How do I store just the vector not the image


I am building a image search engine with text queries.
I am planning to have two fields in my img collection: id and img_url. I’m using multi2vec-clip model and currently feed each image as base64 encoding into the database. I’d like to store just the vector representation of each image and get the most similar entries.
Could any of you legends give me some pointers? Much appreciated.

Server Setup Information

  • Weaviate Server Version: v1.23.9
  • Deployment Method: docker
  • Multi Node? Number of Running Nodes: 1
  • Client Language and Version: python v4

Any additional Information

One possibility would be not configuring a vectorizer for the collection and doing your manual vectorization of the image and just store that (along with the other data you need).

Hi @lored,
If you are using multi2vec-clip then Weaviate will also store the base64 representation of the image.

The only workaround is to generate the vector manually, as suggested by @rjalex.

Github issue

I’ve created a GitHub issue with a similar request. Please upvote, as that might help move it closer to the front of the queue :wink:


thanks guys, I just created another collection, imported my images there and copied the vectors over to the vector only collection.

1 Like