---
title: Live video
description: Start an endless video in the playground, give it a first frame, a director and a voice, and change what happens next while it plays.
---

The **Live video** playground ([/app/playground?mode=live](/app/playground?mode=live)) plays a video that keeps going while you watch. New footage arrives about every 5 seconds, each shot carrying on from the one before, and you can change what happens next at any moment. It is billed to your balance like an API call. To do the same from code, see [Live video](/docs/models/live-video); **View code** in the header shows cURL and Python for your current settings.

## Pick a model {#model}

The header shows the price per minute, the minimum per session, and the longest a session runs (10 minutes). When more than one live model is offered, pick one from the cards under it; you can't switch while a session runs.

Live models run on GPUs that sleep when nobody uses them. The badge under the model shows **Warm** (ready now), **Waking up** or **Asleep**; when it is asleep, your first session waits up to about a minute while a GPU starts. Only starting a session wakes it. See [Hosted models](/docs/models/hosted#cold-start).

## Describe what happens {#prompt}

Write the scene in **What happens**: who or what is on screen, what they do, where, the light and the camera. Up to 1,000 characters; short prompts render fastest. **Write for me** drafts or improves the text with a chat model, billed to your balance.

- **Shape**: **Landscape** or **Portrait**.
- **Longest it runs**: 1, 2, 5 or 10 minutes. Each option shows the most it can cost. The session ends there, or earlier when you stop it.

## First frame {#first-frame}

The picture the video starts from. Its people, place and look carry through every shot.

- **None**: the video starts from your prompt alone.
- **Upload**: drop a PNG, JPEG or WebP picture, up to 15 MB.
- **Generate**: describe the frame, or leave the box empty to draw the scene from **What happens**, then press **Draw it**. Pick the **Image model**; the picture is billed to your balance at that model's price. It is drawn in the stream's shape. **Draw another** makes a new one.

The picture is cropped to the stream's shape from the center. **Remove** clears it.

## The director {#director}

Turn on **Direct it for me** to have a chat model write every next shot from your story and keep the action moving.

- **Director**: the chat model to use. Each shot is part of a short chat call, billed to your balance at that model's price and listed in your [logs](/app/logs). Pick a fast model: it writes two shots per call and has about 10 seconds for it, or the video pauses until it is done.
- **Notes for the director** (optional): style, pace, where the story should go. Up to 2,000 characters.

Without a director, every shot follows your prompt as written until you change it.

## Fixed camera {#fixed-camera}

**Fixed camera** makes the whole stream one locked-off shot: the camera never moves or cuts and the place stays the same; only the people move. Use it for a talk to the camera, a room or a window.

## Voice {#voice}

Each shot makes its own sound. So that a speaker doesn't get a new voice every few seconds:

- **Keep the first voice** (on by default): the voice heard in the first shot with speech stays for the whole stream. Off: each shot picks its own.
- **Or drop a voice clip to use**: about 5 seconds of clear speech from one speaker, as WAV, MP3, M4A, Ogg, FLAC or WebM, up to 15 MB. Only its first ~5 seconds of speech are used. **Remove** goes back to the checkbox.

## Consent {#consent}

You must tick the consent box before you start. With it you accept the [LTX terms](/docs/models/ltx-license) and their use restrictions, and take responsibility for following them. That includes not impersonating anyone without their consent and saying the video is AI-generated wherever you show it. You also confirm that anyone recognisable in your prompt or first frame is 18 or older and agreed. **Read the terms** opens the full text.

## Go live {#start}

Press **Go live**. While a GPU starts, the player shows **Starting a GPU…** with a timer. Then it renders the first shots, and playback begins once two are ready.

Two players take turns: one plays while the next shot loads hidden in the other, so shots run back to back without a gap. If the next shot isn't ready, the player shows **Buffering**. The corner button toggles sound; if your browser blocks it, playback starts muted.

The corner label shows **LIVE** and the shot number; the line under the player counts shots and seconds made.

## Steer it while it runs {#steer}

While the session runs, a box appears under the player. Type and press **Send**:

- **With a director**, you tell the director what should happen ("a dog runs up to her"). It works that into the coming shots and keeps it in mind.
- **Without a director**, your text replaces the prompt: the next shot to start follows it word for word. The shot rendering now keeps its own prompt.

What you sent is listed as chips under the box. If the director fails to write a shot, a note says so and the video keeps the last prompt.

## The shot list {#shots}

**Shots** lists every shot made with the prompt it followed; shots the director wrote are tagged **director**. The one playing is highlighted. With a director, the shot it has written but not yet rendered appears at the end as **Up next**.

## Stop {#stop}

**Stop** ends the session after the shot in progress. The session also ends when it reaches the length you chose, when your balance runs out, or when you leave the page. Shots already made play out, the corner label changes to **Replay**, and a line says why it ended.

## Price {#price}

You pay per second of video made ($0.019 per second for `ltx-live`), at least a minimum per session ($0.42). The bar next to **Go live** shows the running cost while it plays. Director calls and image-model frames are billed separately. When you start, the most the session can cost is held from your balance and released when it ends. Current prices are on [Models](/models); billing details are in [Live video pricing](/docs/models/live-video#pricing).
