4 ms·
For pure image embedding, I find DINOv3 to be quite good. For multimodal embedding, maybe RzenEmbed. For captioning I would use a regular multimodal LLM, Qwen
by daemonologist 11mo ago
For pure image embedding, I find DINOv3 to be quite good. For multimodal embedding, maybe RzenEmbed. For captioning I would use a regular multimodal LLM, Qwen 3 or Gemma 3 or something, if your compute budget allows.