VFusion3D: Learning Scalable 3D Generative Models from Video Diffusion Models
github.comI tried a few samples, I feel that its quality is not as good as stable fast 3d.
I'm guessing this paper is more about "It's neat that this works at all." rather than trying to improve on the state of the art.
And stable fast 3D main point wasn’t quality but speed.
Interesting that they compare to totally different models than Stable Fast 3D’s paper. I’m not clear on relative size of VFusion3D vs Stable Fast 3D, but I do think the training idea is good and novel — getting relatively good quality out of movies is a much easier ask in terms of collecting training data than getting 3D model renderings.