There are news articles about this problem going back to at least 2022
But people have known it well before 2022. The observation that hands appear uniquely hard probably goes back to the BigGAN ImageNet Generator released in late 2018 and widely used on Ganbreeder; that was the first general-purpose image generator which was high-quality enough that people could begin to notice that the hands seemed a lot worse. (Before that, the models are all too domain-restricted or too low-quality to make that kind of observation.) If no one noticed it on Ganbreeder, we definitely had begun noticing it in Tensorfork’s anime generator work during 2019-2020 (especially the TADNE preliminaries), and that’s why we created hand-tailored hand datasets and released PALM in June 2020.
(And I have been telling people ‘hands are hard’ ever since, as they keep rediscovering this… I’m still a little surprised how unwilling generator creators seem to be to create hand-specific datasets or add in hand-focus losses like Make-a-Scene’s focal losses, considering how once SD was released, complaints about hands exploded in frequency and became probably the single biggest reason that samples had to be rejected or edited.)
But people have known it well before 2022. The observation that hands appear uniquely hard probably goes back to the BigGAN ImageNet Generator released in late 2018 and widely used on Ganbreeder; that was the first general-purpose image generator which was high-quality enough that people could begin to notice that the hands seemed a lot worse. (Before that, the models are all too domain-restricted or too low-quality to make that kind of observation.) If no one noticed it on Ganbreeder, we definitely had begun noticing it in Tensorfork’s anime generator work during 2019-2020 (especially the TADNE preliminaries), and that’s why we created hand-tailored hand datasets and released PALM in June 2020.
(And I have been telling people ‘hands are hard’ ever since, as they keep rediscovering this… I’m still a little surprised how unwilling generator creators seem to be to create hand-specific datasets or add in hand-focus losses like Make-a-Scene’s focal losses, considering how once SD was released, complaints about hands exploded in frequency and became probably the single biggest reason that samples had to be rejected or edited.)