ComfyUI Node
Krea2 Edit Prep (gg)
Prepares model and conditioning for KSampler from zero, one or two reference images: no image = plain text-to-image, one image = single-reference edit, two = scene + subject. A boost of 0 leaves that image out.
Krea2 Edit Prep (gg)
- model
- clip
- vae
- latent
- image
- image_b
- image_mask
- image_b_mask
- model
- positive
- negative
◄prompt►
◄grounding_px1024►
◄fit_modefit►
◄image_boost1.00►
◄image_b_boost1.00►
◄negativezero (CFG 1)►
◄image_positionaligned (trained)►
◄image_b_positionbeside (right of target)►
◄rebalancefalse►
◄rebalance_multiplier1.00►
◄image_usegrounding + pixels►
◄image_b_usegrounding + pixels►
Categorykrea2/gg
Inputs (20)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | Krea 2 with an edit LoRA (identity or general) applied | |
| clip | CLIP | — | |
| vae | VAE | — | |
| latent | LATENT | the same latent you feed KSampler.latent_image | |
| prompt | STRING | — | |
| grounding_px | INT | 10240–4096 | lower = stronger edit adherence, higher = stronger identity; 0 = native |
| fit_mode | COMBO | fit | 2 options: fit, crop (legacy) |
| image_boost | FLOAT | 1.000–1000 | fidelity to image; 0 leaves image out |
| image_b_boost | FLOAT | 1.000–1000 | fidelity to image_b; 0 leaves image_b out |
| negative | COMBO | zero (CFG 1) | grounded = empty prompt on the same images, the trained unconditional for CFG > 1 |
| image_position | COMBO | aligned (trained) | aligned = pinned to the output layout; beside = identity without the layout pin, restages more freely; bag of features = experimental |
| image_b_position | COMBO | beside (right of target) | beside keeps the person's identity but copies much less of the photo's pose and lighting |
| imageopt | IMAGE | scene, or the only reference | |
| image_bopt | IMAGE | subject; used as the only reference when image is absent | |
| image_maskopt | MASK | region of image its boost applies to, e.g. a face; used only when image is the only reference | |
| image_b_maskopt | MASK | region of image_b its boost applies to, e.g. the face | |
| rebalanceopt | BOOLEAN | false | applies Conditioning Krea2 Rebalance's default per-layer weights to positive and a grounded negative |
| rebalance_multiplieropt | FLOAT | 1.000–10 | scales every layer; about 0.7-0.8 with the V2 LoRAs, 1.0 with the V1.2 LoRA or no edit LoRA |
| image_useopt | COMBO | grounding + pixels | grounding = the text encoder sees the image (objects, outfit, scene, by meaning); pixels = the image itself goes into the model (faces, layout when aligned) |
| image_b_useopt | COMBO | grounding + pixels | the same choice for image_b |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |
| positive | CONDITIONING | — |
| negative | CONDITIONING | — |