> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mage.space/llms.txt
> Use this file to discover all available pages before exploring further.

# Prompting for video

> Video prompts describe motion, not a still image. What to include, how to carry a Character into a clip, and which controls pin the first and last frame.

The most common mistake in video prompting is describing a photograph. A still prompt gives you a still image that happens to move slightly, because nothing in it told the model what should change over time.

## Describe the motion, not the still

| Reads as a still                          | Reads as a clip                                                                      |
| ----------------------------------------- | ------------------------------------------------------------------------------------ |
| a woman in a red coat at a train platform | a woman in a red coat turns to look down the platform as a train pulls in behind her |
| a cup of coffee on a table, steam         | steam rises from a cup of coffee as a hand reaches in and lifts it out of frame      |

## What belongs in a video prompt

* The subject, same as an image prompt.
* What the subject does, as an action with a direction to it.
* What the camera does, if anything. Say when it holds still, because otherwise the model may decide for you.
* The pacing, when it matters. A slow drift and a quick turn are different clips.

Keep it to one or two actions. A clip of a few seconds cannot deliver a sequence of five events.

## Carrying a Character into a clip

Call a Character with its @ handle the same way you would in an image prompt. Character support is per model: Berry, Cherry, Grok Video, Lemon, Melon, MiniMax-H3, Plum, Raspberry all take Characters. See [Characters](/features/characters).

## Pinning the start and the end

Where the opening and closing image matter more than the description, pin them directly. A first and last frame is supported on Blueberry, Framepack Image-To-Video, Lemon, Melon, MiniMax-H3, Peach Max, Plum, Veo 3.1 Video, Wan Video v2.1. The prompt then describes only the motion between the two frames. On the Mage models this is a separate mode from Characters and References. See [First and Last Frames](/features/first-and-last-frames).

## Prompting sound

Some video models generate audio alongside the clip. Where they do, describe the sound you want the way you describe the picture, and keep it to what would be audible in the shot. For a Character who should sound the same across clips, use a Character Voice rather than describing the voice each time. See [Character Voices](/features/character-voices).

## Limits

Duration and resolution ceilings are set by plan and model. Asking for a longer clip does not extend it. See [video models compared](/models/video-models-compared). Video costs more Gems than images, so test the composition as a still on a cheap image model first, then carry the result into video as a reference.