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