LTX-2.5 Pro

Introduction

LTX-2.5 Pro can cut between two and four distinct shots inside a single generation, holding the same character, lighting, and voice across the cuts. You write the whole sequence as one chronological paragraph and name each cut in prose, and the model returns a small edited scene built from one prompt.

That clip is one call. The prompt runs an aerial establishing shot, cuts to the hands on the wheel, and match cuts to the car cresting the ridge, keeping the same car and a continuous engine track throughout. This guide covers the request, the anatomy of a cut, giving each shot a job, holding identity across cuts, and when a single take is the better call. It builds on the prompting guide, whose single-shot fundamentals every shot here still follows.

The request

Nothing new in the payload. A multi-shot clip is the same videoInference call, and the whole edit lives in the positivePrompt.

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 cinematic car commercial in three shots. A wide aerial establishing shot frames a matte-grey sports car carving along a mountain road at dawn, low engine hum and wind in the air. A hard cut transitions to a tight interior close-up of the driver\'s gloved hands on the stitched wheel, the engine note continuing across the cut. A final match cut jumps to a low three-quarter shot of the same matte-grey car cresting the ridge into the rising sun, the engine swelling. Cinematic teal-and-amber grade, anamorphic look.',
  width: 1920,
  height: 1080,
  duration: 10,
  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 cinematic car commercial in three shots. A wide aerial establishing shot frames a matte-grey sports car carving along a mountain road at dawn, low engine hum and wind in the air. A hard cut transitions to a tight interior close-up of the driver's gloved hands on the stitched wheel, the engine note continuing across the cut. A final match cut jumps to a low three-quarter shot of the same matte-grey car cresting the ridge into the rising sun, the engine swelling. Cinematic teal-and-amber grade, anamorphic look.",
            "width": 1920,
            "height": 1080,
            "duration": 10,
            "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": "c4a9b3d2-6e7f-4801-c2b3-d4e5f6071829",
      "model": "lightricks:ltx@2.5-pro",
      "positivePrompt": "A cinematic car commercial in three shots. A wide aerial establishing shot frames a matte-grey sports car carving along a mountain road at dawn, low engine hum and wind in the air. A hard cut transitions to a tight interior close-up of the driver's gloved hands on the stitched wheel, the engine note continuing across the cut. A final match cut jumps to a low three-quarter shot of the same matte-grey car cresting the ridge into the rising sun, the engine swelling. Cinematic teal-and-amber grade, anamorphic look.",
      "width": 1920,
      "height": 1080,
      "duration": 10,
      "settings": {
        "audio": true
      }
    }
  ]'

runware run lightricks:ltx@2.5-pro \
  positivePrompt="A cinematic car commercial in three shots. A wide aerial establishing shot frames a matte-grey sports car carving along a mountain road at dawn, low engine hum and wind in the air. A hard cut transitions to a tight interior close-up of the driver's gloved hands on the stitched wheel, the engine note continuing across the cut. A final match cut jumps to a low three-quarter shot of the same matte-grey car cresting the ridge into the rising sun, the engine swelling. Cinematic teal-and-amber grade, anamorphic look." \
  width=1920 \
  height=1080 \
  duration=10 \
  settings.audio=true

{
  "taskType": "videoInference",
  "taskUUID": "c4a9b3d2-6e7f-4801-c2b3-d4e5f6071829",
  "model": "lightricks:ltx@2.5-pro",
  "positivePrompt": "A cinematic car commercial in three shots. A wide aerial establishing shot frames a matte-grey sports car carving along a mountain road at dawn, low engine hum and wind in the air. A hard cut transitions to a tight interior close-up of the driver's gloved hands on the stitched wheel, the engine note continuing across the cut. A final match cut jumps to a low three-quarter shot of the same matte-grey car cresting the ridge into the rising sun, the engine swelling. Cinematic teal-and-amber grade, anamorphic look.",
  "width": 1920,
  "height": 1080,
  "duration": 10,
  "settings": {
    "audio": true
  }
}

[
  {
    "taskType": "videoInference",
    "taskUUID": "c4a9b3d2-6e7f-4801-c2b3-d4e5f6071829",
    "videoUUID": "9a2b3c4d-5e6f-7890-abcd-ef1234567890",
    "videoURL": "https://vm.runware.ai/video/os/a14d18/ws/2/vi/9a2b3c4d-5e6f-7890-abcd-ef1234567890.mp4"
  }
]

Two things shape a multi-shot prompt:

• Write the sequence as one chronological paragraph. Skip numbered beats and screenplay sluglines unless you also describe the cut in prose, since the model reads a flowing scene, not a shot list.
• Give the cuts room to read: two to four shots fit an 8 or 10 second clip. Pack more cuts into the same runtime and each shot gets too little time to register.

Anatomy of a cut

A cut is where multi-shot prompts succeed or fall apart. Spell out four things at every cut or the model fills the gaps on its own and the sequence drifts.

• Name the transition. "A hard cut transitions to", "a match cut connects", "the image dissolves into". The words tell the model an edit happens here, not a camera move within the same shot.
• Re-establish the shot. State the new scale, the angle, who or what is in frame, and the lighting if it changed. The model does not carry the previous framing forward for you.
• Keep identity consistent. Reuse the same visual tag for a recurring person or object, so they come back as themselves rather than a lookalike.
• State the audio. Say whether the music, ambience, or dialogue continues, softens, or changes across the cut, so the track reads as one scene instead of resetting.

The two-shot clip below does all four. It names the hard cut, moves from a medium-wide to a close-up, keeps the same cook and kitchen, and carries the room ambience across while the sizzle rises over it.

Giving each shot a job

A sequence reads when each shot does a specific job and they build in order. Two patterns carry most of the work: wide to close (establish the space, then the subject, then a detail) and establish, detail, reaction. The hotel promo below runs wide to close, and each cut hands off cleanly to the next.

The wide sets the place, the medium gives the subject something to do, and the close-up lands the moment the shot was built around. Order the jobs so each cut answers the last. A sequence of three wides, or three unrelated close-ups, has no build and reads as clips stapled together.

Holding identity across cuts

Across a cut the model has no memory of the exact face it drew a second earlier, so you re-identify a subject every time it reappears. Reuse a short visual tag at each shot and the same person comes back. Drop it and you often get a different one. The street-style clip below names "the man in the camel overcoat" in all three shots, and he stays the same man from the sidewalk to the crosswalk to the lobby.

Anchor the tag on something concrete and repeatable, the wardrobe, a color, a distinctive object, rather than a name the model cannot see. The same trick holds a product steady across a spot: "the same matte-grey car" in the hero above returns as the same car in every shot.

When to stay single-shot

Cuts are not always the answer. Stay with one continuous take when the shot lives on unbroken camera motion or an intimate, lip-synced performance, where a cut would break the very thing the shot is about. Animating from a first frame reads best as a single take too, covered in the first and last frame guide, unless you deliberately cut away from the opening image.

Tips

1. Write one chronological paragraph. Describe the shots in order as flowing prose, not a numbered shot list, and let each cut fall where you name it.

2. Name every transition. "A hard cut", "a match cut", "dissolves into". Without the words the model reads a camera move and keeps one shot going.

3. Re-establish each shot. Restate scale, angle, framing, and any lighting change at the cut. The model does not inherit the previous shot's setup.

4. Re-identify recurring subjects. Reuse a concrete visual tag ("the same matte-grey car", "the man in the camel overcoat") at every shot so identity survives the cut.

5. State audio across cuts. Say whether music or ambience continues, softens, or changes, so the track reads as one scene.

6. Keep it to two to four shots. Give each a job, wide to close or establish and pay off, and match the duration to the count with an 8 or 10 second clip.