usamaehsan/ltx-voice

LTX-2.5 voice only: expressive text-to-speech from the audio half of LTX-2.5 (no video). Designed voices from a text prompt, no voice cloning.

Public
1.7K runs

Run usamaehsan/ltx-voice with an API

Use one of our client libraries to get started quickly. Clicking on a library will take you to the Playground tab where you can tweak different inputs, see the results, and copy the corresponding code to use in your own project.

Input schema

The fields you can use to run this model with an API. If you don't give a value for a field its default value will be used.

Field Type Default value Description
prompt
string
Describe the speaker, the emotion and the room, and put the exact words in quotes, e.g. A tired old man says softly: "I waited all night for you." Quiet room, no music.
duration
number
0

Max: 20

Seconds of audio, rounded up to the next 0.5 s. 0 = let the model pick from the prompt.
seed
integer
-1
Random seed. -1 = random.
steps
None
8
Denoising steps from the distilled schedule (8 = full quality).
refine
boolean
False
Extra 3-step refine pass (like stage 2 of the video model).
ancestral
boolean
True
Ancestral sampler (what LTX-2.5 uses).
output_format
None
mp3
Audio format (mp3 = mono 96 kbps, wav = stereo 16-bit).
trim_text
boolean
True
Run the text encoder on the real prompt tokens only (fast).
profile
boolean
False
Speed-test debug: return the torch.compile mega-cache of this replica (a .bin) instead of audio.

Output schema

The shape of the response you’ll get when you run this model with an API.

Schema
{
  "type": "string",
  "title": "Output",
  "format": "uri"
}