Thoughts on high effort AI art?

[tsuki3341]

There have been many waves of hate against AI art, and with Steam banning and itch banning a bunch of nsfw devs (me included) I'm not entirely sure if sidestepping all of this bs and just trying to make a bigger game with AI art and "self publishing" (e.g. patreon) wouldn't be a better way ... but then many people hate on it.

I can't really afford to hire an artist, but I'm a very good developer and pretty good at AI, so the stuff I'd make would not be just generic poop, but not sure if that really matters, or if there are things to watch out for.

 

[Gunizz]

AI is making a lot of progress but still all images look the same, with one single character in the middle of the scene. I haven't seen good complex images with many characters, or original points of view. Myself have messed with it, apart from the notorious problems (hands etc) making anything other than "single person 3/4 figure" led to medriocre or terrible results.
But the technology is improving week after week so in a few months the scenario could be different.

At the moment, AI VNs are really bad compared to anything DAZ.

 

[tsuki3341]

You're totally right that doing lots of characters is difficult, especially big scenes, though it's not impossible. I haven't done any AI generation but recently was quite surprised by how far the quality has gone, for "painterly non-photorealistic" stuff, e.g. one cherrypicked example



or an example with two characters (definitely would need fixing hands/feet and some other details)



IMO the reason most AI VNs are bad is not because of AI though, it's because people making them are just going for the easiest thing, whereas DAZ is actually work in comparison.

Honestly if DAZ didn't have such a shitty slow UI I'd probably be using it, but after having spent a year with it I just can't anymore. I ended up buying a pretty high end PC, and the thing is still as fking slow as Photoshop, it doesn't matter what computer one has, compared to Blender/Houdini doing anything in DAZ is like going to the dentist and having my teeth drilled through my asshole. It's definitely "fast workflow", but UI slow and laggy af.

 

[DuniX]

Depends on what kind of game it is.
If you are using it just as a replacement for Daz or Honey Select to make another Adult "Visual Novel", don't bother.

But if you use it for say a Dungeon Crawler, things could get more intresting.
You can generate all the Monsters that you fight and you can generate the BG CG and illustrations for the diffrent environments and locations.
You may even be able to generate furries and monster girls for those communities.
Fantasy worlds in general tends to work well with AI CG.

Basically AI CG currently works best for those game that used to have random anime images, and it's pretty much a straight improvement to that.
If you look at the "ai cg" tag that is the kind of projects it is used.

 

[QuantumVNs]

You left "high effort" undefined in the OP, so it's not clear where you're drawing that hazy line that you think you can cross. For one off images of one or two characters StableDiffusion can produce lots of fap material. If you want to get consistent characters, posing, background, and body proportions it gets more tricky and AFAIK the current advice involves a lot slapping methods together and hoping that the speed makes up for throwing out 90% of images.

Anyway, good luck. I think there's a lot of promise for the tech, but probably as an enhancement of current methods rather than images from whole cloth.

 

[DuniX]

If you want to get consistent characters, posing, background, and body proportions it gets more tricky and AFAIK the current advice involves a lot slapping methods together and hoping that the speed makes up for throwing out 90% of images.

The methods are much more complex nowadays than that with all kinds of AI wizardry.
I hear about networks, Lora models and whatnot and I go, Yes, AI Wizardry.