[jjwiseman]

This is a super cool project. But it would be 10x cooler if they had generated CLIP or some other embeddings for the images, so you could search for text but also do semantic vector search like "people fighting", "cats and dogs, "red tesla", "clown", "child playing with dog", etc.