Qwen3-TTS setup with Pinokio and ComfyUI

This tutorial covers Qwen3-TTS for local AI speech generation, starting with installation through Pinokio on Windows. The speaker downloads the project, installs it and explains which models serve voice design, reference-audio cloning and predefined speakers. Voice design uses the 1.7B model; the walkthrough also covers smaller and larger model choices for other tasks.

The demonstrations show how voice descriptions affect delivery, how to upload reference audio and transcribe it, and how chunk size and pauses change generated speech. The speaker recommends keeping cloning inputs around 400 characters despite an approximate 500-character input limit. Style instructions for predefined speakers work only with the larger models, according to the walkthrough.

GPU memory management gets a separate explanation: loading several models consumes more VRAM, while individual unload controls and an unload-all button release it. The speaker considers the cloned voice usable but imperfect, suggests a 6 GB VRAM system should work and leaves 4 GB compatibility uncertain.

The final section revisits ComfyUI after a custom-node update resolves an earlier error. It covers Git availability, dependency installation with a separate command for the portable version, restarting ComfyUI and loading a JSON workflow. Model files download on the first run. The speaker cautions viewers to check unfamiliar custom-node developers before installation.