map

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

scroll = zoom · drag = pan

427 of 427 points · last computed 8/19/2026, 7:20:17 AM