Read the original at HF Daily Papers
Researchers propose GRACE, a framework that compresses video autoencoders while maintaining compatibility with pretrained diffusion transformers to accelerate video generation.
Carried by: HF Daily Papers. First seen: .