Gabby and Bull
atcDave
Member Posts: 2,817
My latest team has made into BGII. We all know the story, but I've posted how Gabby will experience it all. Or re-read from BG1. All with AI art and videos.
0
Comments
I do, however had, also think that several of the AI images do not quite fit into the BG1/BG2 "theme". That's hard to describe because even the regular art by real humans differs. The images here specifically remind me of some game engines, where the generated images all seem to be in a similar "style" and that style can look a bit oddly. For instance, the tiger shaman thingy looks like a human priest and then a tiger head was put on top of it as direct replacement.
How easy is it to tweak these AI models? Can one generate or use a specific style that would be as close as possible to the default BG themes?
I assume this would require feeding the model with all available BG art. Has anyone tried this, to then see if this could generate better, or rather, more fitting art? Which AI models are in general used, if AI is used therein? To me these do not all seem to be from the same AI model, but I am very novice when it comes to AI generated content in general, simply because I haven't used it much at all save for demo purposes. How many content creators with regards to older beamdog games make use of AI in general? I know that morpheus used AI for voice generation in some of his mods, for instance. I don't currently know whether anyone has used AI for map generation, for instance, or code generation in scripting for e. g. BG2 or something else.
Another minor observation is that the generated portraits all seem to be drawn in perfection, whereas from a real human there are more oddities, thus making images drawn more unique and a bit messier rather than "pixel-perfect-pureness". An example can be seen by this for the old images on books for the Dragonlance saga and comparing this with AI generated cover for those old fantasy novels.
I’ve used more variety on my IWD runs. For some reason I like realism for BG but lean towards more traditional looking paintings for IWD. But then I often specify the style of Larry Elmore or Clyde Caldwell. Sometimes Keith Parkinson. They were all most active in the 1980s when my interest in D&D was at its highest.
There is always a problem with stability from image to image. Because you start from a verbal description (in most cases) most AIs have no concept of the same character going from one image to the next. And you have limited room to set a scene. So I describe characters in broad terms and hope they come close.
There are some AI editing utilities attached to most renderers. Some work better than others, all are harder to use than the initial image generator. But most posts I’ve made have one or two images with at least some editing.
One of the renderers I use, OpenArt, does allow for the creation of models. I made much use of this on my “Chuck and Sarah” run a year ago. But this has many limitations too, mainly I think OpenArt has a weak understanding of fantasy.
The potential for where all of this can go is sky high. Frighteningly so. The technology is in its infancy, it will get better and more powerful quickly. I think the ability to render by style and taste is one area that has already improved in the 18 months I’ve been playing it.
To me, the most fun of this is that I am absolutely no artist. My ability to draw or paint anything is comically bad. But I’m okay with words, and I know what I like. So I can get images that are sort of how my imagination and taste intersect with the game.
Zeno_42 has done a variety of fantasy styles in a number of CRPGs and is truly far better at this than I am. You can check out his website here.
My own website has many examples of BG and IWD inspired art. I’m planning on doing a “Pool of Radiance” run through next.
If you follow the other links above you’ll see much bigger samples just of my own flawed efforts at this, including videos! For some reason I can’t post them here, I’m only just learning how to get some of what I want with those.