Familiar

About

Audio-visual translation infrastructure for all video, live and recorded -- the first and only that works in real time. World translation: not just changing voice and lips into a new language, but understanding the scene, the sound, and preserving the human. 🦋 Movies & TV, Micro-dramas, Live Broadcasts, Live Shopping, Corporate & Education, Ads, Creators & Podcasts, Platforms. Day-and-date, in every territory. Founders An Zhu Liu, Jibin Song, Mingi Kwon, and Xu Zheng lead a team of PhDs, professors, and researchers from the world's top universities who started the field of joint audio-visual human models and co-authored the current state of the art in real-time human animation. Since 2021: over 10,000 citations and 100+ papers at ICLR, NeurIPS, CVPR, ECCV, and more. Translation has to be a world model. A translator has to know who's speaking: where every sound in the dense polyphony originates (laughter, layered crowds). Laughter here, a short yell there, a voice from off-screen, someone on screen mouthing words before being interrupted. Everyone else scaffolds separate models and pre-processing; a world model simulates the scene. It's a spatio-temporal reasoner: predicting continuity across frames forces it to internalize 3D geometry, spatial reasoning, and permanence -- of objects, people, and the environment itself. The scene state survives the edit: when the shot pans or cuts away, someone off-screen still exists, still owns their voice, and is still there when the camera comes back. Multiple people, side views, off angles, dense polyphony, fast motion -- the hard cases hold. Vs ElevenLabs: they lose the laughter, the music, the ambiance -- 270% more background-sound error -- and ours sounds 27.8% more like you. That means no recasting, no ADR, no M&E stems required -- it works from the final mix. Vs human translators: 100.3% of their quality -- scalable, real time. Published benchmarks: https://thefamiliarlab.com/benchmark

Tags