That if you have 3 separate models, text -> text, image -> text and text -> image you can just glue them together to make it behave like a multimodal. (Just li

That if you have 3 separate models, text -> text, image -> text and text -> image you can just glue them together to make it behave like a multimodal. (Just like gpt4 is rumored to be a few different sub models and not just one giant model)