For many years, my stock answer to anyone who says that computers are "fast enough" or "more powerful than most people need" is that I will only be satisfied when I can create photorealistic movies by talking to my machine.

The session could go something like this:

Create a scene: London docks, around 1919, mid-afternoon. There are a few ships being offloaded, crates stacked around, trucks and horse carts transporting freight.

<scene is created based on historical photos, ML readings of period books, etc.>

OK make it grimer, more soot on everything. And make the season wintertime. Add more stevedores. Ship in the foreground is red, weathered, and flying a German flag.

<changes are made to the scene>

OK pull back and add a pub on the right side of the screen. Let's make it dusk. Sun is setting. Get some good light spill when the door of the pub is opened.

<scene updates>

Let's work on the pub's sign. It's called the Red Swan. Make a sign carved in the shape of a swan, paint it red. Smaller, and let the wood show through where some of the paint is weathered away. Hang it over the door.

<scene updates>

etc.

This should all become possible eventually, and probably a lot sooner than many people think.