2026-05-13T21:09:32.444Z294 KB
Jina v5 Omni Image-Caption Embedding Map
6 image-caption pairs from a small Hugging Face dataset, embedded in one multimodal space and projected with UMAP.
Open studyA single parent page for all published analyses. Each study is stored as a static HTML file in Cloudflare KV and linked from here.
6 image-caption pairs from a small Hugging Face dataset, embedded in one multimodal space and projected with UMAP.
Open studyA local run of jinaai/jina-embeddings-v5-omni-small embeds translated Tatoeba phrases into the same vector space. UMAP turns the high-dimensional vectors into this two-dimensional map; each faint line ties translations of the same English source sentence across languages.
Open study