Easing the creation task for professional uses is definitely "in the pipeline". If you're making a 2D game, diffusion models are already capable enough to act a

Easing the creation task for professional uses is definitely "in the pipeline". If you're making a 2D game, diffusion models are already capable enough to act as a shortcut for detailed painting. It's stuff with generating optimized meshes, animation, scene lighting, other details that is still in the research stage. In many ways it's more straightforward and gives a better demo to address the "render the entire scene" case than to go asset-by-asset and work out a better UX for that than "do everything by hand". Even though the latter is what's really being called for.