Project MonetRequest demo
Home/Blog/Tencent AuK: Open-Source Speech Model, AuK-Flash & Local Setup

AI · Project Monet Briefing

Tencent AuK: Open-Source Speech Generation and Editing Model

Tencent AuK is a 1.5B open-source speech generation and editing family with a Base checkpoint and a four-step AuK-Flash variant.

Published 2026-09-09 · Updated 2026-09-09 · By Project Monet Editorial Team

Tencent AuK and AuK-Flash speech generation and editing workflow with waveform operations

01

What Tencent released

Tencent-Hunyuan open-sourced AuK on September 9, 2026, with code and public model weights for the 1.5B Base model and AuK-Flash. Tencent describes Flash as a distilled checkpoint for fixed four-step inference.

The repository includes CLI and Python inference, Gradio, official ComfyUI integration, prompt-enhancer tooling and a fine-tuning pipeline. A local Python interface is not the same thing as a hosted commercial API; no universal managed API price is documented in the release.

02

One interface for many speech tasks

AuK exposes zero-shot and instruction TTS, speech-content and lyric editing, pitch, speed and volume changes, emotion and timbre edits, de-accenting, nonverbal edits, whisper conversion, enhancement, speech and music separation, and target-speaker extraction through natural-language instructions.

That breadth is a documented capability list, not an independent quality ranking. Listening quality, identity preservation, language coverage and artifact rates still need task-specific evaluation.

03

Local setup and license

Tencent documents Python 3.10 and separate downloads for AuK or AuK-Flash plus Qwen2.5-Omni-3B as the MLLM encoder. The diffusion checkpoint, encoder and VAE are separate runtime pieces, and Tencent does not publish one universal minimum-VRAM requirement.

The repository license explicitly applies MIT terms to the publicly released training code, inference code, parameters and weights. Teams should still review required dependency licenses and exercise consent and rights controls for voice cloning or transformation.

04

Choose the right practical guide

For installation and weights, use How to Run Tencent AuK Locally. For task design, use the speech editing workflow guide.

Creators who prefer nodes can use the official ComfyUI integration guide. Tencent currently documents a 30-second source-plus-target sequence limit for that ComfyUI integration, so do not generalize it into a permanent model-wide duration limit.

Sources

Primary and supporting sources

Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.

Project Monet

Useful signals. Clear decisions. Better digital work.

Project Monet turns relevant shifts in AI, creator tools and the web into practical context—and builds focused websites for businesses ready to grow.

Request a free homepage concept