How Our Photo Tagging System Works

From raw photos to fully organized, searchable assets — automatically.

1
📥

Photos Arrive

Drop them in a folder — the system picks them up automatically.

2
🤖

AI Reads Every Photo

Tags the product AND the scene — shot angle, surface type, setting, season — instantly.

3
✏️

You Review & Correct

A simple web page shows each photo with suggested tags — approve or fix in one click.

4
🧠

System Learns from Your Corrections

Every fix you make trains the AI to get it right next time.

5

Photos Are Ready to Use

Fully tagged, searchable, and organized — ready for any project.

Self-improving loop

The more your team reviews, the smarter the AI gets. Over time, the system requires fewer and fewer corrections — eventually tagging photos almost perfectly on its own.

corrections loop back to train the AI
🛒
Product Photos Come In Pre-Tagged

Images tied to products on our websites (PetStep, HandiRamp, HandiTreads) are imported automatically. The system already knows the product, SKU, and color variant — so AI only fills in what it can’t determine from the catalog: shot angle, setting, photo style. Your team reviews a nearly-complete tag set from the start.

⬆️
Originals Automatically Replace Website Photos

Website images are compressed. When the original high-resolution photo is added later, the system recognises it as the same image and promotes it automatically — carrying every tag forward to the original, and retiring the smaller version. No re-tagging, no manual merging.


Where your team touches the system

Your Points of Contact

📂
Drop Folder A shared Dropbox folder. Drag photos in — everything else happens automatically.
🏷️
Review Portal A web page that shows each photo with AI-suggested tags. Approve what's right, fix what isn't.
🔍
Search Portal Find any photo by product, angle, setting, or keyword — e.g., all outdoor rear-angle CVR-52 photos. Every approved photo is searchable instantly.

What's Running Under the Hood

Storage & Sync
📦 Dropbox
AI Models
🤖 llava-llama3 local
Reads each photo, suggests tags — runs on-site, never leaves the network
☁️ Claude Sonnet cloud
Fallback for low-confidence tags — only called when local model isn't sure
📚 qwen2.5:32b local
Distillation — turns corrections into improved tagging rules over time
Metadata
🏷️ Product + scene tags in every file
Portals
🌐 Review UI
🔍 Search UI
Learning
📝 Correction log
📚 Tagging rules