AI Art General

[NEG]

So, I am not an artist, but I can make art easily. How? AI.

Paid services where almost everything is done for you:
https://novelai.net/(Anime only)
https://www.midjourney.com/ (Discord only, no privacy, strict content policy)
https://www.artbreeder.com/ (mostly face generations, limited Stable Diffusion)

Detailed info (lazily copied from /g/):

► Show Spoiler

>News
https://rentry.org/sdupdates3

>Local install
Nvidia GPU: https://rentry.org/voldy
AMD GPU: https://rentry.org/sdg-link
CPU: http://rentry.org/cputard
Alternative UI (Node-based): https://github.com/comfyanonymous/ComfyUI

>Run cloud hosted instance
https://rentry.org/sdg-link

>Try online without registration
txt2img: https://huggingface.co/spaces/stability ... -diffusion
img2img: https://huggingface.co/spaces/huggingfa ... e-the-rest
Inpainting: https://huggingface.co/spaces/fffiloni/ ... inpainting

>Models, textual inversion & embeddings
https://civitai.com/ (custom stable diffusion models, very useful)
https://huggingface.co/

>ControlNet
https://rentry.org/dummycontrolnet

>Index of guides and other tools
Want to know how to use SD for animation, upscaling, train your own model or merging model checkpoints?

Or need guidance on prompt ideas and when to use which artist then follow these links...

https://rentry.org/sdg-link
https://rentry.org/rentrysd

Quick and dirty quick start guide without reading a dozen rentry's (provided you have an Nvidia GPU with at least 8 or so gb of vram):

► Show Spoiler

1. Go here and follow the installation guide.
If you don't want to mess with git pulls, you can just go to the "Code" button at the top and download the zip. Unzip to a folder near the root directory. E.g., C:\SD\
2. Get a safetensors model from https://civitai.com/ (Why safetensors? Because pickle models could have malicious code.)
3. Copy the prompts you see on the examples of whatever model to get an idea of how it works.
4. Generate.
5. (Optional) Use a VAE to improve generation quality: https://stable-diffusion-art.com/how-to-use-vae/

NEG's 30-Second AI Art Masterclass:

► Show Spoiler

1. Download Krita
2. Go here and follow the install instructions.
3. Generate from within Krita or edit existing generations.
4. Use img2img to edit by feather selecting parts of the image and feathering by 5 to 15 pixels. Masks will be applied automatically and blend the new image area on top of the old one. Generating batches several images is recommended so that you can pick the best one.
5. Never upload another image with messed up eyes or hands again.

► Example Of Fixing Someone Else's Bad Eyes In Krita

[GhostCow]

You can run novelai locally for free, but I've not had good luck with any of the models that other people have made. I'd really like to make my own, but it seems like a ton of work. It would be a lot easier if I had a script to download images from a site like gelbooru that would also create a .txt with all of the tags.

https://kiwifarms.net/threads/stable-di ... t-13052428

[Shillitron]

You can run novelai locally for free, but I've not had good luck with any of the models that other people have made. I'd really like to make my own, but it seems like a ton of work. It would be a lot easier if I had a script to download images from a site like gelbooru that would also create a .txt with all of the tags.

https://kiwifarms.net/threads/stable-di ... t-13052428

LORA is the poor-mans training.

With a relatively modest dataset. You can train a LORA over an established Model (SD1.5, NAI, Anything3 or 4.5 are all popular ones) with really any kinda-goodish video card. (It has a Low-VRAM option that makes training a bit longer but can run on daisy chained potatos it seems)

They seem to be very sensitive to 'over-training' but that's offset by applying a weight when you prompt it which is becoming the defacto standard now anyways. (Running a LORA at 1.0 generally is considered a very aggressive weight, most recommend 0.3 - 0.8)

LORA's are storming the community because they:
- Are easy and fast to train
- Small in Size
- Highly effective to influence a model with a very specific artstyle or concept
- Easy to inject (You can reference multiple LORA's in a single prompt and combine them without needing to merge, usually up to 4 before things get wonky)


This environment is constantly shifting so maybe in a month LORA's will be dead.. but they've seem to completely overtaken Hypernetworks in popularity and people are even able to extract LORA's out of checkpoints fairly effortlessly as well.

[NEG]

You can run novelai locally for free

It's an older version of the model, can only run effectively using Euler A unless you clone their site (which means using something other than Automatic1111), and tends to generate small color errors, depending on settings.

There's also no reason to use it, since there are similar models that meet or exceed the quality level of the pirated model.

[Klerik]