So I’ve been using Stable Diffusion (bundled as Diffusion Bee and as Stable Diffusion.openvino) to make placeholder art for a personal game design project.

While prompt engineering is lots of interesting fun to try and tease useful images out of the latent model, it does produce a lot of images that can’t be used for one reason or another. Some are inappropriate (to my sensibilities) for an all-ages game, others are filled with AI image generation artifacts: