Boyo Chatterbox Turbo Loader

Chatterbox Turbo is Resemble AI's open-source TTS model, released December 2025, and it does something that still feels like a demo: zero-shot voice cloning from about five seconds of audio, with a <150ms time-to-first-sound and programmable emotion. This loader is how you get that model running inside ComfyUI. It's a thin wrapper - one input, one output - but it's doing two things under the hood that are worth knowing about before you build a workflow on top of it.

What it actually does

The loader takes a device dropdown (auto, cpu, cuda, mps; default auto picks CUDA if present, then MPS, then CPU) and hands you a CHATTERBOX_MODEL object for the Generate node. On first run it downloads the model from Hugging Face Hub - that's a real download, minutes on most connections, so the first "fast" TTS is preceded by a not-fast model fetch. It caches afterward.

Here's the part the README doesn't shout about: the source creates a dummy perth module and replaces the model's watermarker with a no-op before generation. Chatterbox Turbo ships with PerTh watermarking - an authenticity marker that stamps generated audio so it can be detected as synthetic. This loader deliberately bypasses it, and it does so unconditionally, not as an opt-in. If you use this node to generate speech, the output will not carry the provenance watermark the model's authors intended. The reason it's coded that way is practical - the watermarking dependency (resemble-perth) is fiddly and breaks installs - but the consequence is real: indistinguishable-from-human voice audio, stripped of its marker, produced by a one-line pip install. Think about what you're doing with that before you do it. It's a legitimate tool with a legitimate "why," and it's also exactly the kind of capability people are uncomfortable with when it comes with no friction.

Setup

The README's audio install path:

cd ComfyUI/custom_nodes && git clone https://github.com/DragonDiffusionbyBoyo/Boyonodes
pip install librosa transformers safetensors huggingface_hub pyloudnorm soundfile
pip install chatterbox-tts --no-deps

The --no-deps is important: it keeps Chatterbox's heavy dependency tree from fighting your existing ComfyUI environment (the pack's requirements file lists a whole stack - diffusers, onnxruntime, funasr, descript-audio-codec, and more - that you don't strictly need for the core path). After restart, the node appears under Boyo/Audio/TTS. Note that despite the README listing resemble-perth as an install step, the code bypasses it entirely - you can skip it, which is what the --no-deps install is betting on.

Troubleshooting

• "Chatterbox TTS not installed. Please install with: pip install chatterbox-tts" - that's the node's own error when the import fails. Run the install above and restart ComfyUI, not just the Python session.
• First-run hang - it's the Hugging Face download. Check the console for progress; give it a few minutes before assuming failure.
• CUDA out of memory on load - the model is a few GB. cpu will work (slowly) if you're VRAM-starved, but TTS is generation; you'll want the GPU for the actual speech synthesis in Generate.

This loader is the half of the pair you set up once and forget - the real action is in Boyo Chatterbox Turbo Generate, where you'll find the emotion tags, the voice cloning, and the actual speech.