Before diving head-first into Act XIV I decided to do a small update first.
Part one was an experiment with AI and motion. I suspect it's because my life's been immersed in game development for two decades now, but for the last week it's felt as if the entire world was talking about AI (in a mixture of rage and wonder, depending).
There's been a lot of progress over the last few years, but November saw two really impressive updates, and I diverted into a rabbit hole for a couple of days playing with it.
The first of these is ChatGTP, an AI camouflaged as a chatbot.
It's remarkably intelligent, knows how to respond to feedback, and is very much a child of the internet. It also knows how to hotwire cars, but it really doesn't want to tell you... that's what happens when you give an AI training net access to the internet and let it go.
The second (and far more controversial) one is MidJourney, a text-to-image generator. Where ChatGTP was given access to the internet at large, ChatGTP was essentially handed a very very large selection of images download from Art websites, like ArtStation.
You interact with the AI like a bot in Discord, "prompting" it with inputs to which it responds.
a fairytale princess marries a dark, handsome prince 2D game, volumetric light, dust in air, realistic, ar 1:1 --v 4 --q 2
(In case you're curious, the tags at the end set the aspect ratio to 1:1, uses version 4 of the AI, and bumps up the quality to max).
There's a number of things that are controversial here. The first one is *where* the data came from; like ChatGTP the MidJourney AI has been quietly ingesting information from the internet, in this case hundreds of thousands of art pieces made by thousands and thousands of artists, most of whom never thought their hard work would be fed to an AI.
Well. Damn. Part of the reason most people didn't worry too much was that as late as 6 months ago the results of AI generated art was pretty low-quality, fiddly and hard to access. With MidJourney all of that changed, and a number of artists (understandably) panicked.
While the results look really impressive (especially in terms of lighting and shading), it's not without it's issues.
it can't do text to save its life, it struggles with details (especially hands, feet, swords, etc), and it doesn't really know composition (here is "red ball on top of a green cube").
You can get it to make a really impressive stand-alone scene, but making a VN using it is (so far) impossible; you couldn't get the specific details and composition you need.
You can also prompt it to use an image as input. Here I gave it my background from the Dark Paths of Elsewhere, and asked it for some variations:
Wow... that ended up looking *really* good. It's abstract enough that you don't need to worry about 6-fingered hands or messed up text, it's primarily derived from my image (which makes me feel a little less guilty about the art that went into the training set), and it looks pretty.
Now a sensible person would have said "neat", and either just forgotten about this, or picked one of the images to replace the existing one, and moved on. Sadly I'm not that sensible. So I crank up AfterEffects and go to work.
There's a couple of things happening here; I'm warping the image from one to the other, then adding special effects on top. I ended up loving it enough that I made two versions (one for the Dark paths, one for the Light paths). I then made 4 small clips (2 looping still images (A and B), and two transitions between them). When it plays, it goes something like: "Loop A 1-3 times, Play A->B, Loop B 1-3 times, Play B->A". It adds a nice bit of randomness, and captures the idea that the Paths are never quite static... they constantly change and morph.
I think AIs are at the cusp of revolutionizing a number of aspects of our society, and along the way we're going to run into various ethical challenges. I don't think it can be *stopped*, though, so at the very least we need to understand it, and know what it can and cannot do.
Apart from that, this update adds in an option for a replayable Chosen 4-some (at night in the KD Bedroom), and cleans up a number of little bugs and typos that snuck in. I've started writing and rendering Act XIV, so that's up next.
Was this kind of "behind the scenes" information useful? Should I do more of it, or less? Let me know in the comments.