What problem does it solve? When chaining ComfyUI steps like style transfer, inpainting, or background replacement, the subject's face often drifts between steps. This Skill locks the face region by generating a prompt graph that re-renders only the non-face area at very low denoise, keeping facial identity pixel-stable. ## Core Features & Use Cases - Face-region locking: Inverts a supplied face mask with InvertMask and applies it via SetLatentNoiseMask so only non-face pixels are regenerated. - Low-denoise re-render: Defaults the KSampler denoise to 0.25 so surrounding content changes subtly while the face stays untouched. - Composable sub-Skill: Triggered by planner steps with goal=preserve_face, or chained as a seal step after style transfer or inpainting; output can feed downstream steps or the evaluating-identity-preservation evaluator. - Use Case: After replacing a portrait's background, run this Skill with the original image and face mask to restore any facial drift while keeping the new background. ## Quick Start Ask the agent to preserve the face in the current image using the provided face mask while re-rendering the rest of the scene at low denoise.
