LTX-2.5 Pro

by LightricksAugust 11, 2026

LTX-2.5 Pro is the higher-capability model in the LTX 2.5 video family, built for production-quality multimodal video creation and transformation. It supports text-to-video, image-to-video, and audio-to-video workflows together with retake, extension, and reframing operations, pairing synchronized audio generation with higher-resolution output and longer-form clip control. It is well suited to commercial video production, concept films, storyboards, speaking-character shots, and other workflows that need stronger quality, broader modality support, and more flexible post-generation iteration than the Fast tier.

Native audio with LTX-2.5 Pro

How to generate synchronized audio with LTX-2.5 Pro: turning on settings.audio, prompting ambient sound and effects, directing spoken dialogue with accent and lip-sync, and balancing the mix.

Introduction

LTX-2.5 Pro generates the soundtrack in the same pass as the picture, not as a separate step afterward. Turn it on and describe what you want to hear, and the model scores the scene: an ambient bed, effects that hit with the action, music, and spoken dialogue that stays in lip-sync. Sound stops being a post-production job and becomes part of the prompt.

One call, with a drum groove locked to the hands. Turn the sound on.

A music-video performance shot, a medium close-up. A drummer in a black tank top plays hard behind a vintage kit in a dim studio lit by a single warm spotlight, sweat flying and cymbals shimmering as the sticks blur across the toms and snare. The camera pushes in slowly and drifts slightly to his right, catching the flare off the cymbals. Moody high-contrast grade, deep shadows, crisp motion detail. The audio is the star: a driving, tight drum groove locked exactly to his hands, the sharp crack of the snare and the wash of the cymbals filling the room, a warm low room tone underneath.

Play the clip with sound: the groove lands on the sticks, frame by frame. This guide covers switching audio on, prompting the ambient bed and effects, directing spoken dialogue, and balancing the mix. It is about the audio LTX generates for you from the prompt. To drive a shot from an audio track you already have, see the audio-driven guide.

The request

Audio is a single switch on the same videoInference call. The sound itself comes from the prompt.

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'lightricks:ltx@2.5-pro',
  positivePrompt: 'A music-video performance shot, a medium close-up. A drummer in a black tank top plays hard behind a vintage kit in a dim studio under a single warm spotlight, the sticks blurring across the toms and snare as the camera pushes in. Moody high-contrast grade, deep shadows. The audio is a driving, tight drum groove locked to his hands, the crack of the snare and the wash of the cymbals filling the room, a warm low room tone underneath.',
  width: 1920,
  height: 1080,
  duration: 8,
  settings: {
    audio: true
  }
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "lightricks:ltx@2.5-pro",
            "positivePrompt": "A music-video performance shot, a medium close-up. A drummer in a black tank top plays hard behind a vintage kit in a dim studio under a single warm spotlight, the sticks blurring across the toms and snare as the camera pushes in. Moody high-contrast grade, deep shadows. The audio is a driving, tight drum groove locked to his hands, the crack of the snare and the wash of the cymbals filling the room, a warm low room tone underneath.",
            "width": 1920,
            "height": 1080,
            "duration": 8,
            "settings": {
                "audio": True
            }
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "videoInference",
      "taskUUID": "d5b1c4e3-7f8a-4912-d3c4-e5f607182a3b",
      "model": "lightricks:ltx@2.5-pro",
      "positivePrompt": "A music-video performance shot, a medium close-up. A drummer in a black tank top plays hard behind a vintage kit in a dim studio under a single warm spotlight, the sticks blurring across the toms and snare as the camera pushes in. Moody high-contrast grade, deep shadows. The audio is a driving, tight drum groove locked to his hands, the crack of the snare and the wash of the cymbals filling the room, a warm low room tone underneath.",
      "width": 1920,
      "height": 1080,
      "duration": 8,
      "settings": {
        "audio": true
      }
    }
  ]'

runware run lightricks:ltx@2.5-pro \
  positivePrompt="A music-video performance shot, a medium close-up. A drummer in a black tank top plays hard behind a vintage kit in a dim studio under a single warm spotlight, the sticks blurring across the toms and snare as the camera pushes in. Moody high-contrast grade, deep shadows. The audio is a driving, tight drum groove locked to his hands, the crack of the snare and the wash of the cymbals filling the room, a warm low room tone underneath." \
  width=1920 \
  height=1080 \
  duration=8 \
  settings.audio=true

{
  "taskType": "videoInference",
  "taskUUID": "d5b1c4e3-7f8a-4912-d3c4-e5f607182a3b",
  "model": "lightricks:ltx@2.5-pro",
  "positivePrompt": "A music-video performance shot, a medium close-up. A drummer in a black tank top plays hard behind a vintage kit in a dim studio under a single warm spotlight, the sticks blurring across the toms and snare as the camera pushes in. Moody high-contrast grade, deep shadows. The audio is a driving, tight drum groove locked to his hands, the crack of the snare and the wash of the cymbals filling the room, a warm low room tone underneath.",
  "width": 1920,
  "height": 1080,
  "duration": 8,
  "settings": {
    "audio": true
  }
}

Response

[
  {
    "taskType": "videoInference",
    "taskUUID": "d5b1c4e3-7f8a-4912-d3c4-e5f607182a3b",
    "videoUUID": "1b2c3d4e-5f6a-7890-bcde-f12345678901",
    "videoURL": "https://vm.runware.ai/video/os/a14d18/ws/2/vi/1b2c3d4e-5f6a-7890-bcde-f12345678901.mp4"
  }
]

settings.audio is a boolean. Set it to true for a synchronized track, or leave it off for a silent render.
The audio comes from the prompt. Name the sound you want in plain language and the model builds it to match the scene. Leave it unnamed and the model fills the track with its own judgment.
Nothing else changes. Audio generation runs on the same size, duration, and frame-rate options as a silent clip.

Prompting the soundscape

Two layers make a scene sound real: an ambient bed that sets the space, and effects tied to the action that land on the motion. Name both, and attach each effect to the thing that causes it. The coastal clip below runs a steady sea ambience under sharp, specific footsteps in the wet sand.

A steady sea ambience under specific, motion-locked footsteps

A calm travel lifestyle scene, a wide shot easing into a medium at golden hour. A woman in a linen shirt walks slowly along a wet sand beach at low tide, small waves sliding in and out around her bare feet, gulls drifting overhead against a soft pastel sky. The camera tracks alongside her at a gentle pace. Warm natural light, airy color grade, shallow depth of field. The soundscape is layered and specific: the wash and pull of the waves, the soft crunch and squelch of her footsteps in the wet sand, distant gull calls, and a light steady sea breeze, no music.

Specific beats generic. "The wash and pull of the waves" and "the squelch of her footsteps in the wet sand" give the model something to sync, where "beach sounds" would leave it vague. Add "no music" when the clip will sit under a soundtrack you drop in later, so the model scores the room instead of writing a theme that fights yours.

Directing dialogue

For spoken lines, put the exact words in quotation marks and the model renders them as speech, lip-synced to the character. Name the language and accent and the delivery, and keep the line short enough to fit the clip's runtime.

A quoted line, spoken in a named accent and lip-synced

A friendly brand spokesperson shot, a clean medium close-up. A young woman with a warm smile stands in a bright modern studio with a soft green backdrop, looking directly into the camera and speaking to the viewer, her hands gesturing lightly. She says, in a warm and clear American accent, "If you can picture the shot, you can hear it too, all in one take." Her lips and expression stay in sync with every word. Bright soft key light, clean commercial color grade, shallow depth of field. The audio is her clear spoken voice up front, with a quiet studio room tone behind it and no music.

The quotation marks tell the model the words are spoken, not a caption on the frame. Direct the delivery with a physical or tonal cue, "a warm, measured voice" or "an excited near-shout", rather than a bare label. The same pattern carries beyond English: name the language ("speaking in French") for another tongue, or note that the character sings for a vocal, and keep to one speaker per line so the lip-sync stays clean.

Balancing the mix

A convincing track has levels, not a pile of sounds at once. Put the voice up front, the action effects with the motion, and the ambient bed underneath, and the mix reads the way an edit would. Volume words shape it further, "a whisper", "a distant murmur", "a sudden shout", so the model knows what sits loud and what sits low.

"No music" is the single most useful audio direction. Without it the model often scores a scene on its own, which is a problem when the clip is headed for a soundtrack you control. Add it to any prompt where you plan to lay in your own music later.

Tips

Switch audio on with settings.audio. Set it to true for a synced track and off for a silent clip. It changes nothing else about the request.
Name the ambient bed and the effects. Give the space a base layer and tie each effect to what causes it, so the model has motion to sync against.
Quote spoken lines. Wrap dialogue in quotation marks and name the language and accent, and the model speaks it in lip-sync.
Direct delivery with a cue, not a label. "A warm, measured voice" or "an excited near-shout" beats "happy", the same way physical cues beat emotion labels for the picture.
Shape the mix with volume words and "no music". Whisper, murmur, and shout set the levels, and "no music" keeps the model off a soundtrack you plan to add yourself.
Keep dialogue short and single-speaker. One voice per line and a line that fits the runtime keep the lip-sync tight.