OUR MISSION
We exist to break every barrier to understanding.
Where this comes from.
Our CEO, An Zhu Liu, arrived at a school where he couldn't speak English to his classmates. He was the only Asian person in the entire school. He had no friends. He learned the language from watching TV; the screen was where English finally reached him. It never left him.
That is why Familiar exists: so what plays on a screen reaches everyone, in the language they understand.
We don't replace artists. We translate them.
Language has always been the wall between people.
- A parent who can't talk to their kid's teacher.
- A patient who can't describe their symptoms.
- A witness who can't be understood in the language the court runs in.
Machine learning for the tedious part.
- Your footage, your voice, your face. Nothing new is made up.
- No prompts. Not randomly generating new.
- All of the data we trained with ourselves are actually generated by the Familiar team.
Not generative AI, but translation to enable creatives.
Multimodal translation does it in one unified model with scene & world understanding: it translates the words, carries the voice, syncs the lips and keeps the scene sound (the laughter, the music, the ambience), into 30 languages from 99 source languages.
Spider-Verse set the precedent: Sony Pictures Imageworks trained a model on its own artists' line work to repeat the tedious part.
We're another tool for the creative process.
We're starting where the wall costs the most: video.
Human dubbing isn't usable for most of the world: even top productions dub only a few languages, because of the cost, the people and the time. Smaller productions, shows and vertical dramas don't dub at all.
Proven over years of operating international channels for Veritasium, 3Blue1Brown, and Welch Labs, at the company our CTO led.
Our Vision: A world where every person is understood.
We stand against AI slop & deepfakes.
Researched and Built with
We're building multimodal translation: the voice, the lips, the scene and the person, so actors and creators reach new audiences in any language.
The voice and the face from one model, so identity can never split: the field we started with JAM-Flow.
The most realistic real-time humans today: SoulX-LiveAct animates a person at 20 frames per second, stable for an hour.
Distillation, guidance, and sampling: the research that makes diffusion models fast enough for real time.
Editing inside frozen diffusion models, from finding h-space to controlling style and removing on-screen text.
Reading the scene before editing it: who is talking, from visual and audio together, in any environment.
