There are image models with censorship at every stage from pretraining to posttraining. Most recently Ideogram released an open weight model that will denoise

There are image models with censorship at every stage from pretraining to posttraining. Most recently Ideogram released an open weight model that will denoise into a grey image with the text "Blocked by safety filter" notice for certain prompts Of course, because it's open weights people have found defeats