JD.com Open-Sources JoyAI-Video-Edit for Real-Time Streaming Video Editing

JD.com has open-sourced JoyAI-Video-Edit, a real-time streaming video editing model that allows users to modify characters and scenes while the video is playing. The model supports arbitrary-length videos and achieves an inference speed of 30 frames per second at 720p resolution. This model shifts video creation from post-processing to real-time interactive editing, opening up new possibilities for live streaming, interior design, and synthetic data generation for embodied AI. It outperforms existing models like LiveEdit and SANA Streaming on benchmarks like OpenVE-Bench. Built on the proprietary JoyAI-Video model, JoyAI-Video-Edit can convert human hand movements in videos into robotic arm actions while preserving spatial relationships, which helps scale up robot training data. The code and weights have been made publicly available on GitHub and Hugging Face.