Unfiltered AI Image Generator Stable Diffusion Sparks Ethics Debate
Stability AI's new open-source image generator achieves rapid adoption but faces intense scrutiny over its unfiltered nature, which allows users to easily create non-consensual deepfakes and explicit content.
A new open-source AI image generator called Stable Diffusion sees stunningly swift uptake in its first week by allowing users to produce highly realistic pictures from text prompts on consumer hardware. Platforms like Artbreeder, Pixelz.ai, NovelAI, and Midjourney quickly integrate the model to enhance their own creative tools and offer greater photorealism to their users.
However, the unfiltered nature of the model leads to less savory applications, particularly on discussion boards like 4chan where users generate nude celebrity images and other forms of AI pornography. Stability AI CEO Emad Mostaque calls the early leak "unfortunate" and points to an included Safety Classifier tool designed to detect and block offensive images, though this safety feature is easily disabled by anyone running the software.
Unlike strictly filtered competitors such as DALL-E 2, Stable Diffusion technically allows the creation of public figures and lacks hard-coded restrictions against explicit material, creating a risky combination that experts warn could lead to widespread non-consensual deepfakes. Industry leaders express grave concern that these synthetic images primarily target women and could perpetuate abuse or falsely implicate innocent people in crimes.