Stable Diffusion 2.0 Quickstart (webUI, local installation etc.)

It's come so far that even Mo couldn't ignore AI any longer, so he reluctantly started diving into diffusion models. When he emerged a month later (and very unkempt) this is what he found out.

This tutorial covers installing Stable Diffusion 2.0 using Automatic1111's webUI, prompting, creating images from images and fixing a broken script with two lines of python. Also included are ramblings about AI in general.

Please let us know in the comments if you'd like to see more AI focused content.

Stable Diffusion webUI:

Stable Diffusion 2.0 Checkpoint:

Config File:

Prompt used for Artwork:
img2img with this image as start:

Prompt:
an abstract minimalist sculpture with the surface of a space shuttle, heat shield, spacecraft details, soyuz, apollo, mercury, gemini, white background, bright, insulation, symmetric, very symmetrical body, tubing

Negative Prompt:
mobile, alexander calder, piet mondiran, jean miró, concept art, asymmetry, banal, mangled, unrealistic, decal, gauge, display, rough surface, glossy surface, blurry, unsharp, out of focus

00:00 - Intro
01:33 - Setting up a new user
02:32 - Installing WebUI
02:58 - Installing Python 3.10.6
03:30 - Installing Git
03:48 - Downloading WebUI using Git
04:58 - Downloading Stable Diffusion 2.0 checkpoint
06:07 - Config File
07:00 - Launching WebUI for the first time
07:46 - Quick guide on parameters
09:55 - My first prompt
10:16 - How to get initial prompts (CLIP Interrogator)
11:54 - Refining our prompt
14:28 - Using images as prompt: img2img
16:44 - Loopback script
18:06 - Issues with the loopback script
18:50 - Fixing the loopback script
20:34 - Restarting & loading new script
22:05 - Infamous last words

[alistairbuchan]

Super useful brief intro into stable diffusion and its potential as part of an artists toolbelt, rather than a replacement. Good level of depth for a taster too. On to your portraits with Stable Diffusion video!...

[alwayssomewhattired]

This is a fantastic video! Thank you so much Entagma

[roybatty]

Houdini + latent diffusion integration + PDG etc would be great. Also you can go into settings and change the slider that will allow the image to be updated in the UI every so many iterations so you can see your progress.

[ShariSolo]

Thank you Mo. Just love your content. Need popcorn with it.

[correctopinion]

13:55 you can also go to settings user interface and show image creation process: 1

[kasali]

a couple of years ago Mr. Horikawa trained pix2pix AI with floor plans generated in Houdini. There must be some way to train StableDiffusion as well. If you could make this "bridge" between Houdini and SD, that would be fantastisch

[chillsoft]

Smaller GPUs should use InvokeAI instead! :) With that said, I just started my first 3D / Houdini job where I'm using a custom A1111 SD workflow for concepts, and the client is superhappy! :) :D Nice to see it get some traction! I've been playing with these things since April (not specifically SD but Image diffusers)!

[habib.bhatti]

I feel AI will become standard Houdini built in plugins. AI is revolutionary. It was the Industrial Age, then the Space Age, then Information Age, then AI age now, and then Mo age

[sergheibac]

Thank you for your video it helped me to install SDiff, but I don’t understand why you got nice images from the beginning and I got ugly and not clear images with the same promt and model. Do you have any ideas?

[Ganexcg]

anyone, if you are getting 9009 error, unistal Python and set add python to the system path setting at the beginning of the setup .

[habib.bhatti]

This is essentially like a mutation generation tool for renders from any 3d graphics design software

[tomo-yiow]

question🙌🙌
Can I do the exact same thing as in the video and can expect it to work with AMD gpu?

[RonaldFong]

Hey Mo, or anybody out there actually, what's the implication of installing Python 3.10.6, and adding it to PATH for the main user? I gather that Mo was trying to avoid messing up his main user account, but I'm wondering if it's mostly a developer concern, and won't affect normal CGI artists? I wish to install SD on all my studio's machines, but I'm afraid it'll affect existing software configurations like Redshift/Houdini/Nuke/Maya. I hope this is not a silly question, thanks in advance!

[wonderland]

Thanks for the guide! So this does take a lot of resource e.g vram, cpu, etc from your PC actively right while running, or just during generation?

[misa]

when I open webui-user.bat I get this error
No Python at

it is solved by installing python 3.11 but if I install python 3.11 it breaks a different part of the program

[nicksushkevich]

git installation gave me a dozen options to choose from on Win10

[rizkiramadhan]

I went through everything without a problem until I had to install WebUI. It says I could not install torch. What should I do?

[vicioustide]

I assume this will actually run from GPU without intensive interactions on the users side? Unlike with InvokeAI currently?

[cruxial]

How hard would it be to use "custom" models such as anything-v-3 or AnimeDiffuision with this approach?

[zahiranazer]

hi, thanks for the video, it was very useful, i followed the steps but webui didn't launch, it left the message exit code 9009 on command prompt and couldn't launch python.