manifold

3D semantic point cloud. each dot is an image; position reflects caption-embedding similarity projected via UMAP to 3 dimensions.

113 of 128 points · stale · seed 42 · last computed 5/29/2026, 6:39:41 AM