Stable Diffusion FAQ

This is a FAQ for Stable Diffusion on Automatic1111, ComfyUI, and whatever else you make.

• The idea is to collect basic wisdom in a readable way. Why here? Well, I didn't see it happening anywhere else 🤷‍♀️ - right now mostly it's to collect resources and answers from and for folks on the Civitai discord server.

1. If you are a complete beginner

• https://education.civitai.com/civitais-100-beginners-guide-to-generative-ai-art/

2. Need more than this FAQ?

• First, consider the NOOB FAQ: https://civitai.com/articles/530

• Next, try https://civitai.com/articles - there's a lot of stuff in there and this is just getting started.

Helping

Just LMK you want to help - pgm

Automatic1111

Prompts

Can I use anime and realism models and checkpoints and whatnot together?

• ...maybe. You'll have to try it out, but keep in mind they may not have been tested for that capability.

Safetensors files, what even are they? Can I look inside?

• The first formats for files to store and transmit loras, checkpoints, and so on were not very secure. Thus came safetensors.

• Yes, you can look inside, try this. https://github.com/by321/safetensors_util

• Also, kohya_ss comes with a command line tool for safetensor metadata extraction, tools/show_metadata.py

• There is also a file format spec: https://huggingface.co/docs/safetensors/metadata_parsing

What are these Samplers of which you speak?

• https://stable-diffusion-art.com/samplers/

also

• https://www.youtube.com/watch?v=JAMkYVV-n18

Can I restore old photos with Stable Diffusion?

• Yes! Here's an instructional video: https://www.youtube.com/watch?v=dHQ4rkAGghs

How does Stable Diffusion Work?

• You'll need to learn a lot of stuff to know it all the way down (like, to electrons) but here's a nicely illustrated higher level explanation: https://jalammar.github.io/illustrated-stable-diffusion/

Can I colorize old photos?

• Yes, try looking at this: https://civitai.com/models/80549/color-based-picture-control-controlnet

Whatever tf is a VAE?

• https://www.reddit.com/r/StableDiffusion/comments/zamt9y/what_is_vae/

SDXL

Perf??

• https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Optimum-SDXL-Usage

My first SDXL images are coming out blurry?

• The default sampler, Euler A, can make this happen. Choose a different sampler. If the image is more mis-colored than blurry, check that you are using an SDXL-specific VAE.

Controlnet for SDXL?

• Yeah apparently, see https://www.youtube.com/watch?v=5uz-HmN36ao

• And maybe look here https://huggingface.co/SargeZT/controlnet-v1e-sdxl-depth

Can I run SDXL on my VHS / Commodore 64 / Timex?

• I really wish you could.

Where do I find these Upscalers? There are so many?

• I dunno try this, I thought the wiki was better but they didn't ask me and moved: https://openmodeldb.info/

Is AI a problem? Are AI generated images harming artists?

• Excellent questions outside the scope of this FAQ. Perhaps we will have good starting locations to learn about and discuss these questions linked here in the future.