JoyAI-Video-Edit targets real-time 720p video editing
The 16B-parameter autoregressive diffusion system edits open-ended video at about 30 FPS on one Nvidia B200.
Why it matters
The work points toward lower-latency generative video editing systems that can operate causally on ongoing streams, rather than relying on full-video offline processing. If reproducible, it could make real-time AI video editing more practical for interactive and live workflows.
The key points
- 1.16B-parameter model edits video without future-frame access.
- 2.System reports 720p editing at about 30 FPS on one Nvidia B200.
- 3.Evaluations show gains over streaming editors and competitiveness with offline systems.
Researchers presented JoyAI-Video-Edit, a 16B-parameter autoregressive diffusion framework for real-time, open-ended video editing without future-frame access or a predefined video duration. The system combines chunk-wise autoregressive adaptation, Source-Anchored Distribution Matching Distillation and Long-Horizon Autoregressive Distillation to preserve source fidelity and reduce temporal drift. Reported automatic and human evaluations found it outperformed existing streaming editors and remained competitive with strong offline systems, while achieving end-to-end 720p editing at about 30 FPS on a single Nvidia B200 GPU.
⚡ Try this today
Review the paper and code before building real-time video editing workflows that require causal, open-ended generation.
Sources & original reporting
This brief summarizes and links to reporting from the publishers below.
Enjoyed this brief? Get the next one in your inbox.
More in Research
Google advances private AI with homomorphic encryption
Google says it is making private AI more practical using homomorphic encryption.
Google says it is making private AI practical
A Google Security post drew Hacker News discussion alongside broader AI critiques.
Researchers release LittleLearner sandbox
A 5B-parameter model and 88B-token Grade 5-filtered corpus test how training scope limits capability.