AgentSkillsCN

comfyui-image-variant

基于现有基础素材,创建资产衍生品(状态变更、伤害等级、UI 状态效果)的转换流程。当某一资源已存在,但需要生成其变体时,可采用此流程。

SKILL.md
--- frontmatter
name: comfyui-image-variant
description: Transformation workflow for creating asset derivatives (state changes, damage levels, UI status effects) from existing base images. Use when a resource already exists but needs a variant.

ComfyUI Image Variant

Produce stylistic or state-based derivatives while maintaining structural parity.

Workflow

  1. Delta Definition: Determine the target change (e.g., "damaged version," "alternative expression").
  2. Technique Selection:
    • Consult variant-config.md for Denoising and ControlNet settings.
    • Use Inpainting for localized changes (e.g., changing just a helmet).
  3. Execution:
    • Load image_to_image_controlled.json.
    • Pass the base image as the initial latent.
  4. Integration: Update .tres Resource files to include the new variant path.

Guidelines

  • Parity: Compare the original and variant at 50% opacity to ensure no "drift" in character anatomy.
  • Naming: Use logical suffixes (e.g., _inj, _up1, _stat_stun).

Artefacts to Update (when applicable)

  • If the new variant introduces or changes resource paths, update the corresponding .tres files and list the change in ./docs/todo/master_todo.md under the related story.
  • If prompts, skills, or docs refer to the asset names or behavior, add a short note in those files listing the updated paths (e.g., ./.github/prompts/, ./.github/skills/, ./docs/).
  • Commit artefact changes separately with a clear message (e.g., enh - doc: add image variant <name>).