京东开源实时视频编辑模型 JoyAI-Video-Edit，支持边播边改
京东开源 JoyAI-Video-Edit，实现实时视频流编辑，能边播边改，速度与质量兼得，或开启视频交互新范式。
京东开源 JoyAI-Video-Edit，首个同时实现流式架构、实时速度与可用质量的视频编辑模型。720P 下每秒 30 帧，支持任意时长稳定流式编辑，可对直播、视频通话等实时画面进行指令修改。此前流式编辑模型要么太小、要么质量差，该模型用 16B 参数追平离线主流。
正文摘录
 New Intelligence Report  This August, the AI video arms race has escalated again. ByteDance released Seedance 2.5, doubling single video generation length from 15 seconds to 30 seconds. The same day, MiniMax H3 went open source, priced at just one-third of closed-source flagships. The whole track is still competing on the same metric: who can generate longer, more coherent, and cheaper in one go. But there's a premise everyone in this race takes for granted: you have to wait for the video to finish generating, watch it, and only then know if you want to change it. If you do want to change it, sorry, back in the queue. Interestingly, the …