ComfyUI-OmniVoice-TTS

Why we included this project

ComfyUI users who want speech generation inside their node graph get a zero-shot TTS engine that covers more than 600 languages. You can clone a voice from a few seconds of reference audio, or design a synthetic voice from a text description of gender, age, pitch, and accent. Multi-speaker dialogue works through simple inline tags, and non-verbal cues like laughter or sighs can be embedded directly in the text. The nodes also handle the practical details: models download automatically on first use, CPU offloading keeps VRAM usage low, and Whisper is cached so repeated runs don't re-fetch it. That makes it a convenient way to get speech generation into a ComfyUI workflow, whether you're doing voiceover work or just testing multilingual TTS ideas.