VEED's Lip Sync 2.0 launches as an API on fal
By Steven Van ·
The model handles 4K clips up to 10 minutes with adjustable sync strength, no fine-tuning needed.
The lip-sync model making the rounds this week now has a programmatic home: Lip Sync 2.0 from VEED is live on fal. Swap the audio on any video and it automatically matches the lips, emotion, style, and timing, with no training or fine-tuning required, exposed as a fast API developers can drop into their own pipelines.
The specs that make it usable
Three numbers matter. It processes footage up to 4K, handles clips up to 10 minutes, and exposes controllable sync strength so you can dial how aggressively it conforms the mouth to the new audio. That last control is underrated, too much and the face looks rubbery, too little and it drifts out of sync, so making it a parameter rather than a fixed behavior is what lets developers tune quality per use case. And because no fine-tuning is required, it's a single API call, not a training job.
Emotion and style, not just mouths
Lip Sync 2.0's pitch is that it matches emotion, style, and timing, not only lip shapes. That's the difference between dubbing that preserves a performance and dubbing that hollows it out. On an API, that quality bar means products built on top, dubbing platforms, localization pipelines, avatar tools, inherit a believable result instead of the uncanny mouth-flap that made earlier lip-sync APIs unusable for real content.
One model, every surface
fal's hosting completes a pattern that played out in a single day: the same VEED Lip Sync 2.0 also became the default in Magnific's Speak feature. This is how frontier media models propagate now, the creator-facing platform wraps it in a UI, fal exposes it as an API, and developers and end users each get the version that fits them. For anyone building video software, the availability on fal means high-quality dubbing is now a component you call, right alongside fal's LTX Reframe and the rest of its media catalog.
Try it
VEED Lip Sync 2.0 is live on fal now, in the model gallery and via API. Test it on a clip with clear emotional delivery and experiment with sync strength to find the setting that keeps the performance intact.