Playing with an AI fantasy art app

[Jubal]

So, I saw people on Twitter playing with the web app Wombo which attempts to produce AI-made art in various styles, so I had a go at some fantasy stuff. Some thoughts:

• Like a lot of AI, it really can't do people, because whilst it can see pictures with people in or has some in its data banks, it can't link them to a concept or understand what the necessary features of one are.
• It does landscape far better, presumably because that can be a bit choppier without issue
• When I did the "Fens of Serraty" from my own setting, I got something much more akin to modern East Anglian fenland: I blame the disappointing under-use of fens in fantasy settings (it's probably safter than simply blaming the Dutch, for one thing...).
• It must be trained or possibly actively train itself on e.g. a google search dataset or similar. Like, the Solas picture might not be a picture of Solas from Dragon Age, but it's one that makes it very clear that some picture or pictures of him were fed into it
• It's quite good at "vibe" even if not much else: Sparta is blue skied and square in buildings, the elf city has hints of towers etc
• Having checked, the same prompt creates different images every time, not repeats.
• I was amused that the spoof name Gilbert of the Riviera came out better than Geralt of Rivia for pictures of Witchers I asked it to do.

Anyway, feel free to share your own amusments

[Gmd]

Thay geralt and solas one is deeply cursed.

[Pentagathus]

Gerald Riveiera looks like a painting by the crazy dude who cut his ear off.
Disclaimer:
Am not an art critic.

[Jubal]

Van Gogh? Yeah, fair. Though it's bordering on Picasso in places.