With jina-clip-v2, users have a single embedding model that delivers competitive performance in both text-only and text-image cross-modal retrieval - across multiple major languages of the world.
Jina CLIP v2 is successor to Jina CLIP v1, with improved performance on text-only, text-to-image and image-to-image retrieval, and a huge improvement on retrieval of complex documents, such as screenshots containing a mix of text, visual elements such as charts and, graphics.

Highlights:

Multimodal: Superior performance on all combinations of modalities, and especially large improvements in text-only embedding performance.

Multilingual: Trained with support for 89 major world languages. Additionally, Jina Embeddings’ 8k token input support makes it possible to process detailed textual information and correlate it with images.

Matryoshka Learning Enabled : Supports flexible embedding sizes – between 2 and 1024, allowing for truncating embeddings to fit your storage needs.

Usage:

Please refer to this link for detailed usage.