NVIDIA Audio2Face-3D
Facial animation and lip sync generated from speech audio
- Best for
- Teams with NVIDIA GPUs who want lip sync and facial animation for dialogue, baked in Maya or Unreal or generated at runtime
- Makes
- Facial animation (mesh deformation, joint transforms or blendshape weights) from recorded or streamed speech
- Made by
- NVIDIA
- Where you use it
- Windows, Linux, Unreal Editor, Autodesk Maya
- Engines
- Unreal Engine
- Pricing
- Not published; the SDK, plugins and training framework are free and open source, and the models are open weights (as of November 2026)
- Free plan
- Yes
- Open source
- Yes (MIT (SDK, Maya and Unreal plugins); Apache-2.0 (training framework); NVIDIA Open Model License (models))
- Export
- Maya ACE exports animation as native Maya keyframes or FBX; the Unreal plugin drives a character's face rig directly in the engine
- Status
- Available
NVIDIA Audio2Face-3D generates 3D facial animation from speech. Besides lip sync, it animates the jaw, tongue, eyes and facial skin, and adds expression based on the emotion it detects in the voice. It takes recorded audio or a live stream, and its output can be applied to a character as mesh deformation, joint transforms or blendshape weights, according to the project's GitHub hub README. It is the facial animation part of NVIDIA ACE, NVIDIA's wider set of tools for AI characters.
NVIDIA open-sourced the stack on September 24, 2025, according to a post on the NVIDIA Technical Blog. It includes a C++ SDK for adding it to your own tools or game runtime (MIT license), the Maya ACE plugin for Maya 2024 and 2025 on Windows and an Unreal Engine 5 plugin for UE 5.5 and 5.6 (both MIT), a training framework for building models from your own audio and animation data (Apache-2.0), and pre-trained models on Hugging Face under the NVIDIA Open Model License. The Unreal sample project includes a MetaHuman setup. Teams can also run it as a NIM microservice, which is covered by NVIDIA's own software license rather than an open-source one.
NVIDIA doesn't publish a price. The main catch is hardware: the SDK README lists an NVIDIA GPU with CUDA and TensorRT as requirements. NVIDIA lists no official Unity or Godot plugin. AIGamer hasn't tested it. For how facial animation fits with dialogue and voice, see our AI NPCs guide.
Limitation: The SDK needs an NVIDIA GPU with CUDA 12 (12.8 or newer) and TensorRT 10.13 or newer, on Windows or Linux only; Maya ACE runs on Windows only
FAQ
Is NVIDIA Audio2Face open source?
Yes. NVIDIA open-sourced it on September 24, 2025, according to the NVIDIA Technical Blog. The GitHub hub README lists the SDK and the Maya and Unreal plugins under the MIT license, the training framework under Apache-2.0, and the models under the NVIDIA Open Model License.
Can I use Audio2Face-3D in Unreal Engine 5?
Yes. NVIDIA's ACE Unreal plugin v2.5 (downloads labeled 2.5.0rc3) supports UE 5.5 and 5.6, and an older v2.4 build covers UE 5.4, according to the GitHub hub README. It needs a separate Audio2Face-3D models plugin, and a sample project shows it on a MetaHuman.
Does Audio2Face-3D need an NVIDIA GPU?
The SDK README lists an NVIDIA GPU with CUDA 12.8 or newer (below 13) and TensorRT 10.13 or newer, on Windows 10/11 or Linux. The hub README mentions a CPU fallback, so check the SDK docs for your setup. Maya ACE can skip the local GPU by connecting to a hosted Audio2Face-3D service instead.
Can I use Audio2Face-3D in a commercial game?
The v3.0 model card on Hugging Face says the model is ready for commercial use under the NVIDIA Open Model License. The hub README says the Audio2Emotion models have a custom license that only allows use with Audio2Face, so read each license before shipping.
Sources
Compiled by AIGamer from NVIDIA Audio2Face-3D's own pages and the sources below, checked 7 November 2026. We haven't tested NVIDIA Audio2Face-3D hands-on.
- NVIDIA on GitHub: Audio2Face-3D hub README
- NVIDIA on GitHub: Audio2Face-3D SDK README
- NVIDIA on GitHub: Audio2Face-3D training framework
- NVIDIA on GitHub: Maya ACE
- NVIDIA on Hugging Face: Audio2Face-3D-v3.0 model card
- NVIDIA Technical Blog: NVIDIA Open Sources Audio2Face Animation Model (Ike Nnoli, September 24, 2025)
- NVIDIA: Audio2Face-3D NIM on build.nvidia.com
- NVIDIA: ACE for Games
- NVIDIA on GitHub: Maya ACE requirements
Image: NVIDIA
Written in our own words; quotes are marked and credited. How we report
