Instead of including base64 images in your documents that, presumably, describe something to do with the images, store those images as separate documents.
Then, in your current documents, replace the image data with references to the images that should exist.
You can also use bindings to add a computed field to an index, if you really don’t want to split into another Collection as Ewan proposed.
@ridenaio Do you still have any questions on this, or do you have enough information to mark a post as a solution? I would be interested in learning more about how you decided to move forward with your data model.