I've been torn about the use of AI generated assets for gamedev. It's not so much about the "ethics" per se because even if generative AI replaces artists, at least it allows better opportunities for solo gamedevs who aren't good artists, and I don't see games entirely generated by AI coming anytime soon (ChatGPT can even write anything more advanced than Space Invaders without severely fucking something up). Also, I don't believe in the claims about it's impact on climate change. I suspect that claim was created by corporations like OpenAI and Meta just to get governments to ban AI tools outside of approved facilities that use "green" energy.

It just boils down to these points:

1. I don't think text-to-image will ever be good. You need to be very descriptive about what you want, but something too descriptive may produce unexpected results. Image-to-image is better because it gives a rough visual estimate of what you actually want. If I were working for an actual artist right now, I would just draw rough sketches and ask them to make them better or to put them in sprite form.

2. Music generation isn't much better. I agree with Konji Kondo that music needs to match the "rhythm" of the gameplay and I don't see any tools where you can upload a video (i.e gameplay recording) and have it generate appropriate background music for it. You also need to be very descriptive of what you want, which requires that you have some knowledge of classifying music.

3. Unless you have a computer with a really powerful enough GPU, you are stuck with using services that may or may not own the assets you generate. Some people don't care, but I see this being an issue for a lot of people. Granted, not every service does this, but this is something you need to check before using.

4. Contrary to all the fear-mongering about job loss from AI, there is a growing concern that AI will start feeding on it's own data (if it hasn't already) and generated results will be worse in quality. This is known as "model collapse". Nobody has a good solution to this. AI models can try to watermark their generated content, but that requires that all models use the same watermarks. So, this means that AI generated content and human generated content will have to coexist with roughly the same amount of content being outputted in order for generative AI to still be usable in the future.

Now, I am using AI assets for the game I'm developing right now, but they are just backgrounds. One of them was actually pointless because it was a space image and NASA puts out all of there Hubble and James Web images in the public domain as they are part of the government, but I wanted something specific and didn't want to dig through NASAs archives of Hubble images. Maybe it's just more convenient to generate a background that fits you needs.

Now call me crazy, but I would like for Ioncom to actually have people besides myself and be an actual company one day. I just need to find an artist that is at least willing to work part time and is someone I can trust not to backstab me. I really don't want to rely on tools that might not work in the future due to it eating itself, but I'm not 100% against it. I guess I should say that I'm in favor of AI assisted art and not AI generated art.