ComfyUI-OpenClaw/web/docs/OpenClawImageToPrompt.md

OpenClawImageToPrompt

Generate prompts from images using vision LLMs.

Inputs

Name	Type	Description
image	IMAGE	Input image to analyze
goal	STRING	Analysis goal (e.g., "Describe for regeneration")
detail_level	COMBO	low, medium, high
max_image_side	INT	Max dimension for resizing (256-1536)

Outputs

Name	Type	Description
caption	STRING	Descriptive caption of the image
tags	STRING	Comma-separated style/content tags
prompt_suggestion	STRING	Suggested prompt to recreate the image

Example Usage

┌─────────────────┐     ┌───────────────────┐     ┌─────────────┐
│ Load Image      │────▶│ OpenClawImage     │────▶│ KSampler    │
│                 │     │ ToPrompt          │     │ (img2img)   │
└─────────────────┘     └───────────────────┘     └─────────────┘

1. Load an image you want to analyze
2. Set detail level and goal
3. Use prompt_suggestion as input for generation

Safety Notes

• S3: LLM responses sanitized
• S4: Image data encoded as base64 (no file path exposure)
• Requires vision-capable model (e.g., gpt-4o, claude-3-5-sonnet)

Troubleshooting

• Verify provider supports vision (OpenAI/Anthropic)
• Check /openclaw/health for API status (legacy /moltbot/health still works)