So, I actually like generative AI (disclaimer I feel I have to include every time: local open models only), and my main problem with that image is how genericized the new face is. If you’ve seen a lot of AI images, it’s immediately recognizable as the default mixed Asian/Caucasian face you get when not prompting something more specific than “woman” due to the datasets dominating the training data. It heavily implies all faces will be similarly genericized.

I don’t think this tech will be viable unless creators can give the AI a reference image of what a character should look like when photorealistic, and that’s just going to increase the workload of running this in realtime.