ChatterBox TTS Loader (Auto)

ChatterBox TTS Loader (Auto) is the lazy person's way into this pack's voice cloning: one node, one device setting, and it downloads the official ChatterBox TTS model from HuggingFace on first use, caches it, and hands you a ready-to-use model. No hunting for .safetensors files, no manual folder layout, no deciding which checkpoint is which. If the Manual loader's four-file ceremony makes you groan, this is your node.

How it works

It wraps ChatterBox's official Python API. On the first run it calls ChatterboxTTS.from_pretrained(device=device), which pulls the model weights from HuggingFace and caches them locally, so subsequent runs load from disk. The device input accepts auto (picks CUDA if available, else MPS or CPU), cuda, or cpu. The loader then does one deliberate tweak: it disables the model's watermarker, which the pack's author found segfaults under ComfyUI's environment due to version mismatches. You get clean, watermarked-free audio, and the model object is handed off as a CHATTERBOX_TTS output.

Inputs and outputs

• device - auto (default), cuda, or cpu

Output: model - a CHATTERBOX_TTS object for the Simple TTS node. That pairing is the intended flow: Auto loader → ChatterBox TTS (Simple) → audio.

Install

Pack install via ComfyUI Manager (search "Latent Astronaut Suite") or:

cd ComfyUI/custom_nodes
git clone https://github.com/latentastronaut/comfyui-latent-astronaut-suite

Restart. On first boot install.py installs chatterbox-tts with --no-deps (its strict version pins would fight ComfyUI's environment) plus librosa - without those, the import inside the loader fails. The model download happens lazily on your first generation, not at install time.

Where people get burned

First-run patience is the big one. The download is multi-gigabyte and the first generation includes model load time, so a long pause with no output is normal - it's not hung, it's downloading. If it errors instead, the usual cause is the dependency step: restart ComfyUI once after install so install.py has actually run, and confirm chatterbox-tts imported cleanly. Don't pair this with the Manual TTS node - that one expects the manual loader's model object, and the API shapes don't match.