map

semantic cluster view. each dot is an image; position reflects caption-embedding similarity. projected via UMAP.

scroll = zoom · drag = pan

112 of 128 points · stale · last computed 5/28/2026, 3:40:36 PM