feat: ML alt text on upload + object-label search (Azure Vision)

[sinagkh]

Summary

Add ML-powered accessibility and search to Moments:

• Auto alt text generated on upload (Azure AI Vision).
• Object search via a new Object category that queries ML-detected labels.
• Ensure every user photo renders with an alt="..." for accessibility.

What changed

• Service: moments/services/vision.py
  • Calls Image Analysis 4.0 (caption,tags). If caption isn’t supported in the region, falls back to:
    • 4.0 tags (for search), and
    • Vision 3.2 describe (for captions).
• Upload flow: moments/blueprints/main.py
  • On upload, analyzes bytes and stores Photo.alt_text + Photo.detected_labels.
• Search: main.search supports category=object (label-based LIKE query).
• Model: Photo gets alt_text and detected_labels columns (+ migrations).
• Templates:
  • Add alt="{{ photo.alt_text or photo.description or 'Image' }}" to all user photos.
  • Photo page shows Detected: labels and a small Generated badge when alt is ML-derived.
• CLI: flask reanalyze to backfill captions/labels for older photos.
• Docs: Updated README.md and .env.example.

How to run (quick)

uv python pin 3.11
uv sync
cp .env.example .env
python3 -c 'import secrets; print("SECRET_KEY=" + secrets.token_hex(32))' >> .env
# set AZURE_VISION_ENDPOINT / AZURE_VISION_KEY in .env

uv run flask --app app init-app
# optional demo:
# uv run flask --app app lorem

uv run flask --app app run
# http://127.0.0.1:5000/
# upload at /upload

[sinagkh]

Verification steps (for TA)

1. Setup

   • Python 3.11 with uv
   • cp .env.example .env
   • set AZURE_VISION_ENDPOINT / AZURE_VISION_KEY in .env
   • uv sync && uv run flask --app app init-app

2. Run

   • uv run flask -e .env --app app run
   • Upload a new photo at /upload

3. Expected behavior

   • The uploaded photo page renders <img ... alt="..."> with a natural-language caption.
   • /search?category=object&q=<one of the detected labels> (e.g., dog, tree) returns the photo.

Notes

• ML runs on upload. To populate older photos: uv run flask -e .env --app app reanalyze.
• No credentials are committed; secrets loaded from .env.