JoyAI-Video-Edit is a real-time streaming video editor from JD (Jingdong) for e-commerce teams and creators who need to change a video while it plays instead of re-rendering it offline. Open-sourced on 5 August 2026, it edits faces, clothing, scenes, and styles on the fly through natural-language commands.
Core Features
- A 16B-parameter autoregressive diffusion architecture that runs at 720p and 30 frames per second with under half a second of latency.
- Any-length streaming edits: unlike earlier tools limited to seconds or minutes, it keeps pace with multi-hour live video.
- Natural-language edits such as swapping an outfit to pink, adding a red cap and sunglasses, or turning live footage into an animated style.
- Reference-image driven editing that copies a look onto the video subject, plus global style transfer and local object replacement.
- Open weights and code on Hugging Face and GitHub, with a free commercial API for merchants.
Use Cases
- Live commerce hosts who swap outfits, products, and backgrounds mid-stream without stopping the broadcast.
- Home-design videos that trial different furniture, walls, and lighting in real time.
- Film and robot teams that synthesize training data by converting human action clips into robot-arm footage.
Pricing
JoyAI-Video-Edit is open-source under a permissive license with weights free to deploy, and JD ships a free commercial API tier for developers and merchants. There is no paid paywall for the base model; costs come from your own GPU hosting. Enterprise support is available for high-volume retail use.
Our Take
Best for retail and live-commerce teams that already live in JD's ecosystem and want instant on-screen edits. The trade-off is that it is an editing model, not a generator, so you still need source footage, and the 720p ceiling trails offline cinematic tools. Compare it with other AI video tools.
