SAM WEISS
My details are in the game. Three ways to get them out.
SUBWAY — WHEN TO LEAVE
1 · PHOTOS
Drop photos here
They never leave this machine — everything below runs in your browser, and nothing is uploaded anywhere.
2 · GROUPING
Two photos group when they look alike and were taken close together. Looks come from DINOv2, which describes the particular scene rather than what kind of picture it is — CLIP would put every landscape in one heap because it reads them all as “scenery”. Time comes from the EXIF capture date, which no image model can know: it is what separates two shots of one breakfast from two viewpoints an hour apart on the same walk. Set the time weight to zero to group on looks alone. Photos with no EXIF date are grouped on looks regardless.
Scoring weights
The keeper of each group is decided mostly by sharpness — every frame of a burst shares a composition, so focus is what actually differs, and the aesthetic model rates them all about the same. It's ranked within each group, so a busy scene doesn't beat a simple one just for having more edges.
3 · RUN
The first run downloads the CLIP encoder — 85 MB, once, then kept in your browser cache. It uses your GPU where there is one and the CPU otherwise; either way it reports its real speed after a few photos and you can stop it. Nothing is uploaded.