vtl-image-analysis
由 rusparrish 提供
Image & Video Generation简介
Measure compositional structure in AI-generated images using the Visual Thinking Lens (VTL) framework.
描述
该技能的功能
Measures compositional structure in AI-generated images using the Visual Thinking Lens (VTL) framework. Detects patterns like center lock, radial collapse, and low tension that indicate default-mode model behavior. When patterns are found, generates alternative prompts targeting specific compositional issues.
Uses deterministic metric coordinates instead of subjective feedback, so compositional diagnosis is repeatable and not influenced by the model's priors.
何时使用
- Diagnosing why a generated image looks generic or flat
- Getting re-prompt variants after a center-locked image output
- Checking visual tension and void distribution in AI art
- Iterating on prompts that consistently produce similar layouts
- Validating compositional quality before finalizing a generated image
示例工作流
User asks: 'Why does this image look so bland and centered?'
- 1
Run vtl_probe.py on the image to extract compositional metrics
- 2
Check valid and mask_status fields; halt and report failure if either fails
- 3
Report the five VTL coordinates and any flags detected
- 4
Run vtl_regen.py with the original prompt and metrics to generate re-prompts
- 5
Present the selected prompt variant as the primary recommendation with alternatives
VTL coordinates showing CENTER_LOCK flag, plus up to 3 revised prompt variants targeting off-center placement and increased tension
安装
clawhub install rusparrish/vtl-image-analysis运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。