Real-Time Video Editing with JD’s Open‑Source JoyAI‑Video‑Edit Sets Global Performance Lead

Real‑time streaming video editing

JoyAI‑Video‑Edit processes 720p video at 30 fps, matching typical playback speed while preserving visual clarity.

It supports stable streaming for arbitrarily long videos, enabling continuous scene replacement, object swapping, character changes, or style adjustments without waiting for the whole clip to finish rendering.

Evaluation

Compared with representative streaming editors SANA Streaming, LiveEdit and Xmax‑X2.0 on a benchmark covering global‑style transfer, local replacement, deletion and subtitle editing, JoyAI‑Video‑Edit achieved uniformly higher scores, matching offline‑model quality while retaining streaming speed.

In the OpenVE‑Bench suite the model outperformed all existing streaming methods, delivering a large margin in every task, demonstrating accurate edit intent understanding and stable results.

Applications

Live fashion shows: presenters can switch outfits on‑the‑fly.

Interior‑design previews: instantly show different furniture, walls, lighting.

Film production: experiment with costumes, environments, visual effects in real time, reducing re‑shoots.

Embodied‑intelligence data synthesis: convert manually captured manipulation videos into robot‑hand or robotic‑arm footage while preserving object positions, spatial relations and motion trajectories, enabling large‑scale training sample generation.

Open‑source release

Model and code are available at https://huggingface.co/jdopensource/JoyAI-Video-Edit and https://github.com/jd-opensource/JoyAI-Video-Edit.