AI 精选动态
智能评分 60
JoyAI 开源实时视频语言交互模型预览
AI 推荐理由
该模型提供了完整的实时视频交互开源方案（模型、训练配方、数据、部署系统），值得关注和尝试。核心解读
JoyAI 团队发布 JoyAI-VL-Interaction-Preview 模型，8B 参数，开源 Apache 2.0 许可。在实时视频监控与警报场景中，与 Doubao 和 Gemini 视频通话助手进行人类成对比较，胜率 100%。训练数据为 4M+ 时间对齐的秒级标注片段。
全文
👀 JoyAI-VL-Interaction-Preview just landed on ModelScope! An open 8B model for real-time video-language interaction. License: Apache 2.0🚀
👉 Try it now: https://t.co/RhVf9MErq7
📄 Paper: https://t.co/STkBjMzeMe
✨ Real-time presence: built for live video scenarios where the right answer has to arrive at the right moment, not after a user prompt
🚨 Strongest zone: wins 100% of human pairwise comparisons on monitoring and alerting against both Doubao and Gemini video-call assistants
🧠 Interaction training: trained on 4M+ time-aligned clips labeled second by second for speak, stay silent, or delegate
🛠️ Open stack: releases the 8B model, training recipe, data, and deployable system for building always-present visual assistants
https://video.twimg.com/ext_tw_video/2068945220020908032/pu/vid/avc1/1282x720/vCCcACHsx4juWWyB.mp4?tag=12