Using these generators as tools to find inspiration seems like the best case scenario. I think people are assigning too little probability to the potential future scenario that current systems are about as good as we're going to get with current ML methods, though some refinements will marginally improve them. Without a major breakthrough, I don't see any AI systems replacing professional artists on video games, e.g., unless it's a very low effort, low quality game.
There are still major breakthrough and improvements every few weeks, so I really wouldn't worry about already having hit rock bottom.
But even ignoring that, we have barely even stared exploring what we can do with the technology as is. A lot of it is still just experiments living in a git repository or need more GPU than the average person has. Give it a few more months or years, and you'll have it integrated into every major photo and video editor software and optimized to run on normal consumer hardware. That simple improvement in accessibility will have very wide reaching consequences just by itself, even without improving the underlying AI drastically.
And no, you won't replace the professional artists anytime soon, after all somebody still need to have the final say into what goes into the game, but it will drastically transform how that artist will work and the amount of content they'll be able to produce.
e.g., unless it's a very low effort, low quality game.
The output of Midjourney and Co. already looks spectacular, easily better than a lot of games out there. I could easily see that replacing or enhancing a lot of art in 2D RPGs, point&click adventures or visual novels.
Everything that needs animation or 3D meshes will take a while longer, but for 2D games it's already more than good enough. It's really more an issue with artists and game developers still needing to catch up on all the rapid new developments that happened over the last few months.
In 2014 we considered it almost impossible for computers to understand the contents of images. It took a few years for that to become possible. Now ML models are creating amazing images themselves. I don't see what your skepticism is based on.
https://xkcd.com/1425/
Comments
Using these generators as tools to find inspiration seems like the best case scenario. I think people are assigning too little probability to the potential future scenario that current systems are about as good as we're going to get with current ML methods, though some refinements will marginally improve them. Without a major breakthrough, I don't see any AI systems replacing professional artists on video games, e.g., unless it's a very low effort, low quality game.
There are still major breakthrough and improvements every few weeks, so I really wouldn't worry about already having hit rock bottom.
But even ignoring that, we have barely even stared exploring what we can do with the technology as is. A lot of it is still just experiments living in a git repository or need more GPU than the average person has. Give it a few more months or years, and you'll have it integrated into every major photo and video editor software and optimized to run on normal consumer hardware. That simple improvement in accessibility will have very wide reaching consequences just by itself, even without improving the underlying AI drastically.
And no, you won't replace the professional artists anytime soon, after all somebody still need to have the final say into what goes into the game, but it will drastically transform how that artist will work and the amount of content they'll be able to produce.
The output of Midjourney and Co. already looks spectacular, easily better than a lot of games out there. I could easily see that replacing or enhancing a lot of art in 2D RPGs, point&click adventures or visual novels.
Everything that needs animation or 3D meshes will take a while longer, but for 2D games it's already more than good enough. It's really more an issue with artists and game developers still needing to catch up on all the rapid new developments that happened over the last few months.
In 2014 we considered it almost impossible for computers to understand the contents of images. It took a few years for that to become possible. Now ML models are creating amazing images themselves. I don't see what your skepticism is based on. https://xkcd.com/1425/