|A big lumbering steampunk battle monstrosity; the first thing I attempted, naturally|
It's mind blowing, in so many ways, and also... limited.
Imagine an idiot savant who's capable of astonishing levels of sophistication, but doesn't understand fundamentals.
Far as I can tell, the software is 'kit bashing' together thousands of source photos, with greater and lesser degrees of understanding. But beneath the slather of detail it pumps out, there's no real structure. A cynical view would be that it's a brittle egg over nothingness.
And yet, what an egg!
You can produce images with a text prompt and the click of a button! In any style you want, with whatever lighting or mood you choose to specify.
The result doesn't always stand up to close inspection. Yet when you look at them in passing, out of the corner of your eye, they can be dazzling.
Patterns, simple objects, jewelry can be examined directly and I doubt I'd know it was a MidJourney creation.
Some images are just flat out astonishing.
|Is that big thing balanced on a bicycle?|
There's something of an art to the AI prompts. Badly written ones will yield unintended or poor results. As you begin to understand the program better, you channel your efforts towards what it does well, and learn work arounds for areas it can't handle. Or leverage the happy accidents.
Animals and people, for example. Most animals are rendered as hideous deformed monstrosities, like they'd been passed through a Star Trek transporter and come back with a leg sticking out of an ear.
Once you know things like that, you can adapt.
|Midjourney does awesome wreckage|
It would take me years, literally YEARS, to produce what I have with MidJourney in the space of a week. I've realized scenes, while not exactly what I'd imagined, are pretty damn close.
It's like Creative Crack.
|The Great Library: wonderfully wonky and atmospheric|
Another software program, Stable Diffusion, is on the verge of being released to the public, and from what I can see, it DOES have structure behind the images. Joints, logical shadows, hands, feet, etcetera. Rather than starting at the surface and working down, it seems to builds up from a solid structure. No doubt it too has limitations, but it can do some things MidJourney can't.
I did not think these software programs would be taking away jobs from artists at first. The lack of structure and the difficulty in getting a specific result mean artists will still be needed to bring the job home.
But MidJourney excels at kickstarting concept art, finding a mood, initial play and exploration. Unexpected results can spark story ideas. If only I had the time to tell them!
It's great for aspiring writers who don't have money or contacts to make decent covers. MidJourney can pump out a generic retro 1970s sci-fi cover easily. Book publishers will use family vacation photos or whatever generic, unrelated sci-fi painting the editor can find. I've bought many sci-fi books where the cover was totally unrelated to the story.
MidJourney can do better.
|A futuristic garden space|
There are copyright and ethical issues. It is training on living artist work, so using it commercially may raise a mess of copyright issues. From what I understand, while you can use images generated with MidJourney (provided credit is provided to MidJourney), so can anyone else. Which means you can't stop someone else from using whatever you came up with.
I've modified a good number of images I've pumped out. I tried like the devil to get images close to Theo Paxstone, but nothing was quite right. I simply cannot get the rendering engine to produce steam mechs the way I'd imagined them.
But as a playground, it's a lot of fun!
Now if you will excuse me, I have more renders to ponder.
|Industrial farms team mech doing a crop burn|