AI Art

USING STABLE DIFFUSION

✧ AI components installation guide and links ✧ https://www.techspot.com/guides/2590-install-stable-diffusion/
        Github download link foranything-v3.0:  https://huggingface.co/Linaqruf/anything-v3.0/tree/main
        Model to create outpainting/ inpainting: https://huggingface.co/runwayml/stable-diffusion-inpainting creating a background around a small picture or inside it
        Model to create new pictures based on your character: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Textual-Inversion creating your own mini model that you can reference in a prompt

✧ Understanding AI settings ✧
        General settings: https://youtu.be/Z3IHmdqUar0 what each setting does (for txt2img/img2img pages)
        UI Sampling methods: https://youtu.be/Oq_YUIBFewg (DDIM/Euler etc) what are the differences

✧ Prompts/Tags for image creation ✧
        Tag Groups Wiki | Danbooru: https://danbooru.donmai.us/wiki_pages/tag_groups - useful also to click on any image or search by tag, then understand how tags work
        Tag Effects | Pastebin.com: https://pastebin.com/GurXf9a4 explains some prompts/tags and how and how heavy they can change things the AI than generates
        NovelAI Anime Girl Prompt Guide: https://lunarmimi.net/freebies/novelai-anime-girl-prompt-guide also includes poses
        More promt examples: https://youtu.be/lFI8JQvPfu8
.      VAE:  https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.safetensors does help make pictures vibrant and less saturated, and also solve weird spot issues

Note on prompts: The AI ​​Prompt cannot take unlimited arguments. You might think so, since there is no character limit, but some arguments are just ignored after a certain length.
The prompt can contain up to 75 tokens. You can calculate how many tokens make up your chain of arguments on the following website: https://beta.openai.com/tokenizer

In case if u dont have god graphics card: comment by @Azu: if you or others ever want to experiment but don't have a great graphics card you can hire a GPU with a cloud service for very very cheap: https://www.runpod.io/blog/stable-diffusion-ui-on-runpod 50 cents/hr for a 3090. There are also 'distilled' models releasing soon that will run much faster and might be better for older GPUs

✧ Useful to know/think about: ✧
    Art jargon and terms for your prompts (as example): https://youtu.be/46uVD97bkHQ (photorealistic / steampunk / intricate / render / high contrast)
    Fundamental Art Styles: https://youtu.be/aBYrgvE6fc4 some hints to understand what outcome to expect using art style prompts (indian / renaissance / baroque / rococo etc)
    Historical artists: https://youtu.be/AGZFUcsehJQ for understanding art styles
.
✧ Setting notes while creating an image ✧

- Check image dimensions setting - Width/Height - or image will be squished
- How'd you get that nice mature face instead of lime anime waifu face? - I'm using DDIM cause of that, if u generate a image with the same seed on DDIM and Euler, Euler will give a less mature image,
   more kid like image, DPM++ SDE also generate nice mature faces, but it takes a lot to run, so I'm rocking DDIM, and adding {{teen}}, {{kid}} to the negative prompts

Note from @kcnkcn for people seeing this through pins, i suggest you scroll up to see another take on settings.
   there is no right solution and maybe other settings work better for you
   so here's my take on your settings:
- i recommend Euler (not A) as a beginner because it's pretty consistent with it's outputs. A has a lot of variation, so as a beginner it might be difficult to understand where you went wrong
- iirc, euler doesnt change tooo much if you increase sampling steps. Unless you really want very detailed results, 30-50 is good because of generation speed. fun fact, NovelAI uses Euler at 28
- for denoising strength, i recommend you start high (maybe 0.4-0.6). if you like an image, drag it back into the img2img with a lower denoising strength for fine tuning
- personal preference, but i prefer random seed (click the die/dice) so you get a little more variation with your results and you dont get almost literally the same image every time
- increase batch count and batch size to increase the number of results per generation attempt. itll allow you to see different results, at a cost of generation time
- anythingv3 is strongly trained on danbooru tags, so visit the danbooru website, ignore all the r34, and figure out what tags are used. if you want a helping hand, click the "interrogate danbooru" button and itll return some tags that it thinks your input is using.

type the tags in the format of tag1, tag2, tag3, .