Feel free to sponsor this project at https://patreon.com/manuelkiessling.
We are building a Discord community around this project:
You can also test drive the Bot there!
Note that for all Stable Diffusion images generated with this project, the CreativeML Open RAIL-M license applies.
The following setup is known to work on AWS g4dn.xlarge instances, which feature a NVIDIA T4 GPU.
This installs the required OS packages:
sudo apt-get update
sudo apt-get -u dist-upgrade
sudo apt-get install \
ubuntu-drivers-common \
libsm6 \
libxext6 \
libxrender-dev \
mariadb-server \
net-tools \
php8.1-cli \
php8.1-xml \
php8.1-mysql \
composer
This makes sure your kernel can talk to your NVIDIA GPU:
ubuntu-drivers devices
sudo ubuntu-drivers autoinstall
sudo reboot
Running nvidia-smi should now look like this:
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 515.65.01 Driver Version: 515.65.01 CUDA Version: 11.7 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Tesla T4 Off | 00000000:00:1E.0 Off | 0 |
| N/A 58C P8 18W / 70W | 2MiB / 15360MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
This sets up the conda environment used by Stable Diffusion:
wget https://repo.anaconda.com/miniconda/Miniconda3-py38_4.12.0-Linux-x86_64.sh
bash Miniconda3-py38_4.12.0-Linux-x86_64.sh
When asked "Do you wish the installer to initialize Miniconda3 by running conda init?", answer yes.
Then, log out of your shell session and log back in - shell prompt should now start with (base).
You can now set up Stable Diffusion:
git clone https://github.com/CompVis/stable-diffusion.git
cd stable-diffusion/
conda env create -f environment.yaml
conda activate ldm
cd -
Important: Download [https://huggingface.co/CompVis/stable-diffusion-v-1-4-original/resolve/main/sd-v1-4.ckpt] to ~/stable-diffusion/sd-v1-4.ckpt.
sudo mysql
GRANT ALL PRIVILEGES ON discord_bot_backend.* TO 'root'@'localhost' IDENTIFIED BY 'secret' ;
git clone https://github.com/manuelkiessling/stable-diffusion-discord-bot.git
cd stable-diffusion-discord-bot/discord-bot-backend
cp .env.local.dist .env.local
Now, set up a new Discord application at [https://discord.com/developers/applications].
Edit file .env.local and set your own bot token - generate it at [https://discord.com/developers/applications/<your-application's-id>/bot].
composer install
php bin/console --no-debug doctrine:database:create
php bin/console --no-debug doctrine:migrations:migrate
Start the following command once, to have your bot register its commands with Discord:
php bin/console --no-debug app:bot:register
Then, run the actual bot command in a separate screen session:
screen
while true; do php bin/console --no-debug app:bot:run; done
CTRL-A D
And the task consumer in another one:
screen
while true; do php bin/console --no-debug messenger:consume async --memory-limit=2048M --time-limit=240 --limit 1 -vv; done
CTRL-A D
Create an invite URL for your bot at [https://discord.com/developers/applications/<your-application's-id>/oauth2/url-generator]. You need scopes bot and application.commands, and permissions Send Messages, Attach files, Read Message History, and Read Messages/View Channels.
You then need to open the generated URL to invite the bot into your Discord server.
Talk to the bot using the /draw slash command.