Nodes/ComfyUI-Krea2Moodboard/Krea 2 Moodboard + Identity Fusion
ComfyUI Node

Krea 2 Moodboard + Identity Fusion

Moodboard style + identity edit fused in a single encode (Neo-parity). Both image inputs are optional: ID only = plain identity edit, moodboard only = plain vibe transfer, both = fusion. Positive only; negative = Krea 2 Identity Edit with empty prompt + same image.

By RedNodeAI·Created about a month ago·Updated 28 days ago· 9
Krea 2 Moodboard + Identity Fusion
  • clip
  • edit_source
  • moodboard_images
  • vae
  • edit_source2
  • sources
  • target_latent
  • CONDITIONING
instruction
strength0.50
extract
reference_processing
style_directivetrue
indirecttrue
budget_px384
grounding_px768
ref_boost1.00
ref_boost_a1.00
fit_modefit
Categoryconditioning/krea2

Inputs (18)

NameTypeDefaultDescription
clipCLIP
instructionSTRINGthe edit instruction, e.g. 'create a photo of this person at a night market'
strengthFLOAT0.500–1
extractCOMBO2 options: style / vibe, subject / concept
reference_processingCOMBO3 options: full image, quadrant crops (2x2), fine tiles (4x4)
style_directiveBOOLEANtrue
indirectBOOLEANtruedelete moodboard rows after encoding; style survives via in-encoder attention. Safest when style refs contain people.
budget_pxINT384128–1536
grounding_pxINT7680–2048longest-side cap for the edit source fed to the encoder
edit_sourceoptIMAGEidentity source image. Not connected = pure moodboard mode.
moodboard_imagesoptIMAGEstyle references (batch for several). Not connected = pure identity-edit mode.
vaeoptVAEconnect to attach the in-context identity latents (required for actual editing)
edit_source2optIMAGE2nd reference for two-ref LoRAs (scene first, subject second)
sourcesoptKREA2_SOURCESchained sources (Krea2 Edit Source Chain) — appended after edit_source/edit_source2 as frames 3..N. 3+ refs is beyond the LoRA's training; identities may blend.
ref_boostoptFLOAT1.000–10reference-fidelity dial: multiplies target->reference attention for the LAST identity ref (the subject). 1.0 = off; >1 pulls harder toward the reference (the v1.2 edit-LoRA author suggests 2-6). Positive only — the moodboard span is unaffected.
ref_boost_aoptFLOAT1.000–10same dial for the earlier identity refs (the scene in two-ref workflows). No effect single-ref. 1.0 = off
target_latentoptLATENTconnect your (empty) sampling latent to enable the v1.2 'fit' geometry: identity refs are fitted in PIXEL space to the output resolution before VAE-encoding — fixes blur from resolution mismatch and removes the match-the-aspect-ratio requirement. With CFG > 1, connect the same latent to the negative edit node too.
fit_modeoptCOMBOfithow identity refs fit a mismatched output AR (needs target_latent + vae): fit = resample to the target grid at a centered offset (v1.2-trained geometry); crop (legacy) = center-crop to the target AR then resize (v1/v1.1 geometry, for older weights).

Outputs (1)

NameTypeDescription
CONDITIONINGCONDITIONING