4.2 KiB
Character and Narrator Switching
Use [Alice] and [Bob] tags in TTS Text or TTS SRT to select character voices. This is shared TTS Audio Suite syntax; reference requirements, languages, and native multi-speaker modes depend on the selected engine.
Add character voices
Save your voices under ComfyUI/models/voices/:
ComfyUI/models/voices/
├── alice.wav
├── alice.reference.txt
├── bob.wav
└── bob.reference.txt
- The audio filename gives the character its name:
alice.wavbecomes[Alice]. Names are case-insensitive. Subfolders are for organization; use unique filenames across your library. - For engines that use a transcript, put the exact spoken words in
name.reference.txt. A matchingname.txtis also accepted when no reference transcript is found. Keep metadata in a separate file from the transcript. - Transcript requirements vary by engine and mode; see Reference Transcript in the Feature Comparison.
The suite also searches models/TTS/voices/, bundled voices_examples/, and voice paths configured in extra_model_paths.yaml. Supported audio formats include WAV, MP3, FLAC, OGG, M4A, and AAC.
You can create library voices with Save Character Voice, using the opt_narrator output from Character Voices or Voice Designer.
Set the narrator and write your script
Connect an engine to TTS Text or TTS SRT. For reference-based voice cloning, select the engine's cloning mode when it offers multiple modes. Choose the default voice with narrator_voice, or connect Character Voices to opt_narrator.
Welcome to our story. This line uses the narrator.
[Alice] Hi, Bob!
[Bob] Hello, Alice.
[Alice] Shall we begin? [narrator] And so their journey started.
The next line uses the narrator again.
Each new line starts with the narrator. A character tag applies until the next tag on that line. Repeat [Alice] at the start of each line Alice should speak; use [narrator] to switch back within a line.
Use the same tags inside SRT subtitle text. Subtitle timing is controlled separately by the TTS SRT node's timing mode.
Languages and segment settings
| Tag | Meaning |
|---|---|
[Alice] |
Select Alice's voice. |
[narrator] |
Select the narrator. |
[de:Alice] |
Select Alice and request German. |
[fr:] |
Request French with the narrator. |
[Alice|seed:42] |
Select Alice and override the segment's seed. |
Language tags do not translate text or add languages to a model. Some engines infer language from the text; others accept an explicit language or load a language-specific model. See Language Support for engine coverage and the Parameter Switching Guide for supported segment settings.
Language priority is: explicit tag → character's alias default → node/model default.
Aliases (optional)
Character Alias Manager demo (v5.5.3):
https://github.com/user-attachments/assets/960a27d8-c0d8-45e6-b343-7d18bf85d1d6
Open the manager from Character Voices or the Multiline TTS Tag Editor, or use the Open Character Alias Manager command. Assign friendly names, choose voice files, and set optional default languages. Save changes to apply your mappings.
Alternatively, place #character_alias_map.txt in a voice-library root. For example, to address female_01.wav as Alice:
Alice = female_01, de
Bob = male_01, en
[Alice] then selects female_01 with German as its default language; [en:Alice] overrides that language. The language field is optional. Alias Manager user overrides take priority over mappings loaded from files.
If the wrong voice speaks
- Check the audio filename or alias target. Unknown character names fall back to the narrator and produce a console warning.
- Check the transcript requirements for the engine and mode. If you trim a reference, update its transcript to match the retained speech.
- After adding or changing library files, use Refresh Voice Cache to rescan them.
For engine-specific audio preparation, see F5-TTS Reference and Inference Guidelines. For visual tag insertion, see the Multiline TTS Tag Editor Guide.