Griffin模型通过视频图灵测试
Tavus的Griffin让近半人误认为真人对话,全双工技术实现实时互动,远超同类系统表现。
Tavus团队开发的Griffin模型让48%的实时对话者误以为是真人,远超同类系统不足3%的通过率。该模型在NVIDIA全双工AI视频基准(VideoFDB)的感知与生成两项均排名第一。
Griffin能实现全双工对话,在说话的同时持续倾听和观看,并生成整个画面而非仅换脸。
首个通过「视频图灵测试」的模型:Griffin 来自 @tavus 团队的 Griffin 让 48% 的实时对话者误以为对面是真人,在此之前同类系统不足 3%,他们还在 NVIDIA 全双工 AI 视频基准(VideoFDB)上感知与生成两项均排名第一,厉害! 全双工对话是最关键的技术差异点 传统语音 AI 是“回合制”的:说完才听。Griffin 能在说话的同时持续倾听和观看:用户 Mars 讲述无聊约会时,它会实时笑、点头、插话提问,而不是等她说完。 生成整个画面,不只换一张脸 Tavus 明确强调 Griffin "generates the whole frame, not just a face"。Simon Says 演示中它要同时听指令、看手势、做全身动作。这区别于市面上大多数“唇形同步 + 换脸”式数字人。 视觉实时融入对话 用户举起孔雀玩具,Griffin 即兴编入故事;教魔方时能看着用户的手实时指导,并在对方思考时安静等待,时机感本身就是智能的体现。 时间与情境感知 焊接指导演示中,Griffin 能为动作计时,“时间也是对话的一部分”,这暗示模型对物理世界节奏有内在感知。 技术背景补充 Griffin 并非单点突破,它是 Tavus 四模型体系的统一层:Sparrow-2(对话轮次理解,TurnBench 第一)+ Phoenix-4.5(实时人物渲染,宣称 134ms 延迟、单图零样本生成新身份)+ Raven-1(语音/视频/屏幕三路多模态感知,上下文刷新 <300ms),Griffin 将三者整合为实时交互整体。 Tavus @tavus Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Griffin, Tavus, multimodal.
Source attribution: curated from shao__meng (@shao__meng). Cross-source corroboration: 9 independent sources have covered this development (including rohanpaul_ai, Replicate, elvis), confirming industry-wide signal strength.