Vicki Boykis Some notes on the Stable Diffusion safety filter Nov 18 2022 In time for NeurIPS 2022, there are a lot of interesting papers and preprints being published on ArXiv. One I ran into recently was "Red-Teaming the Stable Diffusion Safety Filter." Having worked on content moderation before, the concept of how to moderate the content of a deep learning model was interesting to me and I thought it benefitted from a broader look. Let's dive in by starting with the paper title. The concept of "red-teaming" comes from cybersecurity.

In artificial synthesis, diffusion models worked very well, even better than GANs for images. Because of this, they became popular in the machine learning community and are a key part of systems like DALL-E 2, Imagen and Parti that use text to make photorealistic images. The field of computer vision has had the most success with diffusion models. But most of the recent research on diffusion models is not available to the machine learning community as a whole and stays behind closed doors. Here is the READY-to-USE code that will guide you through the most essential parts of diffusion models.