Dolly

Dolly takes the same tex includer from clip, so that that's just fixed. We don't touch it. And then we take this image feature, just another list of numbers, and then we use the diffusion models to turn it back into an image. So it is kind of this amolti stage process. It reuses a lot of the machinery from clip. But again, like gans, do have certain kind of aspects which make them very appealing. They're like, very, very fast.