There’s a pretty solid article that walks through the evolution of image generation models — starts from the early VAE days and goes all the way up to Diffusion. It basically connects the dots on the whole AI art tech lineage. As someone who’s just generating images with these tools every day, I usually just tweak prompts and settings to get the results I want, never really look back at how this whole thing evolved step by step. After reading it, I found it pretty interesting — turns out all these smooth, high-quality outputs we get now are built on a mountain of failed attempts from earlier approaches.
I’m not a tech guy by background, so a lot of the details went over my head, but I could follow the general storyline. Is there anyone here who actually knows their stuff and can break it down in plain English? Like, what was the real bottleneck between VAE and Diffusion models, and why did the latter suddenly blow up in quality?