I played around last night with Midjourney AI until they stopped my free trial.
I started playing after someone posted his prompt about a bar on a spaceship.
There’s a prompt for me that never gets old, and I can reroll it for hours and not get bored of the amazing images, this sends my nostalgia senses tingling with each image, it’s concept art for old sci fi books that were never made into TV shows, Ian M Banks Culture series, Kim Stanley Robinson’s Mars Trilogy, Isaac Asimov’s Foundation series… The list goes on. This is one of my favourites, ‘Bar on a Spaceship’ but you can sub it for pretty much anything and still get amazing results:
A science fiction painting of a a busy rowdy bar in a spaceship with large windows showing low mars orbit. highly detailed science fiction painting by norman rockwell, frank frazetta, and syd mead. rich colors, high contrast, turquoise accents gloomy atmosphere. trending on artstationMike Bridge
Indeed he got some very nice and moody results.
I tried the same on Dall-e but it didn’t work like that at all:
So we can see the difference between Dall-e and Midjourney pretty clearly. For more artsy styles, Midjourney is better. But Dall-e is more powerful overall.
I went back to Midjourney and tried the same prompt, it indeed produced similar looks but I think that guy did a lot more trial and error before coming up with something as nice as the shared pics. Still, the fact that you can eventually end up with those is amazing.
Fed up with that test, I grabbed a prompt from someone else who was trying things out and I liked what the output was.
/imagine prompt: full-body of a Cyberpunk Girl insanely detailed and intricate, pulp art, neon, realistic photo, HDR, art by Greg Tocchini and Laurie Greasley and J.C. Leyendecker --v 4 --ar 2:3
It’s nice to note that the aspect ratio modifier works and it ends up with more useful pics than the square ones.
The upscale button produced weird results. Sometimes it worked right and the lines were a lot clearer, and it also seemed to rework some problem areas of the pic. Other times it completely messed up the look and had to go back. Here are some results I’ve ended up with.
After a certain point the results were repetitive and some didn’t work right, wrong leg poses, weird hands of course as that seems to be the bane of AI art right now, stuff like that.
All in all I think it’s a great way to get something usable if you don’t really care about the quality, like to use it in a blog post or something similar. For more professional work you can hire an artist to clean it up.