I still don't really understand how the model is capable of this much compositionality: where you tell it to take a photograph and then open a door to a different image style. #stablediffusion, no editing

Aug 1, 2022 · 4:59 AM UTC

9
19
4
239
FWIW, this is a case where Stable Diffusion performs better than DALL_E. DALL-E does know who Maxfield Parrish is, but once it decides a picture is a photograph it wants the whole thing to be a photograph.
2
1
1
29
Replying to @Ted_Underwood
Ted: "I'm no artist" Also Ted: *makes incredible art* 🤌
1
8
Y'all are kind, but this is just an experiment. It's not even playing the same game that people like @KaliYuga_ai and @images_ai are playing now; they're pushing the boundaries of what we can do with this.
images_ai @images_ai
1 Aug 2022
"Stargazing"
3
8
more replies
Replying to @Ted_Underwood
Is this StableDiffusion inpainting already? or the prompt included the photograph and the doorway?
1
1
The prompts are in the alt-texts for the images. No editing, no inpainting. It can just do that. I've got to assume that the training dataset has some models of this kind, where a different image style is seen through a window or door.
1
2
more replies
Replying to @Ted_Underwood
A super promising model 💛 Waiting for a release to make a fight vs Dalle 🥋
Replying to @Ted_Underwood
These are really lovely.
1
Replying to @Ted_Underwood
@Yom_GuiTV Stable Diffusion got it's skills☝️
1
This tweet is unavailable
prompts are in the ALT text
Load more