writes:In a previous article, I wrote about trying to get the Deep Dream image-making AI to 'paint' a Vietnam War miniature. (Well, to 'color' a photo of a mini.)
The results were not inspiring – as 79thPA 
commented:
I think the Vietnam attempts look like the SEAL has 40K weaponry.
The AI draws its knowledge from a large database of art and images, but from my experience, its knowledge of history is rather spotty. For example, when I gave the AI the simple prompt of 'Vietnam War', it produced this image:
Apparently, it thinks the war was a riverine action involving some dodgy-looking rivercraft and random things on fire. And getting more specific didn't help. I asked for "Huey gunship firing on landing strip" and got this:
The AI did its best. There are some odd-looking choppers (one of which is missing a rotor!) approaching a settlement or base, something is firing into the air, and there's a giant star embedded in a roadway.
So if the AI doesn't know the Vietnam War, what does it know? I decided to try Star Trek. Specifically, the character of Picard, since he's familiar both from Star Trek: The Next Generation and the current Star Trek: Picard series.
And here is our starting point: a picture of the Modiphius Picard figure, recently released as a 3Dprintable mini.
As you may recall by now, with Deep Dream Generator, you can pass it a photo, a text prompt, and a filter strength setting from 10% to 90%. Higher filter settings allow more changes from the original. Above, we've tried 60% filter strength (using the Cyperspace AI model), and we've managed to get flesh tones. The photo has also been blown up to almost twice original size.
At 70% strength, not much color.
At 80% strength, color is beginning to show up.
At 90% strength, we finally have plenty of color, but not much resemblance to the original figure. There's a hint of Next Generation uniform red, but the colors are definitely unorthodox. And his legs are quite stretched.
So I thought: If it knows Star Trek, does it need a base photo at all? What if I just ask for "Jean-Luc Picard on the bridge, in the style of Star Trek"?
OK, that looks somewhat like Picard, the uniform is vaguely like the traditional uniform, and he is on a bridge! 
Now I've changed the prompt to be "Jean-Luc Picard on the bridge of the starship Enterprise," and you see the result above.
This time, I've added to the prompt "in the style of Star Trek Next Generation." Getting closer, but those are some freaky hands!
This time, I've switched from Cyberspace model to Artistic model.
My conclusion is that Deep Dream knows more about Star Trek than the Vietnam War, but is still not capable of giving us the proper coloration.