Soon after VQGAN-CLIP was released last year, somebody made a bot in #the-faraday-cage on Eleuther’s Discord that ran the model. You asked it to !imagine something, and then VQGAN-CLIP responded with its weird interpretation of your request.
It was thrilling. This was the first playground for text to image models, the first place where people could prompt computers to produce art. And, it was hilarious! Instead of the model being locked up in a Colab, it was a hundred people in a room playing with it together. People were riffing off each other’s ideas, patting each other on the back, laughing at each other’s outputs. This is where the “unreal engine”, “artstation”, and other prompt hacks appeared. Somebody saw somebody else do something neat, and then the prompt meme reproduced.
We’ve come a long way since four-eared faceless rabbits. Earlier this week, we were blessed with the weights from The Great A100 Cluster. So, let’s put them to good use.
This guide will show you how to create your own Discord bot with Stable Diffusion, so you can have a much better robot artist in your own Discord server.
We’ll use Replicate to run the model so you don’t need to have a GPU, and we’ll deploy it to Fly.io so you can keep it running all the time, even when your computer’s off.
