This is a tutorial made for the AI HUB Discord Server, Go check them out for an AI Centered Community.
If you have any questions, please notify me on the discord so I can add it to the article or answer it for you.
1. Prerequisites
To begin generating with AI you have to have a device that can do such tasks.
Below are some services/applications you can use to begin generating
THE LINKS WILL LINK YOU TO THE INSTALL INSTRUCTIONS PAGE
Windows (Free / Open Source)
AUTOMATIC1111 - Requires a Beefy GPU, NVIDIA/AMD Recommended, Beginner Friendly
ComfyUI - Also requires a Somewhat Beefy GPU, Not Beginner Friendly
Stability Matrix - Tool to easily install Stable Diffusion User Interfaces (A Great Place to start!)
Mac (Free)
Draw Things (Found in app store) - Mac/iPhone/iPad optimized interface for Stable Diffusion
AUTOMATIC1111 - Not optimized for Mac M1 - M3, Same as with windows version
Linux (Open Source)
Uses the same apps as on windows, Choose whichever one you want
iOS/iPadOS
Draw Things (Found in app store) - Same as the Mac Version
Android
AI-Art-Generator (Advanced Users Only)
Recommended to use online services (As of now)
Online Services (Free / Paid)
Bing AI - Uses the new DALLE-3
ChatGPT DALLE-3 - Paid currently, Rolling out free to select users
Google Imagen - Integrated with Google image search
This website as well!
1B. How to set up and use (Does not apply to online services)
Prerequisites currently require you to have Python 3.10 installed. Any other version could still work, but most likely will end up not working after an update or other factor. Stability Matrix installs Python for you and sets up everything for you. If you are up for the challenge, installing manually is an option.
NVIDIA GPUs currently are the best for any type of AI image generation as they have CUDA. CUDA is NVIDIA's proprietary machine learning architecture built into most GTX/RTX/Quadro graphics cards. CUDA also gives a massive performance boost over AMD/Intel.
Above I have linked all of the install guides for the respective UI's although Draw Things does not have an installation guide.
To get started with Draw Things, make sure you have at a minimum an iPhone X (New SE model is incompatible). The way Draw Things is possible is because of Face ID and how it uses machine learning to learn your face, so Apple put AI cores into most all modern flagship phones.
On the first boot up, you will be prompted to install a few things, make sure to install them. After installing and getting to the main screen, technically you are ready to start generating. At the top is your prompt and swiping over where the arrow is on the right is where your negative prompt is. (SEE BELOW FOR PROMPTING HELP)
To install more models or change settings like resolution on iPhone it's the button on the top right. Draw Things is very descriptive with each of it's options and very easy to understand.
2. Getting Models & Setting Up
This portion of the guide is only for the following: AUTOMATIC1111, ComfyUI, Draw Things, AI-Art-Generator
If you chose to use Stability Matrix to install your UI, it can be used to install models as well. Go ahead and use Stability Matrix, Civitai, or HuggingFace to find a model you want to use.
One thing most models have on their download page is a list of recommended sampling parameters like so:
These parameters were chosen by the model creator because they give the best output, your preferred UI may have different names for things like Hires. Fix or Sampler. If you can't find those options, you can skip them if you so choose to, although it may result in different images than the ones on the model page.
3. Prompts and Negative Prompts
This portion of the guide is only for the following: AUTOMATIC1111, ComfyUI, Draw Things, AI-Art-Generator
Prompting is single-handedly the most important part of AI, it defines what you want to generate.
Prompts are split up into 2 parts: Positive Prompts & Negative Prompts.
Positive Prompts are what you want to have generate in your image (Aka: a beautiful mountain scene with a sunset)
Negative Prompts are what you don't want to show up in your prompt. This also defines how the image looks like. For most models this: (worst quality, low quality:1.4) works the best.
The 2 Primary Stable Diffusion Models that exist are Realistic & Anime, they both prompt differently.
Realistic Models have prompts that describe a scene with sentences or short phrases: (Aka: a cowboy driving a truck on mars)
Anime Models are prompted with booru tags (Aka: 1girl, red hair, mountain scene)
For Bing AI and GPT 4, you can prompt the same way, just If you want an anime or cartoon style, make sure to put cartoon or anime somewhere in the prompt.
4. Setting up your respective UI
AUTOMATIC1111:
On first launch once you are loaded into the Webui, you can start generating right away. Although, you shouldn't. What you should do first is configure a few settings to get the best possible results with your generations and speed up them as well.
Starting off, once you've first booted up into the Webui, close it and the console window as well. Navigate to where you installed your UI, and find the file named webui-user.bat / webui-user.sh
Right click either of the files and edit it with notepad or any other text editing program.
The only line we want to focus on right now is the line: set COMMANDLINE_ARGS=
These are launch arguments to tell what the program should do differently or install.
Below are some recommended launch arguments to set for a few select systems:
Core iX 8 - 14 Gen / Nvidia 16xx / 20xx / 30xx / 40xx 8-24gb (NO MOBILE)
set COMMANDLINE_ARGS=--medvram --no-half-vae --xformers --opt-sdp-attention
Amd Ryzen 5 - 9 3rd gen / AMD RX (Any modern one is suitable as long as vram is 8gb or more)
set COMMANDLINE_ARGS=--medvram --backend directml --no-half --precision full --opt-sub-quad-attention --opt-split-attention-v1 --disable-nan-check
Relaunch your UI afterwards. (It may take some time to install every thing or change things around)
If you get the error CUDA Out of Memory or another one like that. Try turning down resolution settings or turning off Hires Fix. Many other errors can show up where google searching doesn't help. If so, send your error to the AI Hub discord, we're happy to help
One thing to help with making your images more balanced is turning your CLIP Skip to 2.
To do that, navigate to settings, and find the User Interface section. Scroll until you find this section:
Click the dropdown, and add the CLIP_stop_at_last_layers. sd_model_checkpoint will already be there.
This is a very basic setup guide for AUTOMATIC1111 and you can do so much more with it. If you'd like more out of it check out the Extensions tab. There's loads of useful and fun extensions to help make your experience better.
Make sure to remember how prompting works, and you'll get great results!
ComfyUI: (Help Wanted)
5. Loras / Textual Inversions / VAE
WIP (For now look at other guides) :P