About Stable diffusion Generative AI upscaling

This post is a continus discussion from




The image above is upscaled using one of the A1111 extension called StableSR.

Stable Diffusion checkpoint

SR model

PNG Info
Steps: 60, Sampler: Euler a, CFG scale: 7, Seed: 872426874, Size: 800x904, Model hash: dcd690123c, Model: v2-1_768-ema-pruned, VAE hash: 495c9925a5, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Denoising strength: 0.5, ENSD: 31337, Version: v1.6.0

You have to install the StableSR extension first, in the A1111 extension tab.

After you installed the extension,

You can put the SR model into folder

Then select it under SR Model.


ok I see.
For the moment I only have this folders here:

If you don’t see the folder, I think you haven 't install the extension yet.

Go to extensions tab → Avaliable (Load) → search StableSR → install
After that, click Apply and restart.
Then reload the UI or restart the A1111

That’s it, I’ve installed the extension and placed the template. Now I just have to do what your screenshot shows.

Yep, that’s what I did. I even went to the section to type in the link to download it.

Then you should able to find “sd-webui-stablesr” within Extension folder.


yep. I have it on file. And I’ve put the model in the model folder as planned :wink:

Great ~! :grinning:

Joining in! Up until today I had not bothered testing upscaling within the SD app I use (Diffusion Bee, Mac). Here are some comparisons (text prompt: “rich men north of richmond”):



6X with PAI:

First upscale in Diffusion Bee:

Upscaling the upscale in Diffusion Bee:

In this case PAI was best. I will check out the SD alternatives described above.


00000-v2-1_768-ema-pruned_(Euler a)_872426874-CS1

If you don’t mind make a slight change to the face, it can get better quality.

00013-v2-1_768-ema-pruned_(Euler a)_872426874-CS1

Very nice! Can this be set up on Mac and without too much hassle?

I don’t know about Mac :thinking:

I am using Windows 11 :grinning:

I guess I will Google… :wink:

Maybe not too bad:

My first render! Worked great on Mac so far, thanks for the heads-up.

OK, I need some help with adding options to upscaling such as you described.

I seem to have the extensions loaded but cannot find them in the menu (unless they are called something else there?):

Screenshot 2023-09-24 at 4.54.42 PM

Screenshot 2023-09-24 at 4.55.06 PM

Otherwise, the existing options seem to do quite well (SwinIR 4x here at 6X, chosen at random).

Screenshot 2023-09-24 at 5.01.48 PM

It also seems I must keep Terminal open while doing my business…

Screenshot 2023-09-24 at 5.09.22 PM

Interesting hair details on the enlarged image generated with Stable Difusion

Enlargement x6

For AI generated image, I think it is better upscale within Stable Diffusion, because it usually can generate more new detail.



Doesn’t look realistic, but not bad for a text to image.

1 Like

OMG :scream:.
The details are sublime. What settings did you use?
As far as I’m concerned, upscaling doesn’t do as well as Topaz Photo AI. Because the base image is generated in Fooocus, which is a simple interface based on Stable Difusion Automatic1111.

Super zoom your enlargement

You can almost see through your nose hahaha

Fooocus is easiler to install and simipler to use, however, I still perfer using Auto1111, because it give me more flexibility and many checkpoints / extension to play with. :grinning:

The image above is upscaled using one of the A1111 extension called StableSR.