AI art gen­er­a­tion has been explod­ing in the last year, and the once dif­fi­cult to man­age tools are now incred­i­bly easy to use, mod­u­lar, and ver­sa­tile. Last year I cre­at­ed a Dis­cord bot that gen­er­at­ed AI art from my own pri­vate­ly host­ed DALL‑E 2 pub­lic mod­el. This was quite cum­ber­some to set­up, and it wasn’t very flex­i­ble if I want­ed to make small changes. After explor­ing what’s changed in the last year, I can con­fi­dent­ly say that all of my pain points have been resolved by the lat­est iter­a­tion of art gen­er­a­tion: Sta­ble Dif­fu­sion. In this post, I want­ed to share how easy it tru­ly is to cre­ate your own AI art gen­er­at­ing Slack bot. I chose Slack this time, because at Michi­gan­Labs we use Slack as our com­mu­ni­ca­tion tool, so it’s a lit­tle more rel­e­vant to our own work. But there is a lot of over­lap between Slack and Dis­cord, if you’re using their respec­tive Type­script libraries. With that said, start your timer, because this should only take about 10 min­utes (as long as your inter­net speed is fast enough!).

Require­ments #

You just need a com­put­er with a 4GB+ GPU.

You can use Lin­ux or Apple Sil­i­con, but it’s a bit eas­i­er (with respect to NVIDIA dri­vers) with Win­dows. For my set­up, I like mod­u­lar­i­ty and low-pro­file hard­ware. I use the following:

1. Intel NUC 8 — One of the few mini PCs that has Thun­der­bolt support
2. Raz­er Core X — This is an excel­lent plug-and-play enclo­sure that uses Thunderbolt
3. RTX 3080 — This gen­er­ates about one image per sec­ond, which has been plen­ty fast for my needs

Get­ting Start­ed #

Next, you’ll need to install a cou­ple depen­den­cies and acti­vate a new Slack application.

The Sta­ble Dif­fu­sion Web UI

This sta­ble dif­fu­sion web UI project gives us more than what we’re try­ing to accom­plish, but that’s a good thing! It’s high­ly exten­si­ble, with tons of options for adding more mod­els and cus­tomiza­tions. It can also pro­vide a real­ly great web fron­tend to inter­act with your host­ed mod­el. For our Slack bot pur­pos­es how­ev­er, fol­low these steps to get it ready for a dif­fer­ent interface:

1. Install all of the depen­den­cies from the Web UI readme (the git repo, Python 3.10+, and the Sta­ble Dif­fu­sion 4GB model)

2. Go into the stable-diffusion-webui repo and open webui-user.bat, and change the COMMANDLINE_ARGS so that your file looks like this:

    @echo off
       
    set PYTHON=
    set GIT=
    set VENV_DIR=
    set COMMANDLINE_ARGS=--api --nowebui --listen
       
    call webui.bat
   

   • --api tells the web UI to gen­er­ate a swag­ger doc at the server’s <IP>/doc web­site path. This is option­al, but help­ful when you’re try­ing to learn about what your new api can do 
   • --nowebui makes it start a bit faster by skip­ping the fron­tend. This is optional 
   • --listen tells the api serv­er to run on the local net­work, instead of the local host. This is required for our Slack bot to be able to access it 

3. Run webui-user.bat and the serv­er should start up in a com­mand prompt window

To ver­i­fy that it’s run­ning prop­er­ly, try going to <IP_ADDRESS>:7861/docs in a brows­er. You should see a Swag­ger UI page.

The Slack Bot

Next, we’ll cre­ate a very sim­ple Slack appli­ca­tion using Type­script. This bot will be invoked in a Slack chan­nel by typ­ing /art <PROMPT>, and the bot will reply with the result from our Sta­ble Dif­fu­sion model.

1. Make sure you have Node.js installed

2. Clone my bare-bones Type­script Slack bot tem­plate to fol­low along quicker

    git clone https://github.com/DaveAldon/AI-Art-Slack-Bot
   

3. Inside this repo, install the depen­den­cies with a quick npm install. Then cre­ate an .env file that match­es the .env.example file. For the BACKEND_URL field, add your Sta­ble Dif­fu­sion server’s IP address and port like this: http://<IP>:7861. The API runs on port 7861 by default

4. Go to the Slack appli­ca­tion dash­board, and cre­ate a new app from an app man­i­fest. This will make con­fig­u­ra­tion easier

   

   You can use the man­i­fest I used for this basic exam­ple below:

    display_information:
      name: AI-Art-Bot
    features:
      bot_user:
        display_name: AI Art Bot
        always_online: false
      slash_commands:
        - command: /art
          description: Get AI generated art from your prompt
          should_escape: false
    oauth_config:
      scopes:
        bot:
          - commands
          - users:read
          - files:write
    settings:
      interactivity:
        is_enabled: true
      org_deploy_enabled: false
      socket_mode_enabled: true
      token_rotation_enabled: false
   

   We need the files:write per­mis­sion, which can be dif­fer­ent from ordi­nary Slack bots, because of how we’ll upload our pho­tos into the channel.

5. Install to a workspace

   Press the Install App to Workspace but­ton, and autho­rize the bot to be installed in your workspace.

6. Retrieve the need­ed tokens

   Go to App-Lev­el Tokens in your bot’s set­tings page, and add a new one with what­ev­er name you want, and the connection:write scope. This will be used to send mes­sages to the chan­nel. Place this token in your .env file as SLACK_APP_TOKEN.

   

   Last­ly, we need to gen­er­ate an OAuth token. Go to Install App and copy the Bot User OAuth Token. Put this in your .env file as SLACK_BOT_TOKEN

   

7. Run the bot, and test it out in a chan­nel in your workspace

    npm run start
   

   Go to a chan­nel and type /art a red apple. You should see the bot reply with a gen­er­at­ed image.

   

   If you try it and you get a slack bot error: 'not_in_channel' error, add your bot to the spe­cif­ic chan­nel by typ­ing /invite @<BOT_NAME>.

Con­clu­sion #

If you were able to fol­low along and get your Slack bot to return a pho­to, con­grat­u­la­tions! You’ve just cre­at­ed a sim­ple inter­face into an AI model!

I hope that you enjoyed walk­ing through this tuto­r­i­al, and are tak­ing time to look through the Sta­ble Dif­fu­sion Web UI docs to dis­cov­er more ways you can inter­act with these amaz­ing models!

If you had issues, the beau­ty of how this works is that we can trace every­thing back quite eas­i­ly. There aren’t a lot of mov­ing parts to this.

Trou­bleshoot­ing #

Check that your Sta­ble Dif­fu­sion mod­el is run­ning and acces­si­ble on your local network

Try going to <IP_ADDRESS>:7861/docs in your brows­er. Is this work­ing on your local machine?

If you remove the --nowebui flag, can you go to <IP_ADDRESS>:7861 and see the web UI?

Check that your Slack bot is run­ning properly

Try a basic use-case of just return­ing some sta­t­ic text. Replace the app.command func­tion with this:

app.command("/art", async ({ body, ack }) => {
    await ack({
    response_type: "in_channel",
    text: "test!",
    });
});

Does your bot reply with test! in the channel?