Help with running Stable Diffusion locally
I hope this is the right section for this. I recently discovered Easy Diffusion so I've moved to running SD locally on my own PC, but I'm having some issues getting the most out of it. It's hard to find good articles online that describe in plain, simple text how to do much of anything; most of them boil down to "To do this, just do it" which is frankly beyond useless. I'm trying to find good, concise help for the following few things:
-A good refresher or explainer on how the features work. I've gotten to where I am by simply looking at other prompts by people and making my own by cobbling them together. It works, but I know it could work better if I actually understood what I was doing.
-A simple how-to for training your own models/LoRAs. Every article I've looked at was written for the benefit of machines, it seems like. They tend to either be the same "to do it, just do it" pieces or stuff written for those who clearly already know what they're doing.
Basically I've been using AI to generate commission references I can give to a proper artist later. It's hard to commission work of characters who haven't been drawn before, and I've had poor experiences in the past with submitting written descriptions. Either there's a language barrier or it leads to a lot of unnecessary back-and-forth with the artist making them tweak or redo things which is not fair to them. If I can just show a few picture references I've made, it ensures I get what I'm looking for and the artist has an easier time.
I've seen that @skyrbrian seems to be the subject matter expert, but thanks in advance to anyone who replies!
I use Easy Diffusion, and I had a similar issue of finding relevant documentation since terminology is different. Ultimately ,what I settled on is using information like Hammurobbie provided (https://stable-diffusion-art.com/automatic1111/) while keeping in mind that that documentation uses different terms. For example, in img2img mode 'Guidance Scale' is Denoising strength. In that article, just replace the terms.
To get good image references, meaning relevant to what you want to present to a commissioned artist the key is learning how to use Denoising/Guidance Scale, ControlNets (an opt-in beta feature for Easy Diffusion) and LoRAs. The general routine I follow for generating images with Easy Diffusion now is:
I realize not all of this might align with what you're asking, but this is effectively all I've got in terms of communicable Easy Diffusion knowledge. Everything else is what I've innately figured out by trial and error, but not to a point where I can express it to others.
It still could be helpful, thanks!
I’m not familiar with Easy Diffusion so I can’t answer your specific questions, but if you’ll permit me a moment to shill I’d encourage you to take a look at InvokeAI. I’m linking to their YouTube channel because it’s an informative and well-produced way to see what Invoke is all about. They’ve been putting a ton of energy into its UI and feature set, and the result is a really full-featured and easy-to-use system.
Writing good prompts and training models / LoRAs is an art that you can refine with practice, but there’s something to be said for having (and knowing how to use) good tools. IMO Invoke is one of the best tools available because of its unified canvas, node-based workflow editor, dynamic prompts, etc. I haven’t played with the latest version yet but I’m intrigued by the new batch queue feature too.
I appreciate the recommendation but I'll have to pass on it. I've already got a program with a UI that works fine for me that I can find my way around, the problem is finding someone who can explain in meaningful words how to use the tools properly.
This link will probably help. It's a beginner's guide to automatic1111's repo, but it explains most of the features quite well: https://stable-diffusion-art.com/automatic1111/
This is what I used as well. It was all very unfamiliar but the steps are well written and I got it working with no problems.
Hi, while I am not an the most knowledgeable on stable diffusion or ComfyUI(I am more of a a1111 webui user myself) but I can try and answer your questions if you can give some more information. What specific features are you hoping to learn more about.