NVIDIA ACE Alternative for Unity Developers

If your game is in Unity, NVIDIA ACE is not a drop-in option: it ships Unreal Engine 5 plugins and C and C++ SDKs, and no Unity plugin is listed. ACE is NVIDIA's stack for AI characters that run on the player's PC, bundling small language models, speech recognition, text-to-speech and Audio2Face facial animation. It is behind features such as PUBG's co-player characters, inZOI's Smart Zois and an AI advisor in Total War: PHARAOH.

So who is it for, what would a Unity team have to build to use it, and what is the shorter route? That is what this NVIDIA ACE alternative guide covers.

Who NVIDIA ACE is built for

ACE fits a specific team profile:

• a PC game, usually in Unreal Engine 5 or a custom C++ engine
• engineers who are comfortable integrating native SDKs and managing model files
• a reason to run AI on the player's machine, such as zero server cost per conversation or offline play
• time to build the character layer on top: prompts, memory, guardrails and dialogue structure

If that describes your studio, ACE gives you the parts. If you are an indie or mid-size Unity team shipping to several platforms, most of that list becomes work you have to staff.

What ACE gives you

Models

• Language models: Nemotron small language models and Qwen 3.5 4B, sized for on-device use.
• Speech recognition: Nemotron ASR, including a low-latency streaming model, on-device or in the cloud.
• Text-to-speech: Chatterbox TTS and Qwen3 TTS, both with voice cloning.
• Faces: Audio2Face-3D for lip sync and facial animation, and Audio2Emotion-3D for emotion inferred from audio.

SDKs

• Game Agent SDK: a C and C++ agent framework for native integration, in beta.
• In-Game Inferencing SDK (NVIGI): runs models in-process alongside your renderer, with GPU, NPU and CPU backends.

Engines and licences

Unreal Engine 5 plugins cover speech recognition, language models and TTS for UE 5.5 to 5.7, and Audio2Face-3D for UE 5.4 to 5.6. Custom engines integrate through the SDKs. The UE5 plugins and the Audio2Face-3D SDK are MIT licensed, and each model has its own licence terms.

What you would build yourself on ACE

That is a toolkit, not a product. For a Unity team, the work list looks like this:

• A Unity bridge for native C++ SDKs, or your own inference integration
• Model selection, download, versioning and build-size management per platform
• A memory budget that shares the player's GPU between rendering and AI
• Character definitions: personality, knowledge, guardrails, written as prompts in code
• A way for writers to edit and test characters without an engineer
• Dialogue structure for story beats, and a way for the game to know when a beat is done
• Action parsing so a reply can open a door or hand over an item
• A fallback for players whose hardware cannot run the models
• A plan for platforms outside PC, such as WebGL

Each line is a real task. Together they are a team.

NVIDIA ACE alternative for Unity: which route fits?

Read the matrix from the top. Engine and platform rule out more options than anything else: if you are in Unity, or you need a browser build, the on-device ACE route means writing and maintaining the engine bridge yourself. Next comes who edits the characters. If writers and designers need to change a personality on a Tuesday afternoon and test it before lunch, the character has to live somewhere they can open, not in a C++ prompt file. Offline play is the one row where ACE's design is the point, and it is worth being clear early on whether your game truly needs it or only needs cheap prototyping.

What the Neocortex route looks like in Unity

You start in the browser, not the engine. A character is a graph of nodes: Agent, Personality, Emotion, Knowledge, World, Objects, Intents, Actions, Language, Accent and Voice. Writers fill those in, upload lore as PDF or TXT files, and talk to the character in the editor. That testing never uses credits.

Voice is part of the same graph. Pick a voice, clone one from a short sample you have the rights to, or design one from a text description, and it streams to the player with speech-to-text on the way in.

Then you install the Unity SDK, which supports Unity 6 and comes with six sample scenes, and point a scene at your character. Your game sends nearby objects and recent events. Replies come back as lines with an emotion tag you can drive animation from, plus actions such as OPEN_DOOR with a targetId of door_02. For authored beats, Conversation Flows return checkpoint keywords (flowState) your game listens for.

If you want models on your own machine while you build, you can run local LLMs in Unity through Ollama for text chat, with no NVIDIA hardware requirement. The same character also works in Unreal Engine 5, Godot (beta) and anything that can call our Web API.

Start with one character

Pick the NPC your players will spend the most time with, build it in the editor on the free plan, and run it in a Unity sample scene. When the game is ready to sell, Pro at EUR 29 a month covers commercial use and group chat, and Team at EUR 99 a month adds five seats. You get a playable, voiced character without building the pipeline first.

Competitor details were last checked in October 2026.

Frequently Asked Questions