Simply amazing topic, and beautiful sea creatures!
I don't know how this AI from Facebook is, but I can share some knowledge of why the VQGAN+CLIP circles like a champ. The language model si totally able to say something well connected and related to its previous "comment". But that's all, so many times, along a large text the AI just seems to divagate and at end is saying things that has nothing to do with the topic it started. Other times circles and circles coming back to the same, because it forgot "what said at the beginning.
For what I saw in some videos, next language models gonna come trained in categorize the words, by syntaxy (like pronouns, verbs, adjectives and so), and also to give them a role in a particular "conversation". For example, if it starts talking about in Canary islands there are awesome bananas, it gonna label it like "the topic"... and gonna stick to it, unless the conversation truly takes another topic. So the divagation gonna disappear in a future.
And as in VQGAN+CLIP, the graphical "divagation" comes from the language model, it gonna stop happening as well.
Regards, and grab this !PIZZA!
RE: Whales and Dolphins therapy with AI (IC-GAN ) : The fift dream legend