From voice to performance

Lingotion turns actor performances into actor-clones allowing characters to respond in real-time, and make it fully controllable inside your game. Create acting from reactive storytelling, adaptive dialogue, or procedural content created in your game. Capture each actor’s unique expression of fear, joy, contempt, grief, anger, terror, love, and more – along with the dialect, rhythm, imperfections, and other quirks that make us human.

Derivative AI instead of Generative AI

Lingotion is unique by not generating any new art and instead captures the original artistic expression of an real actor. Lingotions AI acting engine thereby preserves the original copyright licensed from the actors by creating a copyright dervative output. Lingotion provides legal compliance documentation and proof of a copyright safe solution that is not generative AI.

What does Lingotion do?

• Provides acting in real time during gameplay.
• Runs fully on-device – no servers, no latency, no usage fees.
• Full emotional control – what they say and how they say it.
• Powered by real actors – ethical, legally compliant and commercially safe.

Acting for real – ethically

Lingotion brings acting and AI together through licensed digital performances by real actors receiving royalties. We capture the actor’s unique performance style and emotional depth to bring each character to life authentically.

At its core, Lingotion is about performance, preserving the true artistry of acting and making it usable in new, interactive ways while keeping actors in control.

Unleash the possibilities

• Create emotions for characters that speak and adapt based on game states.
• Faster design and prototypes without content bottlenecks.
• Scaling voice coverage without scaling your budget.
• Avoid expensive pickup recordings.

Technical Requirements

- Unity version 6.0

FAQ

Does Lingotion use any cloud services?

No. Lingotion runs entirely on the player’s device, fully integrated within the game engine. It does not use any cloud infrastructure. However, we provide a SaaS cloud service for customers that only wish to create static assets.

Can it run on both GPU and CPU?

Yes. You can choose whether the Lingotion Engine runs on the CPU or GPU.

Can it run in real time on a CPU using only a few cores?

Yes, the Lingotion AI model is a proprietary solution that is super-efficient both in terms of memory usage and compute performance.

Is Lingotion LLM-based?

No. LLMs are large autoregressive models that often require too much memory and GPU or CPU performance to run concurrently with everything else and consume too much energy for battery-powered devices. Lingotion uses a proprietary non-autoregressive AI architecture to achieve the necessary performance efficiency.

Is there any way to synchronize the audio with external events to create a trigger for a position in the text?

Yes. Since the timing depends on the character and emotion, Lingotion provides text-position-to- sample-position synchronization. The SDK and library provide the exact sample position when a word in the input text is spoken.

What does the input look like?

It is a JSON structure that specifies language, emotion, and text for a dialogue or part of a dialogue, similar to a manuscript.

Will it affect graphics rendering if it runs on the GPU?

Only minimally. Lingotion is designed to minimize its impact on real-time graphics performance.

What platforms are supported?

Lingotion currently supports and maintain Windows, iOS, Android, Quest,macOS, and Linux. Support for Xbox, PlayStation, and Nintendo Switch will be added soon.

How large are the AI models?

The AI models come in five sizes depending on the hardware capabilities: X-Large, Large, Medium, Small, and X-Small. The sizes range from 33 million parameters for X-Large down to 3.4 million parameters for X-Small. All models can run on CPU or GPU. X-Large and Large are typically intended for high-end PCs and consoles, Large, Medium and Small are typically for handheld devices and mid- to low-end PCs, and X-Small is for old budget phones with very poor performance.

How much memory does a character model require?

The standard model size per character is 13 MB–132 MB depending on model size. Multiple characters can be trained into a single model to reduce the total asset size. The roadmap includes further reduction of character model sizes.

Can you run multiple syntheses concurrently?

Yes, the SDKs are thread-safe, and it is possible to run multiple syntheses in parallel due to the super-efficient performance of the Lingotion model.

Will it synthesize the entire sentence before getting the audio?

No, to reduce latency, Lingotion starts streaming audio before the entire sentence has been synthesized.

How long is the latency from synth call to receiving audio?

It depends on the hardware capabilities, but it is usually in the range of 100 ms–500 ms. For low-performance devices, it can exceed 1000 ms.