Cant create multimodal search due to can not find the correct models

Hi @System Administrator

Thank you for reaching us for your question and for referencing the related thread. I appreciate the detailed information you provided regarding your efforts with different regions and deployments.

The main reason the Cohere embedding models don't appear in the Vectorize your images step or the multimodal workflow is that, although Cohere-embed-v3-english and Cohere-embed-v3-multilingual are strong text embedding models and can be deployed as serverless endpoints, the Azure AI Search portal wizard mainly supports them for text vectorization or image verbalization, not for direct multimodal (image and text) embeddings.

For genuine multimodal RAG, if you want to vectorize images directly instead of converting them to text with an LLM, the most dependable and recommended approach is to use Azure Vision multimodal embeddings. This solution works smoothly with the wizard and supports both text and images within a unified vector space.

Could you please follow the steps outlined below to ensure efficient implementation

1. Please create or utilize a Foundry resource (multi-service) in a region that supports Azure Vision multimodal APIs. Most regions are compatible; however, East US 2, West US 3, and Sweden Central are recommended options.
2. In the Azure portal, go to your Azure AI Search service > Import data (new) > select your data source > choose Multimodal RAG.
3. At the embedding stage, please select Multimodal embeddings and then choose the Azure Vision option. This option will be available automatically once the necessary permissions are configured.
4. Please ensure that your search service has the appropriate role assignments on the Foundry resource, such as the Cognitive Services User role through managed identity.

Once configured, the wizard will manage the extraction and vectorization of both text and images from your documents, such as PDFs.

If you still want to use Cohere models, they're available for text embeddings and will appear in the text vectorization options after deploying serverless. You can also combine them with image verbalization if needed. However, Azure Vision offers the simplest solution for multimodal tasks.

Reference

• Quickstart for multimodal search in the portal: https://learn.microsoft.com/en-us/azure/search/search-get-started-portal-image-search?referrer=grok.com&tabs=search-perms%2Copenai-perms%2Cdocument-extraction%2Cimage-verbalization
• Supported models and methods table (shows Azure Vision as primary for direct multimodal): https://learn.microsoft.com/en-us/azure/search/search-get-started-portal-image-search?referrer=grok.com&tabs=search-perms%2Copenai-perms%2Cdocument-extraction%2Cimage-verbalization#supported-embedding-methods
• Deploying models as serverless endpoints (for Cohere if you go that route):
  https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/deploy-models-serverless?view=foundry-classic&referrer=grok.com&tabs=azure-direct&pivots=ai-foundry-portal
• Integrated vectorization overview: https://learn.microsoft.com/en-us/azure/search/vector-search-integrated-vectorization-ai-studio

Kindly let us know if the above helps or you need further assistance on this issue.

Please "upvote" if the information helped you. This will help us and others in the community as well.