ComfyUI-Supertonic3TTS Introduction
ComfyUI-Supertonic3TTS adds speech generation to ComfyUI. You can type a line of dialogue, choose a language and voice, and generate audio for a character, narration, or an experimental workflow. The extension uses Supertone’s Supertonic-3 text-to-speech system locally through ONNX Runtime. It can run on a CPU, so a GPU is not required.
The extension provides three nodes: one loads the speech model, one turns text into audio, and one optionally adjusts the finished sound. This lets you build a simple text-to-speech workflow or add speech to a larger ComfyUI project.
Example ComfyUI-Supertonic3TTS workflow
How ComfyUI-Supertonic3TTS Works
Think of the workflow as a small recording studio:
- Supertonic Model Loader prepares the speech engine. On its first run, it downloads approximately 400 MB of model files to
models/supertonic-3/inside the extension’s directory. Later runs use those local files. - Supertonic Text-to-Speech acts as the performer. You give it text and choose a language, voice, speed, and synthesis setting called steps. It produces ComfyUI
AUDIO. - Supertonic Effects is an optional finishing stage. It can adjust the generated audio before you send it to Preview Audio or Save Audio. It also accepts audio from other ComfyUI sources.
For a first workflow, connect Supertonic Model Loader to the model input of Supertonic Text-to-Speech, then connect its AUDIO output to Preview Audio or Save Audio. Type a short sentence, select its language and a voice, and run the workflow. To polish the result, place Supertonic Effects between Text-to-Speech and Preview/Save.
ComfyUI-Supertonic3TTS Features
| Node | What it does | When to use it |
|---|---|---|
| Supertonic Model Loader | Loads the Supertonic-3 engine and stores the model locally after the first download. | Add it at the start of a speech workflow. |
| Supertonic Text-to-Speech | Converts your text into spoken AUDIO. | Create dialogue, narration, or other spoken material. |
| Supertonic Effects | Offers trimming, normalization, pitch adjustment, time stretching, and chorus. | Refine speech or process another AUDIO source. |
Languages and voices. The Text-to-Speech node supports 31 named languages, including English, Korean, Japanese, Indonesian, Arabic, Spanish, and French. Choose the language that matches your text. Its preset voice choices are M1–M5 and F1–F5, described by the author as five male and five female voices. Try a few presets with the same sentence to find one that suits your character or scene.
Custom voice styles. If you have a compatible Supertonic voice-style .json file, enter its absolute path in custom_style_path. An absolute path identifies the file’s full location on your computer. When this field is set, the custom style takes priority over the voice_style preset dropdown. Supertone’s Voice Builder is a relevant resource for voice styles.
Expression tags. Put a tag directly in the text to request a vocal expression. The extension passes these tags to the Supertonic SDK during speech generation.
| Tag | Requested expression |
|---|---|
<laugh> | Laughter |
<breath> | Breath intake |
<surprise> | Surprise |
<sigh> | Sigh |
<scream> | Scream or shout |
<throatclear> | Throat clearing |
<sad> | Sad tone |
<angry> | Angry tone |
<cough> | Cough |
<yawn> | Yawn |
For example, Halo! <laugh> Senang bertemu denganmu! places a laughter cue between two spoken phrases.
Speed and finishing adjustments. The Text-to-Speech speed setting ranges from 0.5× to 2.0× and changes the pace while speech is being generated. For a smaller adjustment afterward, use the Effects node’s time_stretch control. That control processes already-generated audio and can introduce audible artifacts at extreme settings. You can use both: choose the main speaking pace in Text-to-Speech, then make a final tempo adjustment in Effects.
Steps. The Text-to-Speech node offers 5–12 diffusion steps, with a default of 8. More steps can produce smoother speech but take longer. Start at the default; compare a higher setting if the result needs refinement, or a lower setting if speed matters more.
Supported language codes. The node lists the following named languages, plus na as an unknown/fallback option:
| Code | Language | Code | Language | Code | Language |
|---|---|---|---|---|---|
en | English | ko | Korean | ja | Japanese |
ar | Arabic | bg | Bulgarian | cs | Czech |
da | Danish | de | German | el | Greek |
es | Spanish | et | Estonian | fi | Finnish |
fr | French | hi | Hindi | hr | Croatian |
hu | Hungarian | id | Indonesian | it | Italian |
lt | Lithuanian | lv | Latvian | nl | Dutch |
pl | Polish | pt | Portuguese | ro | Romanian |
ru | Russian | sk | Slovak | sl | Slovenian |
sv | Swedish | tr | Turkish | uk | Ukrainian |
vi | Vietnamese |
ComfyUI-Supertonic3TTS Models
The author documents one speech model: Supertonic-3. The M1–M5 and F1–F5 options are voice presets, not separate models. Likewise, a custom style .json changes the selected voice style; it does not replace the speech model.
The Model Loader downloads Supertonic-3 on first use and keeps it in models/supertonic-3/ within the extension’s directory. Because the files are stored locally, subsequent loads skip that download.
Troubleshooting ComfyUI-Supertonic3TTS
The model seems to take a long time to load the first time.
- Check the ComfyUI console for the model download’s 0–100% progress indicator.
- Allow time for the approximately 400 MB first-run download to finish.
- On later runs, check whether the model files are present in the extension’s
models/supertonic-3/directory. The loader should reuse them.
You generated audio, but cannot hear or find it.
- Check that the Text-to-Speech
AUDIOoutput is connected to Preview Audio or Save Audio. - If you added Supertonic Effects, check both connections: Text-to-Speech to Effects, then Effects to Preview/Save.
- Run the workflow again and inspect the output of the Preview/Save node.
Your custom voice style is not taking effect.
- Confirm that
custom_style_pathpoints to a Supertonic voice-style.jsonfile. - Use the file’s absolute path, not just its filename.
- Remember that a populated
custom_style_pathoverrides the presetvoice_styleselection. Clear the field when you want to compare preset voices.
The speaking pace or sound is not what you expected.
- Set the main pace with Text-to-Speech
speed; its range is 0.5×–2.0×. - If you use Effects
time_stretch, bring it closer to a neutral setting and compare the output. Extreme post-processing can cause artifacts. - Try the default 8 steps as a baseline, then compare a higher setting for smoothness or a lower one for faster generation.
Frequently asked questions
- Do I need a GPU? No. The author describes the extension as CPU-friendly and running through ONNX Runtime.
- Can I process audio that was not made with Supertonic-3? Yes. Supertonic Effects accepts any ComfyUI
AUDIOsource. - Are expression tags separate nodes? No. Type supported tags, such as
<laugh>or<sigh>, directly into your speech text. - Does choosing a voice download a different model? The documentation lists one model, Supertonic-3, with ten built-in voice presets.
- Where are the downloaded model files? In
models/supertonic-3/inside this extension’s directory, not in~/.cache/.
Learn More about ComfyUI-Supertonic3TTS
- ComfyUI: Learn about the node-based environment in which this extension runs.
- Supertone’s Supertonic project and Voice Builder: Explore the underlying speech system and the voice-style resource referenced by the author.
- The workflow image in the introduction shows how the extension’s nodes fit together. For hands-on learning, begin with Model Loader → Text-to-Speech → Preview Audio, then add Supertonic Effects once you have a voice you like.
