ComfyUI Node
Eric Krea2 Paint (inpaint / outpaint)
A ComfyUI node in Eric/Krea2 with 16 inputs and 4 outputs.
Eric Krea2 Paint (inpaint / outpaint)
- image
- mask
- paint
- vlm_image
- canvas_preview
- generated_mask
◄adapteranypaint►
◄pad_left0►
◄pad_right0►
◄pad_top0►
◄pad_bottom0►
◄seam_px32►
◄reference_max_edge384►
◄restore_s1on►
◄restore_s2auto►
◄restore_s3auto►
◄adapter_s11►
◄adapter_s20►
◄adapter_s30►
◄structure_release_sigma0.40►
CategoryEric/Krea2
Inputs (16)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | Source image. Kept regions are pinned to it at full resolution. | |
| maskopt | MASK | 1 = generate / edit (ComfyUI mask-editor convention). Source-sized or canvas-sized. Optional for pure outpaint. | |
| adapteropt | COMBO | anypaint | anypaint: arbitrary masks, inpaint+outpaint (needs the AnyPaint LoRA on the LoRA stack). outpaint: older rectangular-only adapter. none: training-free - only pins the kept region (fine for small inpaints, weak for outpaint). |
| pad_leftopt | INT | 00–4096 | Outpaint: pixels added on the left (source scale). |
| pad_rightopt | INT | 00–4096 | — |
| pad_topopt | INT | 00–4096 | — |
| pad_bottomopt | INT | 00–4096 | — |
| seam_pxopt | INT | 320–256 | Band around the mask that is regenerated (not pinned) so the model can blend. Canvas pixels; scaled per stage. Adapter default 32. |
| reference_max_edgeopt | INT | 384128–1024 | Condition image size. 384 is the adapter's training contract - change only to experiment. |
| restore_s1opt | COMBO | on | Pin kept tokens to the source at Stage 1. |
| restore_s2opt | COMBO | auto | auto: 'on' when the source has at least this stage's resolution, else 'structure'. on: pin kept regions every step (exact source pixels). structure: pin while sigma > structure_release_sigma (keeps composition and identity), then let the low-noise steps add detail - avoids forcing an upsampled small source's blur. off: S2 may rewrite kept regions. |
| restore_s3opt | COMBO | auto | 4 options: auto, on, structure, off |
| adapter_s1opt | FLOAT | 10–1 | Adapter reference conditioning at Stage 1 (1 = on, 0 = off). |
| adapter_s2opt | FLOAT | 00–1 | At Stage 2. Default off: the adapter was trained at <=1024 px targets. |
| adapter_s3opt | FLOAT | 00–1 | — |
| structure_release_sigmaopt | FLOAT | 0.400–1 | restore 'structure' (and auto on upscale stages): kept regions are pinned while sigma is above this, free below. Lower = closer to the source, higher = more fresh detail (and more drift). |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| paint | KREA2_PAINT | — |
| vlm_image | IMAGE | — |
| canvas_preview | IMAGE | — |
| generated_mask | MASK | — |