Thanks for the reply, mahi. Unfortunately, if your service is replying on diffusion models, then it has likely been trained using web-scraped datasets (for example LAION and CIFAR-10) which infringe copyrights when they're subsequently used for commercial purposes. For example LAION is not for commercial use, see…