ComfyUI Node
Krea 2 Identity Edit
A ComfyUI node in conditioning/krea2 with 12 inputs and 1 output.
Krea 2 Identity Edit
- clip
- vae
- image
- image2
- fuse_with
- sources
- target_latent
- CONDITIONING
◄prompt—►
◄grounding_px768►
◄ref_boost1.00►
◄ref_boost_a1.00►
◄fit_modefit►
Categoryconditioning/krea2
Inputs (12)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | CLIP | — | |
| prompt | STRING | Edit instruction. Leave empty on the negative node. | |
| vaeopt | VAE | — | |
| imageopt | IMAGE | — | |
| image2opt | IMAGE | — | |
| grounding_pxopt | INT | 7680–4096 | Cap on the longest side fed to Qwen3-VL (the identity LoRA trained with 384-768px). 0 = never resize. |
| fuse_withopt | CONDITIONING | Optional conditioning to fuse in front of this one (e.g. Krea 2 Moodboard for scene/style vibe). Its token rows are prepended; this node's identity reference latents are kept. Matches the Neo moodboard+edit fusion layout. | |
| sourcesopt | KREA2_SOURCES | chained sources (Krea2 Edit Source Chain) — appended after image/image2 as frames 3..N. 3+ refs is beyond the LoRA's training; identities may blend. | |
| ref_boostopt | FLOAT | 1.000–10 | reference-fidelity dial: multiplies target->reference attention. Applies to the LAST ref (the subject in two-ref workflows, the only ref in single-ref). 1.0 = off; >1 pulls harder toward the reference's appearance (the v1.2 edit-LoRA author suggests 2-6); <1 loosens. Set on the POSITIVE node; leave the negative at 1.0. |
| ref_boost_aopt | FLOAT | 1.000–10 | same dial for the earlier refs (the scene in two-ref workflows). No effect in single-ref workflows. 1.0 = off |
| target_latentopt | LATENT | connect your (empty) sampling latent to enable the v1.2 'fit' geometry: 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 it to the negative edit node too so both passes share one geometry. | |
| fit_modeopt | COMBO | fit | how refs fit a mismatched output AR (needs target_latent + vae): fit = resample to the target grid at a centered offset, matching how the v1.2 edit LoRA was trained; crop (legacy) = center-crop to the target AR then resize (v1/v1.1 geometry, for older weights). |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| CONDITIONING | CONDITIONING | — |