Maya1 TTS (AIO) Barebones

This is the plain-widget twin of the Maya1 TTS (AIO) node. Same model, same engine, same audio output - the only difference is the interface. Where the main node renders a custom dark-themed canvas with character presets, emotion buttons, and a fullscreen text editor, this one gives you standard ComfyUI text boxes and dropdowns, stacked vertically like any other node.

So why would you reach for it? The usual reason is that the pretty version broke. The custom JavaScript canvas occasionally renders as a black box - no widgets, nothing clickable - especially on older ComfyUI builds or with aggressive browser extensions. The author built Barebones specifically as the escape hatch for that. If you see the black box, or you just prefer boring predictable widgets, this node is the same functionality without the custom UI risk. A sibling without the gamble, essentially.

How it works

Maya1 is a 3B-parameter open-weight speech model (Apache 2.0, from Maya Research) that generates SNAC audio codec tokens rather than raw audio. Your voice description gets wrapped in a <description="..."> control token, appended to your script text, and the model autoregressively writes codec tokens - 7 per audio frame - which the node decodes to a 24kHz mono waveform. It runs locally on your GPU; there's no API call anywhere in this pack. The one thing the barebones version drops is the visual tag insertion - the emotion tags still work, you just type them by hand.

The inputs that matter

Everything is exposed as flat widgets, and they're the same set as the main node:

• voice_description - natural language voice recipe: age, gender, accent, pitch, timbre, pacing.
• text - your script with manual emotion tags like <laugh> or <whisper>. The tooltip lists 17 of them: laugh, laugh_harder, giggle, chuckle, cry, sigh, gasp, whisper, angry, scream, snort, yawn, cough, sneeze, breathing, humming, throat_clearing.
• model_name - auto-discovered from ComfyUI/models/maya1-TTS/; if it shows "(No models folder found...)", the model isn't downloaded yet.
• dtype - bfloat16 is the default and the speed pick; 4/8-bit quantization saves VRAM (~6-7GB) but is genuinely slower, so only use it under ~10GB VRAM.
• max_new_tokens - ~50 tokens per word of speech; 4000 ≈ 30-40s. Increase it if output cuts off early.
• seed - 0 for random, a fixed number to reproduce a take.

Output

audio - a standard ComfyUI AUDIO output at 24kHz mono. Connect it to PreviewAudio, SaveAudio, or any other audio sink.

Install

You install this pack once and get both nodes - they ship together, so there's no separate install for Barebones.

cd ComfyUI/custom_nodes
git clone https://github.com/Saganaki22/ComfyUI-Maya1_TTS.git
cd ComfyUI-Maya1_TTS && pip install -r requirements.txt

Or use ComfyUI Manager and search "Maya1 TTS". The dependency that matters is snac (the audio codec); transformers and torch you almost certainly already have. Then download the model to the non-standard folder the node expects:

cd ComfyUI
hf download maya-research/maya1 --local-dir models/maya1-TTS/maya1

Restart ComfyUI, and look under Add Node → audio → Maya1 TTS (AIO) Barebones.

Troubleshooting

• "No valid Maya1 models found" - the model folder is empty or in the wrong place; it must be models/maya1-TTS/<something>/ with the shard files, chat template, and tokenizer/ inside.
• OOM - try 4bit, reduce max_new_tokens, or set keep_model_in_vram to false after generation.
• No audio generated - the classic cause is max_new_tokens too low or a temperature outside 0.3-0.5; also simplify a very long voice description.
• Quantization errors - pip install bitsandbytes; 4/8-bit also requires CUDA, so fall back to float16 on CPU.

And one honest note: if you came here from the black-box problem, the fix usually is a hard browser refresh plus a ComfyUI update first. The Barebones node is the reliable fallback, not necessarily the permanent home.