Re:DALL·E: Introducing Outpainting
by bergenty on Wednesday August 31, 2022 @06:30PM
(#32669183)
But aren’t the results from stable diffusion not nearly as good as DALLE2?
[ Parent
| on HN ]
You don't have to drive SD with a text prompt; if you take the model weights you can feed anything you want in there and generate an image out of it.
Since DALLEmini and DALLE2 are more "creative" (since they use a better text transformer) you can use them to generate the input and SD to refine it for more fine detail.