V1 sucks, V2 might work.

Starting with a sketch it paints it to realistic.

Trained on fairly static scenes, more motion not yet tested.

So far it's not flexible without a starting image.