7f2e0812f677d0adaa2dae66808f5da13aefaafc70d9a
  Published         2026-08-07

───────────────────────────────────────────────────────────────────────────────
  TRIGGER WORDS
───────────────────────────────────────────────────────────────────────────────
  › "Check the Description for listed Expressions"

───────────────────────────────────────────────────────────────────────────────
  RECOMMENDED SETTINGS
───────────────────────────────────────────────────────────────────────────────
  Taken verbatim from the author's description:

  LoRA strength       1.25

───────────────────────────────────────────────────────────────────────────────
  ADOPTION & POPULARITY
───────────────────────────────────────────────────────────────────────────────
  Downloads               1,138
  Rating                  100% positive (86/86)
  Comments                0
  Published               29 days ago
  Download rate           23.8/day since release

  Adoption rank           #148 of 200 in this collection  (Modest)
  Momentum rank           #169 of 200 by download rate  (Long tail · fresh release)

  Adoption tier is the percentile of total downloads within this collection;
  momentum is the percentile of downloads-per-day since release. Momentum is
  ranked rather than measured against a fixed rate, because this base model
  is itself new and every LoRA looks fast on an absolute scale. Both numbers
  are CivitAI's, read on the scrape date at the foot of this file.

───────────────────────────────────────────────────────────────────────────────
  CHAINING / STACKING
───────────────────────────────────────────────────────────────────────────────
  Stacks with others      not stated

  The author gave no chaining guidance. The rules below are the
  general 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.
  • Krea 2 Turbo is itself the official step-reduction artefact - the
    technical report describes a final timestep-distillation stage using
    TDM (Trajectory Distribution Matching), chosen for being "simple to
    tune, with minimal hyperparameters", producing an 8-step CFG-free
    checkpoint from Raw. It is a full checkpoint, not a LoRA. Three LoRA-
    shaped accelerators exist around it: 1. Raw-to-Turbo delta adapters.
    Comfy-Org/Krea-2 ships loras/krea2_turbo_lora_rank_64_bf16.safetensors
    - I read its header directly: 535 tensors, rank 64, ComfyUI-native
    diffusion_model.*.lora_down/lora_up keys, no __metadata
  • 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
───────────────────────────────────────────────────────────────────────────────
  + Well established — 1,138 downloads. [source: CivitAI stats]
  + Positively received — 86 thumbs-up with no down-votes. [source: CivitAI
    stats]
  + Explicit trigger word(s) — Check the Description for listed Expressions
    — 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]
  + Author published concrete recommended settings. [source: description]

───────────────────────────────────────────────────────────────────────────────
  CONS & CAVEATS
───────────────────────────────────────────────────────────────────────────────
  - Author never states which generation mode it was trained for; test
    against your own workflow before relying on it. [source: description]

───────────────────────────────────────────────────────────────────────────────
  COMPATIBILITY WARNING
───────────────────────────────────────────────────────────────────────────────
  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
───────────────────────────────────────────────────────────────────────────────
  v2.0

  Trained on over 6000 images featuring 73 characters across 50 Anime Series
  / Art Styles with over 750 additional realistic images. Trained on base
  model. Raise strength as needed. Use the list of expressions I provided and
  modify from there. Feel free to come up with your own expressions and it
  should perform well. Trained on over 70 expressions.

  Strength from 0.5 to 1.5. A few expressions will be refused by krea2s
  safety filter and you have to bump the strength closer to 1.5 or pair this
  lora with a refusal reduction lora or simply use an uncensored checkpoint.
  Certain art styles or toon characters with odd facial proportions might not
  react as well as realistic or anime character with more traditional faces.
  If you're running into issues, please post in the comments or DM me and
  I'll work on improving them for the next version.

  Example descriptions to get you started!

  Adapt these descriptions to fit your needs if the face is not reflecting
  the expression you want.

  Ahegao: The head tips back as the eyes roll upward and lose focus beneath
  fluttering lids, the cheeks burn deep red, and the mouth hangs open with
  the tongue lolling out in overwhelmed bliss.
  Angry: The brows slam down and inward, carving a deep vertical crease
  between them, while the eyes narrow into a hard stare, the nostrils flare,
  and the lips press tight over a clenched jaw.
  Annoyed: The eyelids lower halfway over a flat sidelong stare, one brow
  gives a small irritated twitch, and the lips press into a thin line as a
  short breath escapes the nose.
  Clenched teeth: The teeth are visible and clenched. The brows are furrowed.
  Blush: Heavy red blush on cheeks blooms across the cheeks and the bridge of
  the nose, deepening toward the ears, while the rest of the expression stays
  soft and the eyes drift shyly aside.
  Embarrassed: Heavy red blush on cheeks as the gaze drops and darts away,
  the inner brows tilt helplessly upward, and the mouth wobbles between a
  sheepish smile and a bitten lip.
  Bored: The eyelids sag to half-mast over an unfocused stare at nothing, the
  mouth hangs slack and faintly downturned, and every feature reads as
  drained of interest rather than sad.
  Closed eyes: The lids rest gently shut with the lashes lying flat, the brow
  smooth and untroubled, giving the whole face a look of calm or quiet
  concentration.
  Confused: One brow climbs while the other dips, the eyes wander off-center
  searching for an answer, the head tilts, and the lips purse into a small
  puzzled knot.
  Crazy: The eyes stretch wide around tiny pinprick pupils, dark shadows ring
  the sockets, and a trembling, too-wide grin bares more teeth than any sane
  smile should.
  Drunk: The eyelids droop unevenly over a hazy, wandering gaze, a lazy blush
  spreads across the nose and cheeks, and the mouth hangs in a loose, swaying
  half-smile.
  Expressionless: The face is perfectly neutral - level brows, calm
  unblinking eyes, relaxed cheeks, and a mouth drawn as a single flat line
  giving nothing away.
  Evil: The brows dip low over narrowed eyes that catch a sharp gleam, a
  shadow seems to fall across the upper face, and the lips curl into a thin,
  knowing, asymmetrical smile.
  Facepalm: Beneath the hand pressed to the forehead, the eyes squeeze shut,
  the brows knot together, and the mouth twists into a grimace of pure
  exasperation.
  Glaring: The eyes narrow to sharp slits locked dead on their target, the
  pupils contract to points, the brows press down flat, and the mouth sets
  into a hard, unforgiving line.
  Happy: The cheeks lift high, pushing the eyes into warm crinkled crescents,
  while the mouth curves into a bright, easy smile that reaches every corner
  of the face.
  Laughing: The eyes squeeze into upturned arcs, the brows float up relaxed,
  the cheeks round out, and the mouth opens wide showing the upper teeth as
  joy takes over the whole face.
  Naughty face: The eyes glance up through the lashes with a wicked sparkle,
  the head tilts playfully, and a small biting smirk tugs at one corner of
  the mouth.
  Nosebleed: The eyes go wide and stunned, a fierce blush burns across the
  whole face, and a thin trickle of crimson runs from one nostril toward the
  lip.
  Open mouth: The jaw simply drops, the lips form a soft relaxed oval, and a
  clear gap shows between the upper and lower teeth.
  Parted lips: The mouth is very slightly open, soft and unhurried, with
  relaxed eyelids lending the face a look of longing or a caught breath.
  Pout: The lower lip pushes forward and up over the pressed upper lip, the
  chin bunches, the brows knit, and the eyes cut upward in sulky protest.
  Raised eyebrow: A single brow arches high, opening that eye wider while the
  other stays level, paired with a flat skeptical mouth or the faintest
  smirk.
  Rolling eyes: The eyes are rolling back, the lids drop halfway in
  exasperation, and the mouth goes slack with a silent, long-suffering sigh.
  Sad: The inner corners of the brows draw upward into a gentle slant, the
  glossy eyes look down, and the corners of the mouth sink as the chin
  trembles faintly.
  Frown: The brows knit firmly together and the mouth pulls into a decisive
  downward curve, the eyes steady and dry - displeasure or hard concentration
  rather than grief.
  Gloom: A dark shadow seems to hang over the brow, the eyes go dull and
  hollow staring at nothing, and the mouth sits in a faint, defeated downturn
  as a heavy aura settles on the face.
  Tears: The eyes glass over as droplets well along the lower lids, the
  lashes clump wet, the rims redden, and shining trails slide down both
  cheeks.
  Scared: The brows lift and pull together, wrinkling the forehead, while the
  upper lids rise to show white above the shrunken pupils and the lips
  stretch thin and trembling.
  Horror: The eyes strain open to their absolute widest around pinprick
  pupils, the skin drains pale with a blue shadow across the brow, and the
  dropped jaw quivers in a soundless gasp.
  Scream: The mouth tears open into its widest oval with the tongue and teeth
  visible, the brows arch high and strained, and every muscle of the face
  pulls taut.
  Surprised: The brows shoot straight up, the eyes pop into perfect circles
  with small floating irises, and the lips form a sudden little "o" -
  startled but not yet afraid.
  Serious: The brows level out and press slightly down, the gaze holds steady
  and unblinking, and the lips seal into a firm straight line over a set jaw.
  Sigh: The eyes slide half-closed, the brows tilt up in tired release, the
  mouth is very slightly open.
  Sleepy: The heavy eyelids flutter and sag over an unfocused gaze.
  Smirk: One corner of the mouth hooks upward while the other stays put, the
  eyes hold a level, knowing look, and a single brow lifts just enough to
  signal confidence.
  Smug: The chin tips up and the eyes settle into half-closed, serene
  satisfaction, while a small closed-lip smile of quiet superiority plays on
  the mouth.
  Sulking: The cheeks puff slightly, the gaze turns pointedly away, the lower
  lip juts out, and the pinched brows complete a look of childish, wounded
  stubbornness.
  Thinking: The eyes drift up and to the side in search of an idea, the brows
  draw together in focus, and the lips press thoughtfully - often above a
  hand at the chin.
  Troll face: An impish toothy grin stretches from cheek to cheek, the brows
  arch in gleeful mischief, and the hooded eyes shine with an unmistakable
  "gotcha."
  Wink: One eye snaps crisply shut with a smooth lid while the other stays
  bright and engaged, paired with a tilted, in-on-the-joke smile.
  Wince: The eyes screw shut, the nose scrunches, and the teeth grit behind a
  pulled-back grimace as the head flinches away from sudden pain or
  awkwardness.
  Grimace: The lips stretch wide and thin to bare the teeth in a strained
  rectangle, the nose wrinkles, and the knotted brows finish a look of
  disgust or hard effort.
  Crazy smile: A beaming grin spreads far too wide for the face, showing
  every tooth, beneath enormous shimmering eyes that refuse to blink.
  Evil smile: The lips curve slowly into a closed, predatory arc, the
  narrowed eyes glint beneath a shadowed brow, and the whole face radiates
  hidden intent.
  Fingersmile (Two-handed): Both index fingers press into the corners of the
  mouth and push them up into a hand-made grin, dimpling the cheeks, while
  the wide eyes lock onto the viewer.
  Forced smile: The mouth curves dutifully upward but the eyes stay flat and
  tired, tension pinches the corners of the lips, and the strained brows
  betray the act.
  Glasgow smile: The grin stretches unnaturally far toward the ears, well
  past what a mouth should manage, its impossible width made more unsettling
  by calm, unbothered eyes.
  Grin: A broad, open smile flashes the upper teeth, the cheeks round out
  rosy and lifted, and the sparkling eyes join in without reservation.
  Evil grin: The wide grin bares its teeth - sharp canines catching the light
  - while the eyes narrow in triumph and the chin dips to throw the face into
  menacing shadow.
  Light smile: The lips stay together and lift by only a few gentle degrees
  at the corners, matched by soft, calm eyes - contentment at its quietest.
  Sad smile: The mouth manages a small upward curve while the inner brows
  tilt up in sorrow and the glistening eyes threaten tears - happiness and
  heartbreak sharing one face.
  Seductive smile: The heavy-lidded eyes smolder through the lashes, the chin
  lowers, and one corner of the soft, full lips lifts in a slow, deliberate
  invitation.
  Wide open mouth smile: The lips draw fully back and the jaw drops in joy,
  revealing both rows of teeth, while the eyes curve into bright crescents
  above lifted cheeks.
  Wide open mouth smile while winking: The same beaming, open-jawed smile of
  pure delight, with one eye snapped playfully shut and the other curved into
  a happy crescent.
  Laughing (action): Caught mid-burst - the head tosses back, the eyes shut
  into crescents, the brows fly up, and the mouth opens to its widest as
  laughter visibly shakes the face.
  Defiant smile with mouth wide open: A baring, open grin thrown straight at
  the viewer, brows set low but eyes blazing with challenge, chin lifted in
  fearless provocation.
  Sad face with mouth agape: The jaw hangs slack in heartbreak, the inner
  brows tilt sharply upward, and the wide glistening eyes stare as the lower
  lip trembles.
  Cat face: The mouth curls into a small closed shape like a contented cat,
  beneath wide, round, innocent eyes and softly puffed cheeks.
  Devious cat face: The little mouth has a devious smile while the eyes
  narrow into sly half-lidded slits and the brows angle down - a scheme
  clearly in progress.
  Frowning cat face: The cat-like "w" mouth flips upside down into a grumpy
  curl, the brows pinch together, and the round eyes glower in miniature
  outrage.
  Smiling with tongue out: The lips part in a playful smile as just the tip
  of the tongue peeks through, matched by bright, teasing eyes.
  Winking with tongue out: One eye squeezes shut while the tongue pokes out
  to the side, the grin crooked and cheeky - maximum tease, zero apology.
  Licking lips: The tongue drags visibly across the upper lip mid-sweep, the
  hungry eyes fix on their target, and a faint anticipatory smile curls
  beneath.
  Smiling with furrowed brows and closed mouth: The brows knit with worry or
  emotion while the closed lips hold a gentle upward curve - relief,
  affection, and concern all at once.
  :( (Grumpy): The brows angle steeply down toward the nose, the eyes go flat
  and stern, and the mouth sets into a short, disgruntled downturned line.
  Puffed cheeks: The lips seal tight while held air balloons the cheeks
  round, the brows pinch, and the eyes glare sideways in silent indignation.
  :o (Circle shaped mouth): The lips form a clean, perfect circle beneath
  round widened eyes and lifted brows - astonishment distilled to its
  simplest shape.
  ;o (Mouth open one eye closed): The little circle of surprise on the mouth
  meets one eye squeezed shut - a quirky mix of shock and playful wink.
  Heart eyes: The pupils bloom into glowing pink hearts, sparkles scatter
  around the lashes, the cheeks flush and lift, and the open smile radiates
  helpless adoration.
  Eyes tight: The lids clamp down hard enough to fan crease lines from the
  corners, the brows press low, and the nose scrunches as the face braces for
  impact.
  Swirly eyes: The pupils are replaced with dizzy spirals, the head wobbles,
  and the mouth hangs open in a dazed slack line - completely overwhelmed and
  down for the count.
  Squinty eyes: Eyes squinting almost closed, crinkling the corners, as the
  face peers hard at something or flinches from the light.

  V1.0
  I have a bunch of examples but it would only let me upload a selection.
  trained on over 60 expressions and almost 4000 images. example images were
  on base Krea 2 model with mostly 1.25 strength up to 1.5 for a some (bloody
  nose) trained on booru tags and natural language. works well with the
  expression tag mixed with a good description of what you want the face to
  do. actual effectiveness may vary depending on the type of character and
  art style you're trying to use. This is just the first version. I'll take
  feedback and improve expressions that are either not included or not quite
  strong enough. please let me know what you want to see.

───────────────────────────────────────────────────────────────────────────────
  VERSION NOTES
───────────────────────────────────────────────────────────────────────────────
  Trained on over 6000 images featuring 73 characters across 50 Anime Series
  / Art Styles with over 750 additional realistic images. Trained on base
  model. Raise strength as needed. Use the list of expressions I provided and
  modify from there. Feel free to come up with your own expressions and it
  should perform well. Trained on over 70 expressions.

───────────────────────────────────────────────────────────────────────────────
  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, 1744x2560)
  02_image.jpeg  (image, 1808x2656)
  03_image.jpeg  (image, 2000x2828)
  04_image.jpeg  (image, 1744x2560)
  05_image.jpeg  (image, 1744x2560)
  06_image.jpeg  (image, 864x1536)

───────────────────────────────────────────────────────────────────────────────
  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 03:58 UTC.
  Stats and description are the author's; pros/cons are derived from the
  evidence tagged beside each line.
───────────────────────────────────────────────────────────────────────────────