How to get the same images that are generated from the SD web demo with Automatic 1111?

[erikestany]

Hi,

I use Stable Diffusion from Automatic 1111 (via Google Colab) for the options it offers to modify the generation parameters, but I'm often more interested in the aesthetics of images generated with the same prompts from https://stablediffusionweb.com/#demo .

I've tried generating images from Automatic 1111 with the same default settings in the web demo that can't be modified (prompt, steps, guidance scale, seed), but the results don't match. Since I couldn't find which model this demo uses, I tried the various model versions that can be loaded from the Colab (v1.5, v1.5-imp, v2.1-512px and v.2.1-768px ) and with the different sampling methods that can be selected from Automatic 1111 (Euler, LMS, etc.), but the images do not match either. Perhaps the closest ones are those generated with the v.1.5 model, but they are not the same either.

What am I missing? Is there a way to know which model is used in the web demo? Do you know how to get the same images that are generated from the web demo with Automatic 1111?

Thanks.

[erikestany]

Hi,

Any ideas on this?

Thanks!

[TheLastBen]

generating the same image from different interfaces even with the same seed can be tricky, you'll just have to try multiple settings.