flybirdxx/ComfyUI-Qwen-TTS

What it solves

This project provides ComfyUI custom nodes that enable high-quality text-to-speech (TTS) synthesis, zero-shot voice cloning, and voice design within a node-based workflow. It allows users to generate speech from text, clone voices from short audio samples, or create entirely new voices based on natural language descriptions.

How it works

It integrates the open-source Qwen3-TTS model by Alibaba, supporting both 0.6B and 1.7B parameter models. The system uses a variety of attention mechanisms (such as SAGE, Flash Attention 2, and SDPA) to optimize inference speed and memory usage. It features a specialized set of nodes for different tasks:

• Voice Design: Creates voices from text descriptions.
• Voice Clone: Clones voices from reference audio.
• Custom Voice: Uses preset speakers.
• Dialogue Inference: Manages multi-role scripts and synthesizes complex conversations.
• Role Bank & Voice Clone Prompt: Extracts and saves voice features for reuse across different generations.

Who it’s for

This tool is designed for ComfyUI users who want to integrate professional-grade speech synthesis into their generative AI workflows, such as for creating audiobooks, roleplay scenarios, or character-driven content.

Highlights

• Zero-Shot Cloning: Clone a voice using only a short (5-15s) reference clip.
• Voice Design: Generate unique voice characteristics using natural language prompts.
• Multilingual Support: Native support for 10 languages including English, Chinese, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian.
• Flexible Memory Management: Includes an option to unload models from GPU memory after generation to support users with limited VRAM (< 8GB).
• Optimized Inference: Supports multiple attention implementations for significant speedups.
• Multi-Role Dialogue: Dedicated nodes for managing multiple speakers and synthesizing continuous dialogue scripts.