Generative AI

Build Lifelike Digital Humans with NVIDIA ACE, Now Generally Available

NVIDIA ACE—a suite of technologies bringing digital humans to life with generative AI—is now generally available for developers. Packaged as NVIDIA NIMs, these inference microservices enable developers to deliver high-quality natural language understanding, speech synthesis, and facial animation for gaming, customer service, healthcare, and more. 

NVIDIA is also introducing ACE PC NIM microservices for deployment across the installed base of 100M RTX AI PCs and laptops through early access.

A workflow diagram showing NVIDIA ACE NIMs for PC and Cloud inference. 
Figure 1. NVIDIA ACE NIMs are optimized for PC and Cloud inference

ACE now available for production deployment

Leading game and platform developers are revolutionizing real-time character interactions across diverse industries. Companies such as Aww Inc, Dell Technologies, Gumption, Hippocratic AI, Inventec, OurPalm, Perfect World Games, Reallusion, ServiceNow, SoulBotix, SoulShell, and Uneeq are embracing and integrating ACE into their platforms and applications. 

Video 1. NVIDIA ACE enables digital humans to see, understand, and interact in human-like ways
Video 2. Uneeq leverages NVIDIA Audio2Face NIM for realistic facial animation

To make it easier for developers to build digital humans, NVIDIA ACE 24.06 introduces general availability for many components within our suite of digital human technologies, including NVIDIA Riva, NVIDIA Audio2Face, and NVIDIA Omniverse RTX. They’re available through NVIDIA AI Enterprise

Microservices are on the NVIDIA NGC Catalog and the NVIDIA ACE GitHub repository, including:

  • Riva ASR 2.15.1 adds a new English model with higher quality and accuracy.
  • Riva TTS 2.15.1 improves the representation of German, European, and Latin American Spanish, Mandarin, and Italian. Also included is the beta release of P-Flow, a fast and efficient flow-based model that can adapt to a new voice with very little data.
  • Riva NMT 2.15.1 adds a new 1.5B any-to-any translation model.
  • Audio2Face 1.011 adds more blendshape customization options at runtime, supports more audio sampling rates, and has improved lip sync and facial performance quality with Metahuman characters.
  • Omniverse Renderer Microservice 1.0.0 adds new animation data protocol and gRPC and HTTP endpoints.
  • Animation Graph Microservice 1.0.0 adds support for avatar position and facial expression animations.
  • ACE Agent 4.0.0 adds speech support for custom RAGs, Colang 2.0 support, and prebuilt support for example RAG workflows.

Early-access microservices include the following:

  • Nemotron-3 4.5B SLM 0.1.0 is designed for on-device inference and includes INT4 quantization for minimal VRAM usage.
  • Speech Live Portrait 0.1.0 animates a person’s portrait photo using audio and supports lip sync, blinking and head pose animation.
  • VoiceFont 1.1.1 reduces latency for real-time use cases and supports 4 concurrent batches across all GPUs.

Developer tools and workflows

To help developers easily integrate and deploy ACE technologies, new workflows and developer tools are available on NVIDIA ACE GitHub.

The Kairos Gaming reference workflow features an Audio2Face plugin for Unreal Engine 5 and a configuration sample.

Video 3. Perfect World Games uses NVIDIA Audio2Face for accurate English and Mandarin lipsync

The NVIDIA Tokkio customer service reference workflow features a digital human configurator tool, ACE agent quickstart script, helm chart, a default animation pipeline screen, template scene, and samples.

Video 4. The Hippocratic AI healthcare agent connects better with patients with ACE NIM microservices

Developer tools also include:

  • Unified Cloud Services Tools 2.5 streamlines NVIDIA Cloud Functions application deployment.
  • Avatar Configurator 1.0.0 adds a new base avatar, hairstyle, and clothing options.

ACE NIM microservices come to RTX AI PCs 

In addition to NIMs for data centers, NVIDIA is building ACE NIM microservices for deployment across the installed base of 100M RTX AI PCs and laptops. Our first small language model (SLM), NVIDIA Nemotron-3 4.5B is built to run on device with comparable accuracy to large language models (LLMs) running in the cloud.

Nemotron-3 4.5B SLM is now in early access. Audio2Face and NVIDIA Riva ASR on-device models are coming soon.

An updated version of the Covert Protocol tech demo, developed in collaboration with Inworld AI, uses Audio2Face and Riva ASR running locally on GeForce RTX PCs

Video 5. Covert Protocol, an innovative technology demo that brings AI characters to life

Getting started

Begin your journey with NVIDIA ACE by instantly evaluating our latest ACE NIMs directly from your browser or through API endpoints running on a fully accelerated stack.

Leverage our latest tools and workflows to accelerate integrations. Apply for early access to microservices to see how ACE can transform your pipeline in the future.

For enterprises looking for an end-to-end digital human solution or custom development on ACE, reach out to ACE service delivery partners, part of the NVIDIA Partner Network including Convai, Inworld AI, Data Monsters, Quantiphi, Soulshell, Top Health Tech, and UneeQ.

For questions or feedback about digital human technologies, visit our Digital Human forum.

Discuss (0)