It could have been fed videos of people painting.

Connecting it's capabilities to simulating the recreation of such paintings... that's I suppose interesting. But adding brush strokes to produce an image is probably well covered by video footage in its training data. e.g. all the Bob Ross videos in existence.