Suisha is a stable diffusion discord bot.
Negative prompts, prompt weighting, expanded token input, VAE support, xformers, SD 1.x and 2.x support, highres fix support
Larges amount of customization via the config file
Blocked words list
Supports splitting queue amongst multiple GPUs. The GPUs don't have to be on the same computer or network.
Per role rate limits / per role DM access is currently under development.
Install requirements
pip install -r requirements.txt
Suisha uses Automatic1111's WebUI as the backend. Follow the directions on their repo for setup instructions.
Once you have WebUI set up, run webui.sh with the --api argument. You can also add other
arguments such as --xformers to use xformers memory efficient attention.
You can use the web ui interface that Automatic1111 provides to select the model and VAE to use. Their repo has documentation on how to do so. I also recommend doing a test generation
Create a file called .env in the same folder as main.py. Inside the .env file,
create a line TOKEN = xxxx, where xxxx is your discord bot token.
The default config.toml will work if you didn't change any of the default WebUI settings. However,
if you did, you must update the config accordingly.
Now, you can run the bot
python main.py
You can change the GPU that WebUI uses with the CUDA_VISIBLE_DEVICES environment variable. You set this
environment variable to the GPU you want to use. You can launch an instance of WebUI on each GPU using this
environment variable. Make sure to use and change the --port argument so each instance binds to a different port.
Edit the [webui_ips] section of the config.toml file to include links to all of your WebUI instances.
The format is instance_name = 'http://ip:port/'. The instance name that you choose does not matter.
Permissions and rate limits are planned, but not completed yet.
Currently, the only working permission is the ability to toggle DM access for all users
- Permissions and rate limits
- img2img