Local-First Photosearch

I take photos on too many devices.

Sony, Fujifilm, as main camera and iPhone and an Android phone(Wife's phone).

The pipeline was a mess:

Camera photos → Lightroom on laptop Phone photos → device

After a trip, I would dump everything in a random folder. I didn't have a way to index them.

No way to search across all of the photos.

Looking for “sunset from the boat” meant scrolling through thousands of images.

So I built a local-first photo search engine.

Drop in a folder with photos from anywhere. It indexes everything. You search in plain English.

No cloud. No APIs. No sync headaches.

Under the hood:

• YOLOv8 → objects
• SmolVLM → scene + context
• SQLite → index
• Streamlit → UI
• Apple MPS → runs locally with decent speed

Each image goes through:

• Object detection + scene tagging
• EXIF extraction (date, GPS, camera, focal length)
• Blur detection (Laplacian variance)
• Closed-eye detection (MediaPipe FaceMesh)
• Duplicate detection (perceptual hashing)

Everything is indexed into SQLite.

Query looks like:

baby playing on the beach
night streetfood
portraits 50mm

Here is a short demo of the MVP:

It combines labels + metadata. No embeddings (yet).

~2s per image to index. Slow, but fine for now.

The real win for me is that everything is in one place, and I can actually find what I'm looking for.

Next steps are to improve speed, ranking, maybe embeddings, maybe face clustering.