Origin

Origin

Photos get reposted without credit. Someone screenshots a post, puts it up somewhere with no attribution, it gets scraped onto a wallpaper site, and after a few hops nobody knows who took it.

Origin is a page where you upload any image and it tells you which Picashot post it came from. It has to work on a copy, so the image it's given is usually resized, re-encoded, slightly cropped and generally degraded.

It runs in two passes.

• Every image gets a perceptual hash when it's uploaded, which survives resizing and re-encoding. The first pass finds the closest hashes to the one you gave it. That covers most cases and it's fast.
• If it comes back with nothing — a hard crop, a heavy filter, too many hops — it falls back to running the image through CLIP and searching the resulting vector against every embedding on the platform, using pgvector on an HNSW index.

The second pass is slower, and it's a similarity match rather than a fingerprint match, so the result tells you which of the two found it.

• Origin · picashot.co