preserving-faces
Composable sub-Skill that compiles a face-locking ComfyUI prompt JSON by re-rendering the non-face region at very low denoise while leaving the face mask untouched. Use whenever the planner emits a st
SKILL.md.claude/skills/preserving-faces/SKILL.md
GitHubpreserving-faces
Composable sub-Skill. Given an image and a face mask, emits a prompt that regenerates only the non-face region at low denoise, keeping the face pixel-stable.
When this Skill applies
- Planner step with
{goal:"preserve_face"}. - Or as a "seal" step after
transferring-styles/inpainting-regionswhen identity drift is a risk.
Intent overrides
| VIO | Node.input |
|---|---|
prompt |
6.inputs.text |
negative_prompt |
7.inputs.text |
inputs.image |
10.inputs.image |
inputs.mask (face region) |
11.inputs.image |
params.denoise |
3.inputs.denoise (default 0.25) |
params.*, model.checkpoint |
standard |
Notes
preserving-faces is composable: its output can be piped directly into a
downstream step or used standalone. The downstream evaluator
evaluating-identity-preservation validates the result.