Nomic Embed Vision
FreeDrop-in cross-modal search
FreeFree tier
Inputs: image
LinksX
About Nomic Embed Vision
Nomic Embed Vision v1.5 is a high-performing vision embedding model developed by Nomic AI. It shares the same embedding space as Nomic Embed Text v1.5, enabling seamless cross-modal search and retrieval. The model achieves strong benchmark scores (ImageNet 0-shot 71.0, Datacomp 56.8, MTEB 62.28) and is optimized for multimodal RAG, text-to-image, and image-to-text retrieval. It can be used via the Nomic Embedding API with a simple Python client or with Hugging Face Transformers. The model is open-source and supports hosted inference on Hugging Face.
Key Features
Shares same embedding space as Nomic Embed Text v1.5
Supports multimodal retrieval (text to image, image to text)
Open-source with hosted inference API and Hugging Face integration
High benchmark scores: ImageNet 0-shot 71.0, Datacomp 56.8, MTEB 62.28
Easy to use Python client (nomic import embed)
Compatible with Transformers for local inference
Includes data visualization via Nomic Atlas
Pros & Cons
Pros
- High accuracy on vision benchmarks
- Seamless integration with text embedding for cross-modal tasks
- Free and open-source
- Well-documented with blog, technical report, and sample code
- Easy to get started via API or Transformers
Best For
Multimodal RAG (retrieval-augmented generation)Text-to-image retrievalImage-to-text retrievalCross-modal search and recommendationVisualizing embedding spaces with Nomic Atlas
FAQ
How do I use Nomic Embed Vision for text-to-image retrieval?
Use the 'search_query:' prefix for text queries when encoding with Nomic Embed Text v1.5, then compute cosine similarity between image and text embeddings.
Is Nomic Embed Vision v1.5 free to use?
Yes, the model is open-source and available for free download or via the Nomic Embedding API (API usage may have its own pricing).
What is the model size of Nomic Embed Vision v1.5?
The model has 92.9 million parameters and uses FP32 tensor type.