2084: AI Game Development in the Future
A super short post cataloguing AI models for Game Development
Been a busy few days, so short post today as well. Today, I’ll just for interest sake catalogue how many functions of game development are being simulated partially or wholly by existing AI models:
Code: OpenAI’s Codex is progressing rapidly in generating large amounts of code, and combined with Github’s Copilot, means that in the future game programmers might have to do less programming, and more pipelining. Everyone’s favorite programming task.
Sound Effects: While music might be still hard to generate, small sound effects seem to not be as hard as they are much shorter than most songs. Therefore Meta’s AudioGen which generates a sound effect corresponding to the text you give it is really good and will probably replace stock sound sites
Voice Acting: Uberduck is a pretty good Text2Speech system, and Nvidia’s RADTTS is quite capable. But beyond them, there’s a lot of text 2 speech models being worked on in research, which can be found on Hugging Spaces. Some of them sound uncannily realistic. The combination of OpenAI’s GPT-3 and some of these really good text to speech models could be uncanny, especially if they get any better. Of course they will run into the issue of audio generation being tricky.
Music: While real audio music generation is still not quite there, Aiva is probably a good option for symbolic music generation, as it’ll happily make you nice-sounding music in whatever style you want.
Textures: Midjourney, Stable Diffusion and DALL-E 2 all support making tilesets and textures if you request them.
UI: Stable Diffusion somewhat, if you specifically request them.
Concept Art: Midjourney all the way. Anything you want in any style in less than 30 seconds.