Image Analysis

Image Analysis

Analyze images that users send in chat. Covers vision API usage, large-image handling, and failure recovery.

Trigger

User sends an image and you need to describe or extract information from it.

Workflow

1. Check image dimensions and size first:

   file /path/to/image && ls -lh /path/to/image
   

   MiniMax VLM reliably rejects images >~10MB or >4000px. Pre-resize to avoid wasting turns on 1033 errors.

2. Resize large images with PIL (ImageMagick convert may not be installed):

   python3 -c "
   from PIL import Image
   img = Image.open('/path/to/image.jpg')
   img.thumbnail((1280, 1280))
   img.save('/path/to/image_sm.jpg', quality=85)
   "
   

   Target ~0.3MB, 1280px max dimension.

3. Analyze with MiniMax VLM: Use mcp_minimax_mcp_understand_image with the resized (_sm) version. Start with a simple prompt like "What is in this image?" — complex prompts can trigger 1033 errors when the service is under load.

4. If 1033-system-error persists on small images: The MiniMax VLM service is likely down. In that case, retry once with an even simpler prompt, then inform the user the vision service is unavailable.

Pitfalls

• Large images (>10MB, >4000px) → persistent 1033-system-error from MiniMax VLM. Always resize first.
• ImageMagick convert may not be installed (common in WSL). Fall back to python3 -c "from PIL import Image...".
• execute_code may be blocked in cron or restricted contexts. Use terminal with python3 -c directly.
• MiniMax VLM can have transient outages — 1033 errors even on small images. A retry with a minimal prompt often succeeds.