10 Best Open-Source Lip Sync Alternatives in 2026, Compared by Size

Talking Avatar Open Source lip syncs a 3D .glb avatar from any audio in the browser, with about 10 KB of gzipped code and a 1.04 MB neural model, under the MIT licence. If you need a different engine, a smaller library or a different kind of output, these are the open-source alternatives, measured from the files each project actually ships.

Full-body animation

It animates a close-up face; full-body characters with gestures need TalkingHead or an engine.

Another engine

It runs in the browser only; Unity, Unreal and native apps need a different tool.

Other kinds of avatar

It animates 3D .glb avatars; 2D mouths and photoreal video need other models.

Languages

Its model is trained on English; other languages use the nearest English mouth shape.

Download size

Kilobytes load instantly in a browser; gigabytes mean a GPU server.

Where it runs

Browser, Unity, a command line, or an NVIDIA GPU.

Licence

Check the code and the model weights: they're not always under the same terms.

What lip sync needs

Audio-only lip sync works with any voice agent. Text-driven lip sync needs word or phoneme timings from your TTS.

1. 1. Best for ready-made photorealistic avatarsOur product

   Talking Avatar Realistic Avatars

   Talking Avatar's paid product: ready-made photorealistic avatars that speak your voice agent's audio in real time with lip sync, at 832×468, 25 fps. Commercial use is included.

   Why it fits here

   • Ready-made photorealistic avatars instead of your own 3D model
   • Streamed at 832×468, 25 fps to your page over WebRTC or into your LiveKit room: nothing for the user to download
   • $0.02 a minute, billed per second, no subscription
   • Driven by your voice agent's audio: OpenAI Realtime, LiveKit, Pipecat or any TTS

   Pros

   • A realistic human face instead of a 3D model, without changing your voice stack
   • No model or avatar files for users to download
   • No plan to choose

   Cons

   • Not open source: it's a paid service
   • Ready-made avatars only: it can't use your own .glb
   • Billed per minute, where the open-source libraries here are free
   • Face only: you bring the conversation

   Pricing: $0.02/min, billed per second, no subscription.

   Best for: Teams that want a realistic human face instead of a 3D model.

2. 2. Best free full-body 3D avatar library

   TalkingHead

   Open-source JavaScript library for full-body 3D avatars with gestures, poses and moods. Lip sync comes from TTS word timestamps, or from audio with its HeadAudio module.

   Pricing: Free (MIT licence).

   Best for: Full-body characters with gestures and text-driven lip sync.

   TalkingHead vs Talking Avatar →

3. 3. Best free in-browser viseme detector

   HeadAudio (TalkingHead)

   Audio worklet that detects the 15 Oculus visemes in the browser with MFCC features and Gaussian prototypes; part of the TalkingHead ecosystem. Its README notes accuracy is far from optimal on noisy audio.

   Pricing: Free (MIT licence).

   Best for: TalkingHead users who need audio-driven lip sync.

4. 4. Best for uLipSync profiles on the web

   wLipSync

   uLipSync's MFCC approach ported to WebAssembly and Web Audio, reading phoneme weights in real time; it needs a profile calibrated with uLipSync in Unity.

   Pricing: Free (MIT licence).

   Best for: Web developers who already calibrate profiles with uLipSync.

5. 5. Best for DIY React Three Fiber scenes

   wawa-lipsync

   Small library that scores the 15 Oculus visemes from an audio element or the microphone with hand-tuned rules; you animate the character yourself.

   Pricing: Free (MIT licence).

   Best for: Developers who want full control of the animation and simple lip movement.

6. 6. Best for Unity projects

   uLipSync

   MFCC-based lip sync plugin for Unity using the Job System and Burst Compiler, with calibration profiles and VRM support.

   Pricing: Free (MIT licence).

   Best for: Unity developers replacing OVRLipSync inside the engine.

7. 7. Best for offline 2D mouth animation

   Rhubarb Lip Sync

   Command-line tool that turns a finished recording into a timeline of 2D mouth shapes for After Effects, Moho, OpenToonz, Spine and more. Batch, not live.

   Pricing: Free (MIT licence).

   Best for: Animators and game developers with recorded dialogue and drawn mouths.

   Rhubarb Lip Sync vs Talking Avatar →

8. 8. Best for high-fidelity 3D facial animation

   NVIDIA Audio2Face

   Open-sourced audio-to-face and emotion models with a C++ SDK and Unreal Engine 5 and Maya plugins, accelerated on NVIDIA GPUs.

   Pricing: Free: SDK and plugins MIT, models under the NVIDIA Open Model License.

   Best for: Game and film teams working in native engines with GPUs.

   NVIDIA Audio2Face vs Talking Avatar →

9. 9. Best open-source real-time lip sync for video

   MuseTalk

   Open-source model that re-renders the mouth region of a face video from audio, in latent space. Version 1.5 runs at "30fps+ on an NVIDIA Tesla V100", with Chinese, English and Japanese audio.

   Pricing: Free: code MIT, model weights usable commercially (bundled third-party models keep their own licences).

   Best for: Teams with a GPU who want photoreal lip sync on their own servers.

10. 10. Best open-source quality for offline dubbing

    LatentSync

    ByteDance's lip-sync model built on audio-conditioned latent diffusion. Version 1.6 is trained at 512×512 for sharper mouths and needs about 18 GB of GPU memory to run.

    Pricing: Free (Apache-2.0); hosted versions on Hugging Face and Replicate.

    Best for: Rendering dubbed or re-voiced video where quality matters more than speed.

The open-source SDK is MIT, so licensing is no reason to switch. If you want a photorealistic person instead of a 3D model, keep the same voice agent and send its audio to a ready-made realistic avatar.

1. Create a Talking Avatar Realistic Avatars account, add credits, create an API key and pick a ready-made avatar.
2. Keep your voice agent; send the same audio through the server or browser SDK in the same npm package.
3. Top up credits, or turn on auto top-up; sessions bill at $0.02 a minute, per second.

• A photorealistic face without changing your stack: Talking Avatar Realistic Avatars.
• Full-body 3D avatars in the browser: TalkingHead.
• The smallest audio-driven viseme model under MIT: HeadAudio.
• The smallest code, and you'll tune the animation yourself: wawa-lipsync.
• uLipSync profiles on the web: wLipSync.
• Unity: uLipSync.
• Offline 2D mouth animation for recorded dialogue: Rhubarb Lip Sync.
• High-fidelity 3D faces on an NVIDIA GPU: NVIDIA Audio2Face.
• Photoreal video lip sync on a GPU server: MuseTalk (real time) or LatentSync (quality).