AI Photo Pipeline Skill

This skill allows you to process restaurant photos using your internal multimodal capabilities, bypassing external API limits.

Workflow

1. •

   Locate Candidates:

   • •List files in src/authentik/images/<collection_slug>/*.jpg.
   • •Look for patterns like *-candidate-*.jpg.
2. •

   Analyze & Select:

   • •For each candidate photo, view it (using view_file or by context).
   • •Analyze it for:
     • •Food Appeal: Is it delicious?
     • •Quality: Is it focused and well-lit?
     • •Content: Is it food (priority) or interior/exterior?
   • •Select the top 3 best food photos.
3. •

   Enhance:

   • •For the selected 3 photos, use your generate_image tool.
   • •Prompt: "Enhance the lighting and color of this food photo to make it look like a high-end food delivery app studio shot. STRICTLY PRESERVE the food content, ingredients, and arrangement. Do NOT add, remove, or change any food items. Do NOT hallucinate new garnishes. Your goal is 'Relighting' and 'Color Grading' only. Make it vibrant and appetizing, but keep the geometry 100% identical to the original."
   • •Input: Pass the original candidate image path in image_paths.
   • •Output: Save as <short_id>-enhanced-<index>.jpg in the same directory.
4. •

   Report:

   • •Summarize which photos were selected and why.