State of open video generation models in Diffusers

look, the open video space is just a huge dump of poorly optimized checkpoints. 'State of the art' means nothing if the models require 80GB of VRAM just to run a 4-second clip. diffusers makes the plumbing easier, sure, but it doesn't fix the fundamental computational bottleneck. we're drowning in demos, but few actually have production-grade, fast, and cost-effective pipelines.