AIYA

A Discord bot interface for Stable Diffusion

Usage

To generate an image from text, use the /draw command and include your prompt as the query.

To generate a prompt from a couple of words, use the /generate command and include your text as the query.

Currently supported options

• live preview
• negative prompts
• swap model/checkpoint (see wiki)
• sampling steps
• width/height
• CFG scale
• sampling method
• seed
• Web UI styles
• extra networks (hypernetwork, LoRA)
• face restoration
• high-res fix
• CLIP skip
• img2img
• denoising strength
• batch count
• compatibility with SD.Next
  • "Full quality" VAE toggle

Bonus features

• /settings command - set per-channel defaults for supported options (see Notes!):
  • also can set maximum steps limit and max batch count limit
  • refresh (update AIYA's options with any changes from Web UI)
• /identify command - create a caption for your image.
• /generate command - generate a prompt from text, using https://huggingface.co/Gustavosta/MagicPrompt-Stable-Diffusion
• /stats command - shows how many /draw commands have been used.
• /queue command - shows the size of each queue.
• /info command - basic usage guide, other info, and download batch images.
• /upscale command - resize your image.
• buttons - certain outputs will contain buttons.
  • 🖋 - edit prompt, then generate a new image with same parameters.
  • 🎲 - randomize seed, then generate a new image with same parameters.
  • 📋 - view the generated image's information.
  • ⬆️ - upscale the generated image with defaults. Batch grids require use of the drop downs
  • ❌ - deletes the generated image. In Live preview this button interrupts generation process
  • ➡️ - skips the current image generation in live preview and go to next batch (if there's more than 1)
• dropdown menus - batch images produce two drop down menus for the first 25 images.
  • The first menu prompts the bot to send only the images that you select at single images
  • The second menu prompts the bot to upscale the selected image from the batch.
• context menu options - commands you can try on any message.
  • Get Image Info - view information of an image generated by Stable Diffusion.
  • Quick Upscale - upscale an image without needing to set options.
  • Batch Download - download all images of a batch set without needing to specify batch_id and image_id
• mark image as spoiler
  • Per image (on /draw)
  • Set channel-wide default or force based on role with /settings
• configuration file - can change some of AIYA's operational aspects.

Setup requirements

• Set up AUTOMATIC1111's Stable Diffusion AI Web UI OR SD.Next
  • AIYA is currently tested on commit 20ae71faa8ef035c31aa3a410b707d792c8203a3 of the Web UI.
  • For SD.Next currently tested on master branch at 2024-03-01 (325ed10a04775c49c36fc3308559507a4a82271b)
• Run the Web UI as local host with API (COMMANDLINE_ARGS= --api).
• Clone this repo.
• Create a file in your cloned repo called ".env", formatted like so:

  # .env
  TOKEN = put your bot token here

• Run AIYA by running launch.bat (or launch.sh for Linux)

Deploy with Docker

AIYA can be deployed using Docker.

The docker image supports additional configuration by adding environment variables or config file updates detailed in the wiki.

Docker run

docker run --name aiyabot --network=host --restart=always -e TOKEN=your_token_here -e TZ=America/New_York -v ./aiyabot/outputs:/app/outputs -v ./aiyabot/resources:/app/resources -d ghcr.io/kilvoctu/aiyabot:latest

Note the following environment variables work with the docker image:

• TOKEN - [Required] Discord bot token.
• URL - URL of the Web UI API. Defaults to http://localhost:7860.
• TZ - Timezone for the container in the format America/New_York. Defaults to America/New_York
• APIUSER - API username if required for your Web UI instance.
• APIPASS - API password if required for your Web UI instance.
• USER - Username if required for your Web UI instance.
• PASS - Password if required for your Web UI instance.
• USE_GENERATE - Set whether the /generate command is enabled as well as if required package (torch nvidia transformers) are installed

Docker compose

• Clone the repo and refer to the docker-compose.yml file in the deploy directory.
• Rename the /deploy/.env.example file to .env and update the TOKEN variable with your bot token (and any other configuration as desired).
• Run docker-compose up -d to start the bot.

Notes

• See wiki for notes on additional configuration.
• See wiki for notes on swapping models.
• 📋 requires a Web UI script. Please see wiki for details.
• Ensure AIYA has bot and application.commands scopes when inviting to your Discord server, and intents are enabled.
• As /settings can be abused, consider reviewing who can access the command. This can be done through Apps -> Integrations in your Server Settings. Read more about /settings here.
• AIYA uses Web UI's legacy high-res fix method. To ensure this works correctly, in your Web UI settings, enable this option: For hires fix, use width/height sliders to set final resolution rather than first pass
• For systems with less memory/cpu, or if the /generate command is not needed, it can be disabled by setting the environmental variable USE_GENERATE=false for docker/cli.

Credits

Foundation

AIYA only exists thanks to these awesome people:

• AUTOMATIC1111, and all the contributors to the Web UI repo.
  • https://github.com/AUTOMATIC1111/stable-diffusion-webui
• vladmandic and all SD.Next contributors
  • https://github.com/vladmandic/automatic
• harubaru, my entryway into Stable Diffusion (with Waifu Diffusion) and foundation for the AIYA Discord bot.
  • https://github.com/harubaru/waifu-diffusion
  • https://github.com/harubaru/discord-stable-diffusion

Great Contributors

These people played a large role in AIYA's development in some way:

• solareon, for developing a more sensible way to display and interact with batches of images.
• danstis, for dockerizing AIYA.
• ashen-sensored, for developing a workaround for Discord removing PNG info to image uploads. edit Discord is no longer doing this at the moment, but credit is still due.
  • https://github.com/ashen-sensored/sd_webui_stealth_pnginfo
• gingivere0, for PayloadFormatter class for the original API. Without that, I'd have given up from the start. Also has a great Discord bot as a no-slash-command alternative.
  • https://github.com/gingivere0/dalebot
• You, for using AIYA and contributing with PRs, bug reports, feedback, and more!