Practical overview
When it fits
Use it when you want Qwen3-TTS generation or fine-tuning as nodes inside a ComfyUI workflow rather than as a separate speech interface.
Who it is for
- ComfyUI users assembling text-to-speech workflows with Qwen3-TTS.
- Creators who want to use preset voices, describe a new voice, or clone from reference audio.
- Users preparing datasets to fine-tune a voice model.
Documented uses
- Generate speech from text using a preset speaker and chosen language.
- Create a voice from a natural-language description and use it to speak text.
- Generate speech in a cloned voice using reference audio and optional transcription.
- Prepare reusable voice prompts for repeated cloning workflows.
- Fine-tune a voice model from paired WAV recordings and TXT transcripts, then compare model audio.
Requirements & limits
- The model type determines the task: CustomVoice for preset speakers, VoiceDesign for text-described voices, and Base for cloning or fine-tuning.
- The extension supports 0.6B and 1.7B models and downloads only the selected model when it is first used.
- The requirements installation is manual, and the pinned transformers version may conflict with other custom nodes.
- For cloning, the README recommends short reference clips; longer audio can cause generation hangs.
- The README identifies slower inference on Windows without FlashAttention 2 and offers SDPA or eager attention options.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install the custom nodes in ComfyUI
Add the Qwen3-TTS node suite and install its Python requirements.
Before you start
- An existing ComfyUI installation.
- Python or the embedded Python supplied with a portable ComfyUI installation.
Steps
Clone the repository into ComfyUI/custom_nodes.
Run pip install -r requirements.txt from the extension directory; for portable ComfyUI, use its embedded Python to run the requirements file.
Expected outcome
The extension files and their listed Python dependencies are installed in the ComfyUI custom-node location.
Guide limitations
- ComfyUI does not install the requirements automatically.
- The upstream qwen-tts package requires transformers==4.57.3 and may downgrade an existing version.
- CUDA-compatible PyTorch is needed for GPU acceleration.
Build a preset-voice speech workflow
Use a CustomVoice model and generator node to produce speech from text.
Before you start
- ComfyUI with the Qwen3-TTS custom nodes installed.
- A Qwen3-TTS CustomVoice model, downloaded when selected if it is not already local.
Steps
Add the Qwen3-TTS Loader and select a CustomVoice model.
Connect the loaded model to the Custom Voice generator.
Choose a preset speaker, enter text and select a target language or Auto; optionally add an emotional instruction.
Expected outcome
A ComfyUI workflow is configured to generate speech with a selected preset voice.
Guide limitations
- The README describes the workflow and model download behavior but does not specify a guaranteed generation time or hardware requirement beyond CUDA-compatible PyTorch for GPU acceleration.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 18 October 2026
- Original website title
- DarioFT/ComfyUI-Qwen3-TTS · Official homepage title · retained in its source language