Extensions/Comfyui-Auk-T8
ComfyUI Extension

Comfyui-Auk-T8

AuK Local speech generation and editing nodes for ComfyUI V3

By T8mars·Created a day ago·Updated a day ago· 1
T8mars/Comfyui-Auk-T8
Nodes2
On cloudLocal install
CategoryAuK Local
Stars1
Updateda day ago

Nodes (2)

AuK Local 连接

The AuK Local 连接 Node Is Where You Find Out What Broke

AuK Local
AuK Local 生成 / 编辑

16 Speech Jobs, Two Text Boxes, and a Hard 30-Second Wall

AuK Local
Readme
<div align="center">

AuK Local · T8star-Aix for ComfyUI

ComfyUI V3 bridge nodes for AuK speech generation and editing

中文说明 · Model weights · One-click package

</div>

This repository contains the ComfyUI side of the AuK Local integration package. The nodes call the isolated AuK Local service at http://127.0.0.1:7860, so AuK, Qwen, and their Python dependencies stay outside the ComfyUI process.

Features

  • Two ComfyUI V3 nodes: AuK Local Connection and AuK Local Generate / Edit.
  • 16 tasks covering instruction TTS, zero-shot voice cloning, speech and lyric editing, pitch/speed/volume/emotion/timbre editing, de-accenting, nonverbal and whisper conversion, enhancement, and source separation.
  • Standard ComfyUI AUDIO output plus the final instruction and run metadata JSON.
  • AuK-Flash and AuK Base selection, deterministic seed, CPU offload, cancellation, retry recovery, and a 30-second input-plus-output guard.
  • Three ready-to-load workflows in example_workflows.

Install

ComfyUI Manager

Search for AuK Local · T8star-Aix in ComfyUI Manager and install it, then restart ComfyUI.

Git

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-Auk-T8

This bridge has no extra pip dependencies. ComfyUI supplies PyTorch and Torchaudio.

Run

  1. Download the single AuK Local + ComfyUI integration package.
  2. Start 启动AuK服务.cmd in the package and keep its window open.
  3. In ComfyUI, load one of the workflows in example_workflows.
  4. In AuK Local Connection, leave service_url as http://127.0.0.1:7860. If the node was installed by Manager, set the advanced token_file field to the package's absolute data/session-token path. The package installer configures this path automatically.
  5. Choose the task and run the workflow. Use your own audio in Load Audio for cloning or editing examples.

The service listens on loopback only. The workflow stores the token file path, never the token itself.

Models

The exact model mirror used by this release is hosted at t8star/Auk-Comfy:

  • AuK-Flash — fast four-step generation.
  • AuK — base model for higher-quality generation and editing.
  • Qwen2.5-Omni-3B — prompt understanding used by the local service.

The Hugging Face model card links back to this ComfyUI repository and records the upstream repositories, revisions, file sizes, and SHA-256 hashes.

Example workflows

  • AuK-01-描述生成语音.json — instruction TTS without reference audio.
  • AuK-02-参考声音克隆.json — zero-shot voice cloning.
  • AuK-03-语音文字编辑.json — edit spoken content while keeping the source voice.

Compatibility

  • ComfyUI >=0.3.48 with the V3 custom-node API.
  • Windows 10/11 x64 for the published integration package.
  • Verified with Python 3.10, PyTorch/Torchaudio 2.7.1 + CUDA 12.8, and an NVIDIA RTX 5090 Laptop GPU with 24 GB VRAM.
  • Audio output: 24 kHz float WAV.

Links

License

The node code is released under the MIT License. Model files in the Hugging Face repository retain the license files supplied by their original authors.