Sneaky AI model deliberately corrupts training images to sidestep copyright

Text-to-image AI models trained on original images can memorize them, generating replicas that raise an issue of copyright infringement. A new AI model has been developed that’s trained on only corrupted images, removing that particular legal headache.

[...] Researchers from the University of Texas (UT) at Austin and the University of California (UC), Berkeley, have developed a diffusion-based generative AI framework that is only trained on images that have been corrupted beyond recognition, removing the likelihood that the AI will memorize and replicate an original work.

Academic paper: arxiv.org/abs/2305.19256