Effect of Image Size on Stable Diffusion
As for whether the sampling method is DDIM or PLMS, there is some difference, but it is not so obvious to the human eye
https://gyazo.com/0ed967f2faddaba3bb4b3868949df758https://gyazo.com/aa6d3fed9f7d97b07a877b23ea94c4ac
Rather, the effect of size is so great that it's a completely different thing.
https://gyazo.com/5c593cd3ca14658dd22190e76bffc88d
https://gyazo.com/2ae797fdd1a543c059438c06bf86b94f
https://gyazo.com/86e6ad3429432b611f572eb83a41845e
https://gyazo.com/0ed967f2faddaba3bb4b3868949df758
If the composition is the same even if it is small, I tried the "generate many at high speed at a small size and then select the best ones for higher resolution" approach, but it didn't work at all.
LDM in Stable Diffusion does diffusion model in latent space
Not a super-resolution (e.g., video) approach after diffusion modeling in the space of small images
Perhaps that's why.
---
This page is auto-translated from /nishio/Stable Diffusionの画像サイズの影響 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.