/Patent Tracker
NvidiaPatent applicationAI & softwareInterest 6/10

Nvidia patents video AI that keeps motion smooth across frames

Patent drawing
Patent drawing from the official filing.

Nvidia describes a method for training video generation models to handle movement more consistently. Instead of adding random static to each frame separately, the system shifts the noise pattern to match how objects move, ensuring the visual 'grain' stays aligned with the action.

Why it mattersThis technique could help AI create video where moving objects look stable and realistic rather than flickering or shifting unnaturally between frames.

In the filing’s own words

Two and three dimensional video diffusion models configured via training to be equivariant to spatial transformations of the input by training the model using warped noise, where equivariance of the models is learned using a conventional video diffusion loss function modified to replace independent noise inputs with warped noise inputs.

What exactly they claim (claim 1)

1. A process comprising: generating a warped noise input by progressively applying a warping operator to noise added to video frames; and training a video diffusion model using a video diffusion loss function modified to utilize the warped noise input.

Facts

Company
Nvidia (NVIDIA Corp.)
Official title
VIDEO DIFFUSION MODELS WITH TEMPORALLY CONSISTENT NOISE
Publication number
US20260301303A1
Application number
19575327
Filed
Mar 23, 2026
Published
Oct 1, 2026
Inventors
2
Claims
21

Classified under

G06T 15/04 Physics & computingG06T 15/20 Physics & computingG06T 17/20 Physics & computing
Read the full filing (official PDF) ↗ Open in the Patent Tracker

More from Nvidia in AI & software

Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.