NVIDIA just released AnyFlow on Hugging Face The first any-step video diffusion model that generates high-quality text-to-video with any inference budget - 4 steps or 50, quality scales smoothly without degradation.

Model: Project page:

@_akhaliq Thanks for the update

Nvidia is literally changing the game for video generation speed now.

The flexibility on inference steps is what makes this genuinely useful in production — you can tune speed vs quality at runtime rather than committing at architecture design time. Curious if anyone's benchmarked wall-clock diff between 4 and 50 steps on consumer hardware yet.