Any guidance on creating smaller images? 256x256 or 384x384?

[MatthewWaller]

Is your feature request related to a problem? Please describe. I’m looking for ways to speed the process and save memory

Describe the solution you'd like I wondered if it is possible to create smaller images.

Describe alternatives you've considered I tried setting the images size with poor results out of the box. Thoughts?

[Lewington-pitsos]

I spent like 60 minutes googling this yesterday, I checked a bunch of different forums and the consensus seems to be that due to the fact that SD was trained on 512x512 resolution, lower resolutions will not work.

To achieve this I think you would have to re-train or at least read the code and work out something that's not very obvious. I can confirm that there are no guides on doing this at present.

One sneaky trick I tried was "four separate images of a forest" and it generated like 4 images in a 2x2 grid, but that's crazy janky.

[patrickvonplaten]

As suggested by @patil-suraj, the v1-1 might work better for 256x256 as it was trained on it: https://huggingface.co/CompVis/stable-diffusion-v1-1

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[elephantpanda]

I've tried 384x384 and not got bad results. 256x256 though is bad.