AuK: Open 1.5B Speech Generation and Editing Model by Tencent
ComfyUI Wiki
AuK is a 1.5B open speech model covering zero-shot TTS, voice cloning, content and acoustic editing, paralinguistic editing, enhancement, and separation.
A
AuK
TTSVoice CloningSpeech EditingEnhancementSeparationBase 1.5B speech foundation model by Tencent Hunyuan. Handles TTS, voice cloning, lyric rewriting, pitch and speed editing, emotion and timbre transfer, denoising, and speaker or vocal separation, all through one instruction interface. Runs in ComfyUI through the official ComfyUI-AuK nodes.
| Developer | Tencent Hunyuan + Shanghai Jiao Tong University |
| Release Date | 2026-09-09 |
| Architecture | 1.5B diffusion transformer + Qwen2.5-Omni-3B encoder |
| License | MIT |
| Parameters | 1.5B |
| Output | Up to 30-second audio per run |
Capabilities
- Zero-shot TTS: speak target text in a reference voice, or describe a voice with no reference audio (instruct TTS)
- Content editing: replace, insert, or remove spoken words; rewrite lyrics in singing while preserving melody and voice
- Acoustic editing: pitch by semitones, speaking rate, volume
- Paralinguistic editing: emotion, timbre, de-accent, nonverbal sounds, whisper conversion
- Enhancement and separation: denoise, dereverberate, keep one speaker, extract vocals from music, target speaker extraction
ComfyUI Usage
Run the official ComfyUI-AuK nodes with AuK Model Loader and AuK Generate / Edit:
- Install the
comfyuiextra in ComfyUI's Python environment:pip install -e ".[comfyui]" - Link
comfyui/ComfyUI-AuKintoComfyUI/custom_nodes - Download
auk_base.safetensorsand theQwen2.5-Omni-3Bencoder directory - Open the bundled
auk.jsonworkflow and set checkpoint and encoder paths in AuK Model Loader
The source/reference plus generated target must fit within 30 seconds per run. See the official ComfyUI guide for Prompt Enhancer configuration and Flash settings.
Related
Guides and workflows related to this model series.
No articles found.
Comments
Sign in with GitHub to join the discussion.