I don't normally use diffusion models, but I had a dream about a new type of diffusion model that could generate perfect text.
It generated long and detailed texts by segmenting the image and separately generating each letter one after another, akin to inpainting each letter.
Now I'm curious if there's any model that actually works like this.