Deployed cohere embedding model not available in Azure Search import and vectorize image wizard

Hi @Zhang, Chong,

Thank you for submitting your question on Microsoft Q&A.

That message typically indicates that the wizard cannot detect any embedding model deployment that supports image vectors and is accessible for use. During the “Vectorize your images” step, Azure AI Search only lists model deployments that meet all of the following requirements

Requirements for a model deployment to appear

1. Supports image embeddings The model must be designed to generate image vectors (not just text embeddings).
2. Deployed in the correct mode For Foundry catalog models, it must be deployed as a serverless API deployment, not batch or other types.
3. Accessible to the Search service Proper permissions, such as RBAC or valid keys, must be configured so that Azure AI Search is allowed to call the model.

1. Use a supported image-embedding model and the correct deployment type

In the Import data (new) > RAG flow wizard, Azure AI Search will only display model deployments that support image embeddings and are deployed in a compatible way.

The Supported embedding models list includes Foundry catalog models that handle both text and images (multimodal), such as:

Cohere-embed-v3-english

Cohere-embed-v3-multilingual

Azure AI Vision multimodal embeddings (through a Foundry resource)

However, for the Cohere models to appear in the wizard, they must be deployed as a Serverless API deployment. If your Cohere model is deployed in a different mode (batch, managed endpoints, or any non-serverless option), the wizard will not surface it for image vectorization, even if the deployment exists in your project.

How to Fix this is If you don’t see your model available:

• Re-deploy Cohere-embed-v3 as a Serverless API deployment
• Then return to the wizard and repeat the Vectorize your images step

2. Ensure region compatibility for Cohere in Foundry

Cohere models on Azure AI Foundry are only available in specific hub regions East US, East US 2, North Central US, South Central US, Sweden Central, West US, West US. If your AI hub/project (where you deployed Cohere is in an unsupported region, the deployment won’t be usable. Create your AI hub in one of the supported regions and deploy the model there.

3. Grant access from Azure AI Search to the model deployment

The wizard builds a skillset that calls your embedding endpoint. It needs permission:

• RBAC: Assign your Search service’s managed identity a role that can invoke the endpoint for Foundry deployments, use the recommended role in the quick start; for Azure OpenAI, “Cognitive Services OpenAI User”.
• or API Key: If you use key‑based auth, supply the key in the skillset created by the wizard.

Lack of permissions can make otherwise correct deployments do not show up or fail during selection.

What likely happened in your setup

• You have gpt-4 (chat), text-embedding-ada-002 (text), and Cohere‑embed‑v3‑multilingual in your Foundry project.
• On the “Vectorize your images” step, the wizard looked for serverless image‑capable embedding deployments it can call.
• If your Cohere v3 deployment isn’t “serverless API”, or your hub is in an unsupported region, or Search doesn’t have permission, the wizard shows “No deployments available…”

We can resolve the issue by following the steps below.

Verify deployment type In Azure AI Foundry > Your Project > Deployments, check your Cohere deployment. If not Serverless API, redeploy Cohere‑embed‑v3‑multilingual as Serverless API.

Check hub region Ensure your AI hub is in one of Coherer’s supported regions e.g., East US or West US 3. If not, create a new hub in a supported region and move the project/deployment there.

Grant access In Azure AI Search, enable Managed identity and assign the necessary role to the Foundry endpoint (or configure key‑based access in the wizard’s skillset). The quick start documents the role mappings and keyless connections.

Run the wizard again Azure portal > Azure AI Search > Import data (new) > choose your Blob data source > on Vectorize your images, select your Cohere v3 serverless deployment or Azure AI Vision multimodal if you prefer Vision. The “No deployments…” warning should disappear once a valid multimodal deployment is detected

Reference:

https://learn.microsoft.com/en-us/azure/search/search-get-started-portal-import-vectors?tabs=sample-data-storage%2Cmodel-aoai%2Cconnect-data-storage%2Cvectorize-text-aoai%2Cvectorize-images

https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-configure-vectorizer

https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-generate-embeddings?tabs=dotnet

https://docs.cohere.com/docs/cohere-on-microsoft-azure

Kindly let us know if the above comment 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.

Hi,

Thanks for the info.

For the "Re-deploy Cohere-embed-v3 as a Serverless API deployment", I'm following the link https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/deploy-models-serverless?view=foundry-classic&tabs=azure-direct&pivots=ai-foundry-portal#prerequisites, but I'm not seeing the option in a project created in AI Hub:

Hi @Zhang, Chong,
Thanks for sharing the screenshot. Based on what it shows, you’re in the Preview features section of Azure AI Foundry. This panel only contains toggle switches for different preview capabilities for example: AI-assisted safety evaluations, Azure OpenAI Assistants API, Navigation with recent context, Recommended model deployments during project creation

All of these options are enabled in your view. However, this panel does not include anything related to choosing a deployment type, such as Serverless API. Because of that, you won't find the “Serverless API deployment” option here, which explains why it’s missing when you try to redeploy the Cohere model.

Azure AI Foundry doesn’t provide a setting in the Preview Features panel to manually choose a deployment type. Instead, the deployment type is automatically selected based on the model and the deployment flow you’re using.

For Cohere models, the system normally defaults to a Serverless API deployment. However, in certain situations such as older hubs, legacy workspaces, or customized configurations the platform may fall back to a different deployment type like batch or a managed endpoint.

This is why you may not see the Serverless API option even though all preview features are enabled.

What you need to do:

Verify your existing Cohere deployment:

• Go to Azure AI Foundry > Your Project > Deployments.
• Open your Cohere deployment and check the Deployment type.
• If it shows Batch or Managed Endpoint, it will not appear in the redeployment wizard as a serverless option.

Redeploy the model using the correct workflow:

• Open the Model Catalog and select cohere-embed-v3-multilingual.
• Start the deployment from here. The option for Serverless API appears only during deployment creation, not in the Preview Features panel.

If you still don’t see the serverless option, it may be due to: Your hub being in a region that doesn’t support Cohere models, or the feature not being enabled yet for your subscription.

Check your hub’s region: Cohere models are supported only in the following regions: East US, East US 2, North Central US, South Central US, Sweden Central, West US. If your hub is in a different region, you’ll need to create a new hub in one of the supported regions to access serverless deployments.

Validate permissions: Make sure that Azure AI Search (or any service calling the deployment) has the appropriate permissions either RBAC or key-based access to interact with the model deployment

Reason for the wizard shows “No deployments available”:

The Cohere model you deployed is a text embedding model, not an image embedding model. The “Vectorize your images” step in the wizard only looks for deployments that can generate image embeddings. Because Cohere doesn’t support image inputs, it will never appear in that list even if the deployment is fully configured and working.

If your goal is to vectorize or search images, you’ll need to deploy an image-capable embedding model from the Model Catalog, such as CLIP, GPT-4o-mini embeddings, or any other model designed for image embeddings.

Kindly let us know if the above comment 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.

Hi @Zhang, Chong,
I'm just reaching out to see if your issue has been resolved or if you've had a chance to review my previous comment?

Hi,

Thanks for the detailed response.

I deployed an OpenAI-CLIP-Image-Text-Embeddings-vit-base-patch32 model. But it's still not showing up in Azure Search flow for "Vectorize images".