Save a voice, the audio equivalent of a LoRA

tendre.red keeps your voices in a library that works just like the LoRA library. Instead of re-uploading a sample every time, you import a voice once and it is saved as a reusable asset you can attach to any character, exactly as a LoRA stores a reusable face.

1. Import a voice

Open Library > Voices and click Import a voice. Pick a clean recording, 10 to 30 seconds, a single speaker with little background noise (wav, mp3, m4a, ogg, flac or aac). The file is copied into the app's voice library and given a name you can change later by double clicking it.

2. It is saved like a LoRA

Where a character LoRA stores a face you reuse across generations, a saved voice stores a timbre you reuse across clips. It stays in your library for good, with no training step and no queue to wait on: the reference sample itself is the asset.

3. Assign it to a character

Attach the voice to a character's profile so every clip of that character adopts the same timbre, or pick it from the Voice dropdown when you render a clip. The Used by column shows which characters rely on each voice.

4. Reuse it across clips

When you generate a video with the Cloud engine, the LTX 2.3 Full model, and Sound on, tendre.red converts the spoken dialogue to your saved voice and lip syncs it, so the character keeps one consistent voice from one clip to the next. The faster LTX 2.3 Turbo model renders with sound too, but always uses the engine's default voice.

For the full import and management walkthrough, see Character voices. To add ambience or effects instead of a voice, see Sound effects.