An AI tool could construct a virtual scene and then take a simulated photograph or make a simulated photorealistic painting. While I would prefer that people always use words like virtual or simulated to describe it, I think it’s likely that it will often get dropped. I’m thinking of the camera in metal gear solid. At some point people get used to something and start using less precise language.

That's as much as a photograph as someone manually creating a scene in blender and running it through the Pixar Render farm or whatever to get an ultra realistic scene - as the sibling comment stated what you have is a render

Similarly the camera in MGS would be a screenshot, as in a shot of the screen

That's called a render.

The trained neural net form of it would perhaps be a virtual render.

A painting is actually painted with pigment on something. An LLM by itself cannot grab a brush. Maybe with something else.

It has virtual brushes. This predates AI by quite a lot. It's important to be careful to call it a virtual or simulated painting but perhaps not all the time. Maybe every single instance it's brought up but not always repeated – if you wrote a paper about virtual painting perhaps 10 of 100 occurrences of the term painting would come with a term like virtual or simulated.

A simulated painting is not a painting, so there is no contradiction.