watch demo
ProRes support
4444 XQ, 4444, 422 HQ, 422, 422 LT, 422 Proxy
OpenEXR image sequences
UHD and 4K DCI resolution
lossless preservation
bitrate, metadata, and resolution across video and image sequence deliverables
HDR10 and ACES color pipeline
alongside Rec.709
accelerated turnaround
compared to traditional visual dubbing workflows
flexible delivery
client-side storage integration or managed private S3 infrastructure
workflow — localization
sync. visual dubbing
localization that looks like it was originally shot in the target language — shot-by-shot, every frame.
where we integrate
pre-color grade, post-VFX
conformed, VFX-final before the DI.
post-DI, pre-release
finished master after the digital intermediate.
QC rounds
round 1
visual dubbing pass
perfecting sync across every shot.
round 2
near-final frame
tightening to as close to final frame as possible.
round 3
final delivery
last notes in, individual shots delivered.
workflow — adr
sync. dialogue replacement
flexible, iterative ADR integration built around how post-production teams actually work — quick turnarounds, direct pipeline handoff.
process
01
receive assets
EXRs with handles, OV reference, and target language audio.
02
proxy delivery
proxy render for QC review — fast turnaround.
03
implement feedback
QC notes actioned in one focused revision cycle.
04
final EXR delivery
lossless, frame-accurate output into your pipeline.
EXR
final delivery format
proxy-first
review before commit
iterative
QC-driven loop
functional understanding of the scene
sync-3 is a foundational model with world knowledge. it understands lighting conditions, obstructions, and scene context natively. the result is natural visual dubbing no matter how complex the shot.
no 3d meshes or hand-crafted features
unlike other approaches that rely on 3D mesh reconstruction with manual feature engineering, sync-3 is trained by watching millions of people speak. the result: realistic mouth and facial motion for any new speaker, in any context.
language agnostic
sync-3 is trained on hundreds of languages and accents, enabling it to generalize to unheard language variants without additional training.
performance preservation
The model observes each speaker within every shot and understands their style on-the-fly: unique movements, idiosyncrasies, expressions.
zero-shot, any speaker
we require no character-specific training and deliver near-instant visual dubbing on any video, any character, and any shot. this is what enables the speed and scalability our pipeline requires across full seasons and features.
flexible performance control
sync-3 supports the full range of production needs, from silent lips to moving lips (or vice versa), preserving original expressions or adjusting them as needed. the model uses both input video and audio as context to produce performance-driven results.