AI Generated Art Using Google Collab and Stable Diffusion Workshop

Stable Diffusion dark concept art of aliens at a science fiction convention lurking in the shadows, sharp details, octane render
Stable Diffusion prompt: dark concept art of aliens at a science fiction convention lurking in the shadows, sharp details, octane render

One of the things I did for the 2023 Capricon Science Fiction Convention was to teach a workshop on AI Generated Art Using Google Collab and Stable Diffusion. The goal of the workshop was to demonstrate for attendees how to use Google Colab to run Python programs that would use APIs (Application Programming Interface) to access the online libraries and code necessary to create both AI generated art and text. The workshop was a part of the Capricon Science Fiction Convention Art programming track.

Using Google Colab (Colaboratory) as the computing platform offered two key benefits:

  1. Using Colab meant that no software or programs would need to be installed on the user's own computer.
  2. Google Colab offers access to the GPU hardware necessary to significantly speed up the generation of images.

Three Google Colab notebooks were provided for the participants with all notebooks focused on providing the core essentials. The first notebook was for using the OpenAI GPT-3 and DALL-E libraries to generate text and images. The second and third notebooks offered several different ways of using Stable Diffusion to generate images.

The following files were provided to workshop attendees:

stable-diffusion-colab-workshop.pdf - the slides used for the workshop presentation.

CapriconOpenaiAPI.ipynb - the Google Colab notebook (a IPYNB file) containing the code for running both the OpenAI GPT-3 text generation tool and the DALL-E image generation tool.

CapriconStableDiffusionExample.ipynb - a Google Colab notebook (a IPYNB file) containing the code for using Stable Diffusion to generate images.

CapriconStableDiffusion.ipynb - another Google Colab notebook (IPYNB file) containing another set of code for using Stable Diffusion to generate images.

With a focus on simplicity and audience interaction, while minimizing discussion of the technical details of deep learning, transformers, and large language models, the workshop appeared to succeed in its core objective of making the GPT-3, DALL-E, and Stable Diffusion libraries accessible for those without a programming background.