
JoyAI-Video-Edit brings real-time video editing
JD’s open-source 16B autoregressive diffusion model edits live or uploaded video from natural-language instructions, without requiring the full clip or a predefined duration. It delivers reported 720p throughput of 30.19 FPS on a single Nvidia B200 GPU under Apache 2.0.
JoyAI-Video-Edit pushes video editing closer to an interactive capability than an offline post-production step, though its hardware requirements remain far beyond typical developer workstations.
- –Supports object manipulation, appearance changes, background replacement, stylization, motion edits, and reference-guided editing
- –Causal generation helps preserve subject identity, motion, and temporal consistency as new frames arrive
- –Reported benchmarks place it ahead of streaming-editor baselines including XMax-X2.0, SANA-Streaming, and LiveEdit
- –The Apache 2.0 release enables commercial experimentation, self-hosted workflows, and research integration
- –Nvidia B200 testing makes the 30 FPS claim impressive but limits immediate accessibility on consumer hardware
DISCOVERED
2h ago
2026-08-16
PUBLISHED
2h ago
2026-08-16
RELEVANCE
AUTHOR
AI Search