---
title: Create video
description: Cast saved characters and places, write each shot with @tags and dialogue, render the shots and join them into one video with sound.
---

**Create video** ([/app/playground?mode=videos&tool=create](/app/playground?mode=videos&tool=create)) builds a video out of shots. Each shot is a prompt that tags the characters and places in it, and renders 5 to 15 seconds of video with sound and dialogue. When every shot is rendered, you join them into one film. Renders are billed to your balance like API calls. To do the same from code, see [Video shots](/docs/api/video-shots).

The header shows the price per second, the extra price per picture, the 5–15 s range per shot, and that joining is free. Prices are explained under [Video shots pricing](/docs/api/video-shots#pricing) and listed live on [/models](/models).

## Cast characters and places {#cast}

A shot can only tag what is **In this video**. On a computer, the **Characters & scenes** panel on the left lists your saved characters and scenes; on a phone, tap **Characters & scenes** above the cast. Tap a card to add it to the video, tap it again (or **×** on its chip) to take it out. **Manage** opens the [library](/docs/playground/library).

- **New character**: start **From a photo** or **From the description**. An image model draws a character sheet (full-body views, close shots and face close-ups) that every shot follows. Give them a **Name** and, optionally, a **Voice**, then **Save character**.
- **New scene**: give the place a **Name**, describe it under **The place**, then **Draw it** or **Upload** a picture (PNG, JPEG or WebP, up to 20 MB). A place works best without people in it.

Drawing a sheet or a place is billed at the image model's price, shown next to the button.

## Set up the video {#setup}

- **Shape**: 16:9, 9:16 or 1:1, for every shot.
- **Style**: the look of the whole video, for example "cinematic 35 mm film, warm dusk light". Up to 200 characters.
- **Dialogue language**: the language the characters speak in.

Changing the shape marks rendered shots as **Edited**, so they render again before you join.

## Write a shot {#shots}

Type what happens in the shot box. Type `@` to tag a character or scene and pick it from the list. A name with spaces can also be written as `@[Rooftop bar]`. Each tagged character or scene sends its picture with the shot:

- Tag at least one character or scene. If the video has exactly one character and the shot has no tags, that character is used.
- A shot can use up to 9 pictures.
- A tag that isn't in this video is flagged as **Not in this video**. If it matches something in your library, **Add {name} to this video** fixes it in one tap.

Put spoken lines in quotes: `@Leo: "You made it!"`. A line is spoken by the first character named in its sentence. A character with a voice speaks in it; up to 3 voiced characters per shot bring their own voice.

Set the **Length** slider from 5 to 15 seconds. Under it you see the price of the shot and the seconds it renders. **Write for me** in the box drafts or improves the prompt with a chat model, billed to your balance.

Use the arrows or drag a shot's handle to reorder it. **Duplicate** copies it, the bin deletes it, and **Add a shot** adds one at the end.

## Start a shot on your own picture {#first-frame}

There is no separate first-frame upload. To open a shot on a picture you have, save it as a scene and say in the prompt that the shot starts on it:

1. **New scene**, name it (for example "Opening frame"), **Upload** the picture and save.
2. Add it to the video.
3. Write: `The shot opens exactly on @[Opening frame] as its first frame. @Priya turns from the window and walks toward the camera.`

With the [rewrite box](#rewrite) ticked, the picture is given to the model as the shot's real first frame (or last frame, if the prompt says the shot ends on it). The video starts on it, a close match though not pixel for pixel, and keeps the picture's brightness and colours. The shot needs one more picture, such as a character; a lone opening picture is used as a reference instead. People in the picture are not kept from it: to keep someone's look, also save them as a character and tag them. Pick the shape closest to the picture. The picture counts toward the price like any other. See [Starting from a picture](/docs/api/video-shots#first-frame).

Instead of wording, each shot has **First frame** and **Last frame** pickers (shown for models that take frames, once the video has two or more characters and scenes): choose the picture from the video's cast, or leave it on Automatic to let the prompt decide.

## Plan with the director {#director}

**Plan with the director** writes shots for you. Describe **The story** in a sentence or two, choose the **Number of shots** (or let it decide), and pick the **Director model**. It can also create new characters and places for the video, up to the limits you set in **New characters, up to** and **New places, up to**. Press **Plan shots**, then **Add after my shots** (or **Add all**) or **Replace storyboard**.

On a single shot, **Direct** writes or rewrites it to fit the shots around it, with an optional note such as "make it funnier". **Undo** restores your text. Director calls and any pictures it draws are billed to your balance.

## Render {#render}

Check the terms box first (see [consent](#consent)). Then press **Render** on a shot, or **Render {n} shots** in the bar at the bottom to render every shot that is ready. The bar shows the estimated price of what is left to render.

A rendering shot shows its progress and elapsed time; after a quiet spell, the GPU takes about a minute to wake up. A finished shot plays in its card, with its rendered length, how long it took, and its charge. If you edit the prompt, length or shape afterwards, it shows **Edited since this render** and the button becomes **Re-render**. Failed and cancelled shots cost nothing.

## Join the shots {#join}

When every shot is rendered and there are at least two, press **Join**. Joining is free. **Your film** appears below the shots: the shots in order, each with its own sound. **Download** saves the mp4. If you change shots after joining, the film says so and **Join** makes a new cut.

## Your videos {#projects}

Each video is a project, saved as you type (**Saved** next to the name). Rename it in the title field, switch with **Your videos**, start another with **New video** (it keeps the cast, shape, style and language), or delete it with the bin. Projects are free. From code, see [Projects](/docs/api/video-shots#projects).

## The rewrite box {#rewrite}

**Rewrite my prompt for MiniMax H3**, above the terms box, is ticked by default, and your browser remembers if you untick it. You don't need to follow MiniMax H3's [prompting guide](https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md). Before a shot first renders, a vision chat model reads that guide, the shot's pictures and your prompt, and writes the whole H3 prompt from them. It looks at the pictures, works out from your words what each one is (a character to keep, a place, or a frame such as the shot's first frame) and keeps everything you ask for, such as who doesn't speak, a camera move or no music. That prompt is what renders. Rendering starts 10–45 seconds later, and the chat call is billed to your balance. Untick the box only if you write H3 prompts yourself. See [the rewrite](/docs/api/video-shots#rewrite).

## The terms box {#consent}

Rendering needs the checkbox above the film. By checking it you accept the [MiniMax H3 terms](/docs/models/minimax-h3-license), including not using the model or its videos in the EU, the UK, South Korea or the US, and you confirm that everyone shown is 18 or older and agreed to this use of their likeness. The box is not remembered between visits.
