Sound LoRA: train a reusable voice

A Sound LoRA is the audio equivalent of an image LoRA: train it once on samples of a voice or a distinct audio signature, then reuse that exact sound in any clip without re-importing anything.

1. Gather audio samples

Collect several clean recordings of the voice or sound you want to capture. A handful of short, varied clips beats one long noisy recording; variety in tone and pacing gives steadier training.

2. Create the Sound LoRA

Open the Sound LoRA panel, create a new one, give it a name, and import your samples. tendre.AI prepares the audio for training automatically.

3. Train

Press Train and let it run on your GPU. Duration depends on how much audio you provided and your hardware; it works unattended in the background.

4. Use it everywhere

When training finishes, the Sound LoRA appears in your voice list. Select it from the Character voices tool on any clip, no need to import the original sample again.

For a one-off line without training a reusable voice, use Character voices directly with a sample. To add ambience instead of speech, see Add a soundtrack.