>>941268292
>It's moving fast
This really isn't true. Look at the sublimation moments for image diffusion methods. From 2019 to 2020 you went from Google Deepdream to SD. This was the first time convincing anatomy was possible.
Then from 2020 to 2023 you got SDXL, which is the basis for all models extant today. Even alternative methods like Chroma or Flux are doing the same level of complexity just with different hardware demands at the end of the day.
The state of SD is actually very stagnant. Compare it to video diffusion, where you have new breakthroughs almost monthly. Images are harder to commercialize so there is no real reason to push for another sublimation moment for image generation. Any improvements now are just end user tweaks that make an now years-old base more accessible.