Saganaki22 Unveils Higgs_v3-TTS-ComfyUI For Local Voice Cloning
Higgs_v3-TTS-ComfyUI is a new release that brings multilingual conversational text to speech into ComfyUI. It allows users to generate spoken audio in over 100 languages using a local AI model. The tool also supports zero-shot voice cloning and multi-speaker dialogue generation.
Developer Saganaki22 who also made the Zonos2_TTS-ComfyUI node created this integration to bring the Boson AI higgs-audio-v3-tts-4b model directly into local workflows. They built the custom nodes to handle audio input and output using standard ComfyUI audio connections. This effort makes it easier to run advanced speech generation without relying on remote servers or paid APIs.
Advanced audio generation tools
Key Features
- Generates multilingual conversational text to speech.
- Zero-shot voice cloning with reference audio.
- Inline controls for emotion and style.
- Multi-speaker dialogue with separate voices.
- Automatic longform text splitting and chunking.
- AIMDO DynamicVRAM support for memory management.
This tool is ideal for creators who need to generate dialogue or narration entirely on their own hardware. Teams producing localized content can benefit from the support for 100 different languages. Anyone needing specific emotional delivery or sound effects will find the inline text tags highly useful.
Development notes and limitations
The current version requires about 11 GB of VRAM to run smoothly using bf16 on CUDA. Users should note that strong emotion tags placed at the very beginning of a prompt can cause the cloned voice to drift. The developer handles model switching internally, meaning there is no manual unload node or keep-loaded toggle required.
"The loader handles model-switch cleanup internally." - Source: GitHub
Project Links
