1 · Input
PNG, JPG, JPEG, WEBP or GIF. Drag & drop or browse. Images never leave your machine except inside the embedding request you trigger.
One string per line. Sent as the OpenAI-compatible input field to POST /v1/embeddings — a single line goes as a string, several as an array. Pick a text-capable model (CLIP, SigLIP, jina, Qwen3-VL); image-only encoders like the default DINOv2 will reject text.
2 · Embeddings
Cosine similarity
Pairwise cosine similarity between the uploaded images (1.00 = identical direction).