Qwen3 TTS Voice Clone Text to Speech ComfyUI workflow for Videcool

The Qwen3 TTS Voice Clone Text to Speech workflow in Videcool provides a powerful and flexible way to clone voices and generate speech from text prompts. Designed for speed, clarity, and creative control, this workflow is served by ComfyUI and uses Qwen3 ASR together with the Qwen3 TTS voice cloning model to synthesize speech that follows the characteristics of a reference speaker.

What can this ComfyUI workflow do?

In short: Voice cloning and text-to-speech synthesis.

This workflow takes a reference audio sample and uses it to clone a voice, then synthesizes new speech in that cloned voice based on text input. It first loads the reference audio, transcribes it with Qwen3 ASR, then passes both the reference audio and the transcribed reference text into the Qwen3 voice cloning model. The result is synthesized speech audio that preserves important speaker characteristics while following the new text prompt.

Example usage in Videcool

Figure 1 - Qwen3 TTS Voice Clone ComfyUI workflow in Videcool

Download the ComfyUI workflow

Download ComfyUI Workflow file: qwen3_voice_clone_text_to_speech-api.json

Image of the ComfyUI workflow

This figure provides a visual overview of the workflow layout inside ComfyUI. Each node is placed in logical order to establish a clean and efficient voice cloning and synthesis pipeline. The structure makes it easy to understand how the audio loader, Qwen3 ASR model, transcription node, Qwen3 TTS loader, voice cloning node, and MP3 output node interact. Users can modify or expand parts of the workflow to create custom variations or integrate voice cloning into larger audio or video production pipelines.

Figure 2 - Qwen3 TTS Voice Clone workflow

Installation steps

Step 1: Install the ComfyUI custom nodes or extension package that provides the Qwen3ASRLoader, Qwen3ASRTranscribe, Qwen3Loader, and Qwen3VoiceClone nodes required by this workflow.
Step 2: Ensure ComfyUI can access the ASR model repository Qwen/Qwen3-ASR-1.7B, because the workflow loads this model from Hugging Face with precision="bf16", attention="auto", and forced_aligner="None".
Step 3: Ensure ComfyUI can access the TTS model repository Qwen/Qwen3-TTS-12Hz-1.7B-Base, because the workflow loads this model from Hugging Face with precision="bf16" and attention="auto".
Step 4: If preferred, configure local model paths supported by the custom nodes, because both the ASR and TTS loader nodes include a local_model_path field that is empty in the attached workflow.
Step 5: Download the qwen3_voice_clone_text_to_speech-api.json workflow file into your home directory.
Step 6: Restart ComfyUI so the custom node and model files are recognized.
Step 7: Open the ComfyUI graphical user interface (ComfyUI GUI).
Step 8: Load the qwen3_voice_clone_text_to_speech-api.json workflow in the ComfyUI GUI.
Step 9: In the Load Audio node, select a reference audio file containing the voice you want to clone. The attached workflow uses sample_audio.mp3 as the example reference input.
Step 10: Enter the text you want the cloned voice to speak in the Qwen3-TTS Voice Clone node. The workflow automatically uses the output of the Qwen3-ASR Transcribe node as ref_text for the voice clone stage.
Step 11: Optionally adjust generation settings such as seed, language, max_new_tokens, and ref_audio_max_seconds. The attached workflow uses seed 522960123618311, language="Auto", max_new_tokens=2048, and ref_audio_max_seconds=30.
Step 12: Run the workflow to synthesize cloned speech and save the output as an MP3 file through the Save Audio (MP3) node. The default filename prefix is audio/ComfyUI and the MP3 quality is set to V0.
Step 13: Open Videcool in your browser, select the Voice Clone tool, and use the generated audio in your video projects.

Installation video

The workflow requires a reference audio sample and text input, plus a few basic parameter adjustments to begin cloning voices. After loading the JSON file, users can select the reference audio, enter the text to be spoken, and adjust generation settings such as seed, language, token limit, and maximum reference audio length. Once executed, the ASR stage transcribes the reference audio and the TTS voice clone stage synthesizes new speech, which is then saved as an MP3 file that can be used across other Videcool tools.

Prerequisites

To run the workflow correctly, install the ComfyUI custom nodes that provide the Qwen3 ASR and Qwen3 TTS voice clone nodes, then make sure both model repositories or local model files are accessible to ComfyUI. These components ensure the workflow can load the reference audio, transcribe it, process the target text, synthesize cloned speech, and save the result as an MP3 file. Proper installation into your ComfyUI custom node and model environment is essential before running the workflow.

ASR model repository used by the workflow
https://huggingface.co/Qwen/Qwen3-ASR-1.7B

TTS model repository used by the workflow
https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-Base

Custom node classes required by the workflow
LoadAudio
Qwen3ASRLoader
Qwen3ASRTranscribe
Qwen3Loader
Qwen3VoiceClone
SaveAudioMP3

How to use this workflow in Videcool

Videcool integrates seamlessly with ComfyUI, allowing users to clone voices and generate speech directly without managing the underlying node graph. After importing the workflow file, simply select a reference audio file, enter your target text, and click generate. The system handles the backend interactions with ComfyUI, including reference transcription and voice-conditioned speech generation, making voice cloning intuitive and accessible even for users who do not want to work directly with ComfyUI nodes.

ComfyUI nodes used

This workflow uses the following nodes. Each node performs a specific role, such as loading reference audio, transcribing the sample, loading the TTS model, generating cloned speech, and finally saving the output. Together they create a reliable and modular pipeline that can be easily extended or customized.

Load Audio
Qwen3-ASR Loader
Qwen3-ASR Transcribe
Qwen3-TTS Loader
Qwen3-TTS Voice Clone
Save Audio (MP3)

Base AI model

This workflow is built on two Qwen3 model components: the ASR model Qwen/Qwen3-ASR-1.7B for reference transcription and the TTS model Qwen/Qwen3-TTS-12Hz-1.7B-Base for voice-conditioned speech generation. The ASR stage extracts reference text from the sample audio, and the voice clone stage uses both the reference audio and transcribed text to synthesize new speech that follows the speaker characteristics of the input sample.

Voice cloning quality and parameters

Voice cloning quality depends on the clarity of the reference audio sample and the generation settings. The attached workflow uses automatic language detection for transcription, no timestamps, a maximum reference audio length of 30 seconds, a large token limit of 2048, and a fixed random seed for repeatability. For best results, use a clear reference recording with minimal background noise and enough speech content for the ASR stage to transcribe accurately.

Conclusion

The Qwen3 TTS Voice Clone Text to Speech ComfyUI workflow is a robust, powerful, and user-friendly solution for cloning voices and generating speech in Videcool. With its combination of ASR-based reference transcription, Qwen3 voice-conditioned TTS generation, a modular ComfyUI pipeline, and seamless platform integration, it enables beginners and professionals alike to produce useful cloned speech audio for narration, dubbing, and creative voice workflows.