sync. labs is a research driven company modeling how humans perform and interact through video. studio-grade visual dubbing and dialogue editing are the first applications of our work.
- 2023founded
- 100,000+hours edited
- 32team members
studio-grade visual dubbing
adapting an actor’s performance across languages.
visual dialogue editing
rewriting a dialogue in post-production without a reshoot.
we train audiovisual generative models that learn how a character delivers a line in a given shot, and regenerates their facial expressions to sync with any new dialogue from any given audio stem while preserving their performance.
our models are language agnostic and perform across languages, dialects, and speakers without having to train or fine tune on any extra footage or audio.
built to support professional codecs up to 4k 16-bit HDR, with delivery in ProRes 4444XQ or even raw, uncompressed EXRs.
sync. labs is working with production teams across the United States, Korea, India, and Mexico to rewrite dialogue in post-production.
at the core of our workflows sits studio-1, the world’s most advanced lipsync and facial reanimation model built for professional video. it has the ability to reason across a video, and intelligently edit what needs to be edited while preserving what doesn’t across complex scenarios and edge cases.
sharp anglessharp angles and side faces
preserve emotionsacting and emotion preservation
detailsclose ups and details
multiple speakers
low lighting
with state of the art workflows around voice cloning and audio dubbing, take any story and adapt it perfectly to every user, while keeping the original creative intent intact.
when you work with sync. productions you get a guaranteed result. we take delivery of your files and work with your post-production teams and localization partners to pass mastering QC and launch on platform.
today, we teach models to understand and preserve the relationship between how we speak and how we perform. we’re expanding our capabilities across video, voice, and text.
*no infringement intended, research preview*
