One image+text embedding model: photos and queries land in the same space.
L2-normalized, so a dot product is the cosine similarity.