# Koe — Performance in every word. | Hanzo Creative AI

> Directable speech synthesis, character performances, dubbing and multilingual dialogue. Request research access and studio workflows.

Hanzo / Koe · VoiceProprietary Foundation · Private Weights

# Performance in every word.

Directable speech synthesis, character performances, dubbing and multilingual dialogue.

[Request Access](https://hanzo.ai/research-access?models=koe)[Studio Workflows](#studio)[Enterprise Infrastructure](#infrastructure)

Model ID: `koe-1`

Cinematic Dialogue

Documentary Voice

Interactive Agent

HANZO STUDIO · KOE

Mel-Spectrogram Locked · 30 Languages

Cinematic Dialogue · Verified Plate OutputDirectable Character Performance“Nuanced emotional dialogue delivery with realistic breath micro-dynamics, vocal fry modulation, and resonant timbre.”

Natural Prosodic Cadence Adherence

studio.hanzo.ai / koe · koe-1

CONFIDENTIAL INFERENCE · VOICE

FLOW-MATCHING ACOUSTIC STAGE · ZERO-SHOT 3.0s CLONINGTTFT LATENCY: 180ms

ACTING PRESET:Cinematic Dramatic

Intimate Documentary

Gritty Character

Warm Commercial

Zero-Shot CloningReference Audio3.0s Clean Sample

Sub-Second LatencyFirst Token< 200ms

Synthesis EngineHanzo Koe-1 EngineNeural Flow-Matching

Interactive canvas rendered in Hanzo Studio graph mode. High-throughput generation available on Hanzo Cloud.[Request Private Quota](https://hanzo.ai/research-access?models=koe)

Enterprise Infrastructure

## Production Platform & Dedicated Compute for Koe

Deploy Koe (koe-1) with sovereign GPU enclaves, zero customer data retention, real-time Hanzo Studio nodes, and high-throughput OpenAI-compatible endpoints.

Dedicated Compute · Hanzo CloudSovereign Enclave

## koe-1 Dedicated Cluster

Hardware: NVIDIA H100 / B200 Confidential ComputingIsolated enterprise inference clusters with sub-second time-to-first-chunk, guaranteed IOPS, and strict zero-data retention agreements for studio intellectual property.Zero customer data retention or public weight leakage

Dedicated enterprise throughput with committed SLAs

OpenAI-compatible REST and bidirectional streaming WebSocket API

[Platform Console](https://platform.hanzo.ai)[Request Production Quota](https://hanzo.ai/research-access?models=koe)

Studio Workflow Integrationstudio.hanzo.ai

## Hanzo Studio Native Node

Deployment: Visual Node Graph & TimelineFirst-class generative node in Hanzo Studio. Connect prompt generators, audio sync markers, keyframe animatics, and 3D camera paths directly in the node graph.Visual node editor with live parameter modulation

Lossless pipeline export: OpenUSD, GLTF, ProRes, 24-bit FLAC

Custom LoRA adapter slots and fine-tuning checkpoints

[Explore Hanzo Studio](https://hanzo.ai/studio)

#### Enterprise Production & Governance Specification

Operational Parameter

Enterprise Standard

Studio Commitment

Model Class & Version

`koe-1`

Proprietary Commercial Architecture

Hardware Isolation

Confidential GPU Enclave

Zero multi-tenant memory sharing

Data Retention

Zero Data Retention (ZDR)

Zero training on customer assets

Commercial Rights

Worldwide Perpetual License

100% customer ownership of generated media

Inference Delivery

REST / WebSocket / gRPC

Sub-second streaming latency

Studio Tooling

Hanzo Studio Graph Mode

Direct native node integration

Capabilities

## Production Creative Capabilities in Koe

Engineered for high-throughput studios, VFX supervisors, and generative artists requiring deterministic creative control.

01

### Director-Grade Emotional Modulation

Direct nuances in tone: breath control, hesitation, whispers, intense delivery, irony, and pitch contour through natural language instructions.✦ Full dramatic range expressiveness

02

### Cross-Lingual Voice Cloning

Preserves vocal identity, formant fingerprint, and emotional timbre across 40+ international languages and regional dialects.✦ 40+ supported languages

03

### Viseme-Synchronized Dubbing

Phoneme timing alignment matched to video mouth shapes and facial cadence for seamless localization.✦ Sub-frame lip sync alignment

04

### Actor Likeness Rights & Ledger Governance

Every synthesized voice carries cryptographic imperceptible watermarks and logs provenance to sovereign rights ledgers.✦ Guaranteed performer rights protection

Plate Status & Verification

## Research Samples Forthcoming

In accordance with Hanzo’s transparent evaluation standards, output galleries are published only after rigorous human blind testing and studio-grade verification. We deliberately decline using generic stock assets or simulated media as claimed model outputs.

[Join Partner Evaluation Cohort](https://hanzo.ai/research-access?models=koe)[Read Research Methodology](https://blog.hanzo.ai)

Technical Notes

## Sovereign Architecture & Confidential Compute

How Hanzo secures private creative intelligence across training, quantization, and real-time generation.

#### Model Specifications

Voice ArchitectureFlow-Matching Acoustic Latent Model

Audio Quality48 kHz / 24-bit Hi-Res Vocoding

Language Support40+ global languages & regional accents

LatencyStreaming API < 120 ms time-to-first-audio

Safety & RightsCryptographic watermarking & consent registry

#### Security & Data Integrity

Flow-matching acoustic model paired with neural vocoder delivering broadcast prosody and breath realism.

Proprietary weights are strictly private. Zero public weight distributions.

Mandatory identity verification and voice authorization prior to custom voice model enrollment.

Designed for broadcast film dubbing, AAA gaming dialogue systems, and high-consequence narration.

Industry Use Cases

## Where Studios Deploy Koe

Designed to integrate seamlessly into existing production render pipelines and creative agency workflows.

Global Media Localization

### Actor-Preserving Film Dubbing

Dub episodic television and movies into international languages while retaining the original actor’s voice characteristics.

Interactive NPC Dialogue

### Real-Time Branching Character Speech

Deliver thousands of hours of high-performance character dialogue in video games without recording bottlenecks.

Audiobook & Podcasting

### Multi-Voice Dramatic Performances

Cast an entire ensemble of distinct characters from a single manuscript with consistent chapter-to-chapter character voices.

Exclusive Research Intake

# Evaluate Koe in Your Studio

Apply for enterprise research access to test private checkpoints, integrate studio node workflows, and benchmark against production pipelines.

[Request Research Access](https://hanzo.ai/research-access?models=koe)[View All Models](https://hanzo.ai/models)