01
ARKit 52
The standard names used by Apple ARKit, Ready Player Me, Character Creator and most avatar tools.
Lip sync API · early access
Lip sync for your own 3D characters.
The network that moves our hosts’ mouths, as a service of its own. Send what your character says, with its audio or phoneme timings, and get ARKit mouth curves fitted to your model, ready to play at 30 fps.
Running live
in your browser
in your browser
- 29
- ARKit mouth curves per frame, at 30 fps
- 512 KB
- the whole model; runs in the browser, Node or Python
- Real time
- a few milliseconds per sentence, on a CPU
- Your rig
- curves calibrated to the shapes your model has
01How it works
One call. Curves that fit.
Most lip sync gives every face the same generic shapes. We calibrate to the blend shapes your model actually has, so the mouth reads right on your character, not just on ours.
- 01
Upload your model, once
GLB, FBX or VRM. We read its blend shapes, map them to ARKit names and measure how strong each one is, so a weak pucker or a strong jaw still reads right.
- 02
Send the speech
The text, plus its audio (any voice: ElevenLabs, OpenAI, a recorded actor) or the phoneme timings your text-to-speech already gives you.
- 03
Play the curves
One row of weights per frame, as JSON, CSV or a glTF animation. Drive your morph targets with it in three.js, Unity, Unreal or your own engine.
POST https://api.lobbyface.com/v1/lipsync
{
"model_id": "mdl_…",
"text": "Hi, I'm Riya. I welcome your visitors the moment they arrive.",
"audio_url": "https://…/line.wav",
"format": "arkit",
"fps": 30
}{
"fps": 30,
"duration": 3.37,
"curves": ["jawOpen", "mouthClose", "mouthFunnel", …],
"frames": [
[0.47, 0.008, 0.002, …],
[0.562, 0.007, 0.002, …],
…
]
}Early access · the request shape may still change · numbers are real output
02Output
The shapes your rig speaks.
Pick the format your engine already uses. Every format comes from the same timing, so switching rigs doesn’t change how the speech feels.
02
Your model’s shapes
Only the blend shapes your character has, calibrated from the upload.
03
Oculus 15 visemes
A drop-in path for projects built on Oculus Lipsync, which Meta has retired.
04
VRM A I U E O
The five vowel shapes of VRM avatars and VTuber tools.
03Who it’s for
Your character. Your voice.
- Web and three.js developers building talking characters
- AI companion and assistant apps with their own voice
- VTuber and VRM avatar tools
- E-learning and training with animated presenters
- Game and XR studios (Unity and Unreal plugins later)
Where it stands
Today the network runs inside our own hosts, on the visitor’s device. We are opening a hosted API and a browser package to a first group of developers, English first.
- jawOpen
- mouthPucker
- mouthClose
- mouthSmile
- + 25 more
04FAQ
Developer questions.
Does it listen to the audio?
It works from phonemes: which sound is said, and when. Send audio with its text and we align them on our side to get those timings; if your text-to-speech already gives timings, send them and skip that step.
Which voices and languages?
Any English voice, synthetic or recorded. Other languages need their own phoneme set and training; tell us which you need.
Is it fast enough for real time?
A sentence takes a few milliseconds, so the curves are ready before the audio starts playing. Streaming for live conversation is on the roadmap.
Can it run on our side?
Yes. The model is 512 KB and runs in the browser, Node or Python. A self-hosting licence is planned for teams that want it on their own servers.
Will it fit an Oculus Lipsync or Ready Player Me setup?
That is what the output formats are for: Oculus’s 15 visemes, the ARKit names Ready Player Me avatars use, and VRM’s five vowels.
What will it cost?
Early access is free. After that we plan a free monthly allowance, then pay per minute of speech, with a self-hosting licence for larger teams.
What happens to our audio and models?
Audio is deleted after processing. Models are kept only while you keep their model id, and are never used for anything else.
Tell us about your character.
Your rig, your voice and where it runs. We’ll set you up in the first early-access group and test the curves on your model.