This is a controversial topic, but I find no shame in using AI-generated images in games. In fact, I feel we're about to see a proliferation of this use of AI in the gaming industry.
Notably, however, not all AI image generation is desirable, or even useful. In fact, most of the things I tried to generate from it, such as gaming assets, icons, and similar things, were not consistent enough to be usable; I bought assets or drew my own for this.
I did discover that AI can be very good at generating some styles of image for games... but I discovered this by chance. The key, as I see it, is not to try to use AI to solve an existing problem, but rather to take advantage of scenarios where AI happens to produce good results.
Case in point, I had no major plans of completing this project until I started playing around with image generation and realized that AI can generate some really... magical landscapes. They are not always structurally correct, and sometimes things look way off, but no more than a landscape in a Dr. Seuss book might. This means that when my goal is to generate majestic, fantastical landscapes or pictures of places that are, to say the least, not meant to be real, the results actually go beyond expectations. Since my game is about exploration, my inspiration for the whole idea of Landmarks in the game was because I happened to become fascinated by the results of my generative AI images. I put some of them as images in this post as some examples.
Realistically, commissioning an artist to make this scale of landscape images would be infeasible in terms of time and certainly cost for a hobbyist project; it would take months to draw images of similar quality from scratch, and I dare say that, despite some flaws in the AI's interpretation, it would take quite an artist to make work of similar quality. And I can't even imagine the cost. At that rate this would be a feature left out of the game entirely.
All that said, there are some things which the AI absolutely cannot do properly, with any level of consistency:
Animal anatomy (I couldn't make a horse carriage where the horses didn't look off)
Tools, including weapons. I had a 0% success rate even rendering simple axes or swords. Forget about people holding weapons or tools.
Landmarks with specific attributes: I was never able to get a good picture of a "broken wagon on the road". I tried about 40 different prompts with different engines before giving up entirely. Similar for things like animal skeletons on the road; they always drew them upright, on all fours.
Sprites, sprite-sheets, icons, etc.. It can do them, but they are too inconsistent and cannot be relied on to maintain a proper style.
Character art, to a degree. You can get a single frame of a character, but forget getting additional expressions for the same character.
It's worth noting that these images actually don't play too much of a role in my game, functionally. They are static images framing a location, but you cannot interact with these visual environments beyond just looking at a single still image. Nonetheless, my hope is that this still evokes some wonder in the player, like it did for me. After all, it's much better to have a visualization of the Nigh Bazaar, Cozy Village or Giant Rock Pillar you might run across.
My views on generative image AI, ultimately, are that it will be used like any other tool. It's too inconsistent to replace functional graphic or design artists, though it might replace concept artists. Like any tool, it's not perfect for all cases: you shouldn't use a drill to glue pieces of paper together, and likewise you shouldn't expect that these image AIs are useful for all image tasks. That does not, however, mean they are not useful for something. I mean, come on, look at these landscapes.
Hozzászólások