Image Search
Visual product search via CLIP-based image embeddings. Self-hosted, no external API costs. Requires pgvector.
Image Search
Image Search lets customers upload a photo to find visually similar products. It uses a CLIP model running on your own server — no external API calls and no per-request costs.
Requirements
- PostgreSQL with the
pgvectorextension - Sufficient server CPU/RAM to run inference (a model download happens on first use via
@huggingface/transformers)
Enable pgvector and run the migration:
# On the server
sudo -u postgres psql -c "CREATE EXTENSION IF NOT EXISTS vector;"
# In the API directory
npm run migration:run:image-searchInstallation
Install from Admin → Plugins → Store. No license key required.
Settings
| Setting | Description |
|---|---|
| Enabled | Turn image search on or off |
No API keys needed — the model runs locally.
How It Works
When a product image is saved, the plugin generates a CLIP visual embedding and stores it in pgvector. When a customer uploads a photo in the search bar, the same model embeds the image and a nearest-neighbour lookup returns visually similar products.
First inference after a server restart takes longer while the model loads into memory.