AI Vision Vectorization in Promptflow

Hi @Seah Bryan,

Thank you for reaching out to Microsoft Q&A forum!

To effectively access your image search index in Promptflow, it’s essential to use the same AI Vision model for embedding your query images that you used during the index creation. This is because multimodal embedding converts images into vector representations that capture their features, and using different models can lead to compatibility issues.

When you vectorize an image with Azure AI Vision, it creates a specific vector embedding that allows for semantic similarity searches. Since vector search finds matches based on these embeddings, it’s crucial to ensure that the query image is embedded using the same model as the indexed images. This approach will enhance the accuracy and relevance of your search results in Promptflow, as it allows for precise comparisons between the vectors.

If you haven't already, make sure to integrate the Azure AI Vision API within Promptflow to generate these embeddings for your queries. This way, you'll maintain consistency across both the indexed images and your search queries, ensuring optimal performance in your image retrieval process.

For more info: Multimodal embeddings
Also refer to: Image retrieval using multimodal embeddings.

I hope this help. Thank you.

Hi @Seah Bryan,

Following up to see if the given response was helpful. Thank you.

Hi @santoshkc ,

Thank you for your response, and my apologies for the late reply. While I do understand the concept of using the same AI Vision embedding model to query against the search index, my question was pertaining to how exactly to access this model in Promptflow.

So far, I have found a way to access it by creating a Python node and using the SearchClient and VectorizableTextQuery classes to achieve, but it returns an error stating that AI Vision models are not supported in the current api.

This is my code:

from azure.search.documents import SearchClient
from azure.search.documents.models import VectorizableTextQuery
from azure.core.credentials import AzureKeyCredential

search_service = "<search_service_url>"
index_name = "<vector name>"
credentials = AzureKeyCredential(<search_index_key>)

search_client = SearchClient(endpoint=search_service, index_name=index_name, credential=credentials)
text_vector_query = VectorizableTextQuery(text="Stealing", fields="image_vector")

results = search_client.search(vector_queries=[text_vector_query], top=3)
for result in results:
  print(result)

This is the error:

HttpResponseError: (FeatureNotSupportedInApiVersion) Vectorization of queries against fields using the 'aiServicesVision' vectorizer kind is not supported in the current api version.