Just pulled up the repo and it turns out it’s the same guy who made Fooocus, which is still the best Stable Diffusion setup I’ve ever seen.
Incredible.
Yeah it’s awesome.
There’s a lot of optimization left on the table, too. Check out this fork, which doesn’t quite work yet (seems a python config file is missing?) but is integrating some optimizations (like torch.compile) and extra options (like last frames or negative prompts): https://github.com/wongfei2009/FramePack
Can someone please make and then share an example?
If generating ai content is so easy now, and deepseek can run locally on anything, what are all those datacenters working on?
training but at the same time having 100 million people daily asking a variety of questions will put up a server load
