This is a smart move and goes to show people move so quick on this. I wonder how long before these models overtake stock image websites, get used in games or video production, generate T-shirt prints and whatever other usecase people can come up with.

On that note: Iirc back when I learned about GANs, when they were still rather new there was a clear path towards improvements of them by making the model bigger and feed it _much_ more high quality training data. When I look at the outputs of stable diffusion or Dall-E there are still often visible artifacts and most prominently faces are often weird. Is there a clear path towards improvement here aswell or are we hitting a wall with the "just more" paradigm somewhen?