GETTING STARTED
(mostly written by AI)
Last updated: March 4, 2025*
Installation
Install Stable Diffusion Forge with One-Click Installer
What is Stable Diffusion Forge?
Stable Diffusion Forge is an optimized AI art tool offering faster performance and enhanced features compared to Automatic1111. It supports both SDXL and traditional models for high-quality image generation.
Prerequisites
- System Requirements:
- Windows
- 8GB VRAM minimum (For SDXL Models)
- NVIDIA GPU with CUDA support.
**AMD won't work with this method, AMD needs some extra install steps for webui, alternatively try this UI, supposedly optimized for windows and easy install (don't have amd so can't check how bad or good it is, if you try it feedback welcome)
https://www.amuse-ai.com/
Installation Steps
- Download the One-Click Installer
Visit the Stable Diffusion Forge GitHub page to download the simplified installer. - Extract and Run
- Extract the
.zip
- Run
run.bat
then once finished runupdate.bat
to finalize install - You should be good to go.
HOW TO USE UI
Using Stable Diffusion Forge WebUI for XL Models
- Selecting an XL Model
- In the "Checkpoint" dropdown menu, select the XL model you want to use.
- Ensure that the XL model file (
.safetensor
) has been placed in the appropriate directory (models/Stable-diffusion/
) beforehand.
- Setting Up Your Prompt
The prompt is the key to guiding the AI in generating the desired image. Here's how to craft effective prompts:
a. Positive Prompt
- For anime based models you want to use danbooru tags, and example prompt:
- Character and Artist: 1girl, ookami mio, noco (adamas)
- Quality tags (Verbiage you need in most modern anime models): masterpiece, best quality, newest, highres,
- Rest of Prompt: standing, waving, looking at viewer, shirt, shorts, white background,
So full prompt would be:
1girl, ookami mio, noco (adamas),masterpiece, best quality, newest, highres,standing, waving, looking at viewer, shirt, shorts, white background,
IMAGE:
- You can try sentences but mostly work on realistic models.
- Example: A cinematic shot of a futuristic cityscape at sunset, glowing neon lights, highly detailed, 8k resolution
b. Negative Prompt
- Use the "Negative Prompt" field to specify elements you don’t want in the output.
- Good default but add or remove as necessary: sketch, low quality, worst quality, bad anatomy, lowres, jpeg artifacts, artist name, worst aesthetic, signature, patreon username, bar censor, toned, muscular, extra digits, fewer digits, mosaic censoring
- Adjusting Parameters
To optimize results for XL models, tweak the following settings (varies for person to person depending on taste so experiment:
- Sampling Method: Choose a sampler and scheduler, good one to try is Euler, SGM Uniform. Try others depending on your preference for detail and speed.
- Steps: Set between 20-35 for most XL models. Higher steps improve quality but increase rendering time.
- CFG Scale: A value between 5-7 works well for balancing creativity and adherence to the prompt.
- Resolution: XL models perform best at higher resolutions. Some dimension people like to use 1024x1024, 896x1152, 832x1216`.
- Generating Images
- Once all parameters are set, click the "Generate" button.
- The WebUI will process your request and display the generated image(s) in the results panel.
RECOMMENDED MODELS
Illustrious Based Models
Illustrious-XL leverages the Danbooru2023 dataset , which contains a vast collection of tagged illustrations and character art. This allows the model to excel in generating detailed anime-style images
The dataset ensures that the model has extensive knowledge of characters and artworks commonly found on platforms like Danbooru
Some Notable Versions of this Model
- NoobAI-XL: This is the model with dataset up to October 2024 so all characters with at least 100 images on danbooru around that date you will be able to generate (same for artists who have at least 250 images)
- NTR MIX: Above model but merged to make it a bit more stable with anatomy and composition.
- WAI-NSFW_ILLUSTRIOUS: This is based of the base model Illustrious so the data is a bit older around June 2024 but is a fairly popular merge in the community.