Same idea, different forms.

OmniSearch shows how multimodal embedding models build a vector space that holds different media types together. Because these four objects all describe the Sahara, they end up close to each other in that space, sharing similar semantic meaning.