Image Embedding Support

[faradaytrs]

Feature Description

Currently, the Vercel AI SDK only supports text embedding models. We would like to request the addition of support for creating embeddings for images as well.

As image-based machine learning and multi-modal AI applications become more common, the ability to generate embeddings for images would greatly enhance the SDK’s capabilities. This feature would allow developers to leverage image similarity, classification, and other AI-driven tasks directly within the Vercel ecosystem.

Use Cases

1. Image Similarity: Create vector embeddings from images and compare them for similarity.
2. Multi-modal Applications: Combine text and image embeddings for more comprehensive AI models (e.g., image captioning, visual question answering).
3. Image Classification: Use embeddings for classification tasks where images need to be mapped to specific categories.

Additional context

We believe this feature would provide significant value and open up new possibilities for developers using the Vercel AI SDK. Looking forward to seeing this feature in a future release!

[lgrammel]

Please list providers that support image embeddings.

[faradaytrs]

Please list providers that support image embeddings.

For example Amazon Titan Multimodal Embeddings G1 model can do it, there are other models as well, if you need more i can make a research.

[lgrammel]

We'll probably hold back on this for now until major providers such as OpenAI support it.

[cjlryan]

Would also love this - currently have to duck out of the AI SDK for multimodal embeddings through VertexAI.

[bastotec]

Would also love this - currently have to duck out of the AI SDK for multimodal embeddings through VertexAI.

Same here