THE TEAM

The Team That Walked Away from Avatars

FAMILIAR LABS RESEARCHALL ARTICLES

ABSTRACT

The money in human video went into avatars: new faces and voices for software. The researchers who started joint audio-visual human models went the other way: Familiar Labs carries the performance a person already gave into 13 languages, voice and face intact.

Who is on the team?

FULL TIME

7

in Toronto

PAPERS

22

every one in the research overview

PEER-REVIEWED

18

ICLR, CVPR, ECCV, NeurIPS and more

CITATIONS

Over 10,000

since 2021

Three co-founders, Andrew Liu (CEO), Mingi Kwon (CTO) and Jibin Song (Chief Scientist), with a Principal Researcher and Members of Technical Staff; Prof. Xu Zheng and Prof. Youngjung Uh advise. Their JAM-Flow, at ECCV 2026, is the first joint audio-visual human model; every paper is in our research overview.

Where did the money in human video go?

Into avatars: presenters for training videos, agents you talk to face to face.

AVATAR AND DIGITAL-HUMAN ROUNDS, AS REPORTED

Synthesia
ROUND
$200M Series E at a $4B valuation
ANNOUNCED
January 26, 2026
Nuance Labs
ROUND
$50M Series A
ANNOUNCED
September 14, 2026
Tavus
ROUND
$40M Series B
ANNOUNCED
November 12, 2025
Hedra
ROUND
$32M Series A
ANNOUNCED
May 15, 2025
HeyGen
ROUND
$60M Series A at a $500M valuation
ANNOUNCED
June 2024
DeepBrain AI, Seoul
ROUND
$44M Series B
ANNOUNCED
August 2021
Silicon Intelligence, Nanjing
ROUND
Over RMB 1 billion across more than ten rounds
ANNOUNCED
Reported July 5, 2025
AVATAR AND DIGITAL-HUMAN ROUNDS, AS REPORTED
COMPANYROUNDANNOUNCED
Synthesia$200M Series E at a $4B valuationJanuary 26, 2026
Nuance Labs$50M Series ASeptember 14, 2026
Tavus$40M Series BNovember 12, 2025
Hedra$32M Series AMay 15, 2025
HeyGen$60M Series A at a $500M valuationJune 2024
DeepBrain AI, Seoul$44M Series BAugust 2021
Silicon Intelligence, NanjingOver RMB 1 billion across more than ten roundsReported July 5, 2025

Good companies, each building what it believes in; the sources are in the references.

Why did the team walk away from avatars?

++

We never generate a person, and we never replace an actor.

A team that started joint audio-visual human models could build avatars as well as anyone; it chose the opposite of making a new human. A photoreal stand-in for an actor feels empty, because once we know it is AI, the empathy drops (Human to Human, Not Avatars).

What does the frontier look like from this side?

Carrying a person is its own research problem; the published studies measure how close the team has come.

FIG. 01 · THE PUBLISHED STUDIES · AUGUST 2026
182.9%

Higher speaker resemblance than YouTube auto-dubbing

50 PAIRS · 5 LANGUAGES · FROZEN AUGUST 8, 2026
70.3%

Fewer important ideas changed or lost than YouTube auto-dubbing

THE SAME 50 PAIRS · AUGUST 8, 2026
100.3%

Of expert human translators' quality: a match

FRENCH · CHINESE · HINDI · INDONESIAN · AUGUST 2026

What is the team pushing next?

  1. TODAY

    Dubbing in 13 languages

    Up to 4K, in 5.1 or stereo; the translation choices yours to review.
  2. IN THE WORKS

    The text in the frame, and more of the choices

    Signs and screens localized; the audio and visual decisions yours to override.
  3. IN THE WORKS

    Live broadcasts

    Sport, news and streams.
  4. LATER

    Two-way, in meetings and in the room

    Wherever a human interpreter can be.

REFERENCES

  1. [1]Synthesia raises $200M Series E led by GV (Built In, January 26, 2026)
  2. [2]Nuance Labs bags $50M in funding (SiliconANGLE, September 14, 2026)
  3. [3]Tavus raises $40M Series B led by CRV (Scale Venture Partners, November 12, 2025)
  4. [4]Hedra raises $32 million (Cartoon Brew, May 2025)
  5. [5]HeyGen raises $60M valued at $500M (Maginative, June 2024)
  6. [6]DeepBrain AI raises $44M Series B (DeepBrain AI, August 2021)
  7. [7]Silicon Intelligence completes its Series D (36Kr, July 5, 2025)
  8. [8]JAM-Flow (ECCV 2026)
  9. [9]Familiar Labs vs YouTube auto-dubbing: the frozen-cohort study

Multimodal Translation

Familiar Labs builds generationally better dubbing with a Multimodal Localization Foundation Model (MLFM), the first foundation model for multimodal translation: one model decides what to localize across text, audio and visual, and shows you the choices. Your voice and lips, carried into new languages. They're a team of 7 with PhDs, backed by Samsung, RedBird, Gold House. Our mission

BACKED BY

  • Wheelhouse
  • Maximum Effort
  • Stanford
  • Yonsei
  • University of Waterloo