Skip to content
Log inSign up free

Prompting guide

How to prompt Grok Video

How to prompt Grok Video, xAI's Grok Imagine, on Mage: a quick start, a fill-in template, and expert tips for camera, references, edits, and sound.

Try Grok VideoGrok Video model page

One continuous tracking shot
Made with Grok VideoOne continuous tracking shot

Quick start

Your first great Grok Video prompt

Everything a beginner needs in one minute: what to know first, the rules that matter most, a template to fill in, and prompts to copy.

What to know first

Grok Video makes one continuous shot with a soundtrack, from a prompt, from reference images, or by editing a short clip you upload. Write the shot the way a camera sees it: the angle, what happens in order, the place, the light, and what we hear. It has no spoken dialogue and no first frame on Mage, so describe sound instead of speech, and treat your images as references, not as the opening picture.

The rules that matter most

  1. Describe one shot, in order

    The shot type and angle, then the action, joined with "then", then the place, the light, a style line, and the sound. A short clip has room for one or two beats, not a story.

  2. End with the sound

    Every clip comes with a soundtrack. Name it last: rain on asphalt, crowd murmur, a slow piano note.

  3. Leave dialogue out

    Grok Video on Mage does not lip-sync speech. For a character who talks, use Cherry 2 Pro or Lemon.

  4. Give each reference one job

    "The woman in @image1 walks the length of the pier wearing the coat from @image2." Two or three references with clear roles beat five that overlap.

  5. Edit one thing at a time

    Upload the clip, name the one change, then list what stays: "Keep everything else exactly as in the original, including her face, hands, pose, and the background."

Fill in this template

Replace each part in square brackets with your own idea, and leave out a part you do not need.

Template
[Shot type and angle] of [subject] [doing what], then [what happens next]. [Where]. [Light]. [Style or mood]. [Sound].
The template, filled in
Low-angle wide shot of a skateboarder in a yellow hoodie rolling down an empty parking-garage ramp, then kicking the board up into his hand at the bottom. Concrete pillars, painted arrows, puddles at the foot of the ramp. Hard afternoon sun slicing through the open side, long shadows. Gritty skate-video look. Wheels roaring on concrete, the sharp clack of the board, a long echo.
  • [Shot type and angle] How the camera sees it: a low-angle wide shot, an extreme slow-motion macro shot, a slow push-in.
  • [subject] Who or what the shot is about, with one or two visible details: a skateboarder in a yellow hoodie.
  • [doing what] The first action, led by a verb: rolling down an empty parking-garage ramp.
  • [what happens next] The second beat, if there is one: kicking the board up into his hand at the bottom.
  • [Where] The place in a few concrete details: concrete pillars, painted arrows, puddles at the foot of the ramp.
  • [Light] Where it comes from and what it does: hard afternoon sun slicing through the open side, long shadows.
  • [Style or mood] One short line: gritty skate-video look, cozy handmade stop-motion, cinematic.
  • [Sound] What we hear: wheels roaring on concrete, the clack of the board, a long echo.

Prompts to start from

  • An establishing shot

    Aerial wide shot of a small fishing harbor at dawn. A single red ferry pulls away from the pier, then turns slowly toward the open sea, leaving a long white wake across the still water. Stacked lobster traps, a row of pastel houses on the hill, gulls circling the masts. Soft pink light low on the horizon, mist over the water. Calm documentary feel. The ferry's diesel hum, gull cries, water lapping at the pilings.

    Why it works: Angle, action in sequence, place, light, mood, and sound, in that order: one clear shot with nothing left to guess.

  • A slow-motion product moment

    Extreme slow-motion macro shot: the cap flips off a cold green glass bottle of lemon soda, then a fine spray of droplets bursts into the air and a thin mist curls out of the neck. Condensation runs down the glass; ice glitters in a metal bucket behind it. Bright summer backlight, sharp highlights on every droplet. Fresh commercial look. A crisp pop, the fizz of bubbles, a short splash.

    Why it works: Slow motion and a macro lens give a short clip enough to show, and the sound line names every action.

  • A miniature that holds still

    Stop-motion claymation, visible fingerprints in the clay, slightly choppy frame rate, a small tabletop set. A round clay hedgehog in a blue apron waters three potted sunflowers on a windowsill, then sets down the watering can and watches as the flowers slowly lift their heads toward the sun. A checkered curtain, a cracked teapot and a sleeping clay cat sit beside the pots. Continuity: every object stays exactly where it is for the whole clip; nothing appears, disappears, duplicates, or changes shape. Warm morning window light, cozy handmade look, a gentle ukulele melody and the trickle of water.

    Why it works: The continuity sentence keeps props from multiplying or wandering, the most common flaw in a busy scene.

  • Your Character in a saved outfit

    Slow push-in on a rooftop garden at golden hour. Then @mara walks in from the left between the planters, wearing @linen-shirt with wide white trousers, stops at the railing and looks out over the city, then turns her head toward the camera with a small smile. Warm low sun behind her, the skyline soft and hazy. High-quality slow motion, highly detailed face. Wind in the leaves, distant traffic, a soft acoustic guitar.

    Why it works: The camera move comes first, each handle has one role, and the prompt describes what they do rather than what the pictures show. Replace the handles with your own Character and Reference.

  • An edit of your own clip

    Make the woman's raincoat bright yellow. Keep everything else exactly as in the original, including her face, hands, pose, the umbrella, and the background.

    Why it works: Upload the clip as the Video: one change and a list of what stays is all an edit needs.

Advanced

Advanced Grok Video prompting

Techniques for more control once the basics work.

Text, references, or an edit

What you attach decides what Grok Video does. With only a prompt, the prompt is the whole shot. With images, saved Characters, or References, those supply the people, products, clothing, or places, and none of them becomes the first frame. With a clip uploaded as the Video, Grok Video edits that clip.

An edit keeps the source clip's length, framing, and resolution, so the duration, aspect ratio, and resolution settings do not apply, and references cannot be added. Trim a long clip to the part you need before you upload it.

  • Grok Video on Mage has no first or last frame. When a clip must open on an exact still, use Lemon.
  • Draft at the lower resolution and a short length to find the motion, then render the keeper at the higher resolution.
  • Clips with reference images run shorter than clips from text alone; the model page shows the limits.

Structure: one shot, in this order

Write one shot in prose, in this order: the shot type and angle; the subject and its action, in sequence, joined with "then"; the place; the light; a short mood or style line; and the sound. Grok Imagine's own example prompts share this shape.

When objects must hold still, add a continuity sentence: "Continuity: every object stays exactly where it is for the whole clip; nothing appears, disappears, duplicates, or changes shape."

  • Keep a short clip to one or two beats. A clip of a few seconds has no room for a story; a longer clip has room for one more beat, not five.
  • With references, open with the camera move, then bring each referenced subject in with its role and its place in the scene.
  • There is no negative prompt. Say what should happen and what should stay, in positive sentences.

Camera and motion

Name one camera behavior and give it a direction: "The camera dollies from the sunlit doorway to the window." Grok Video reads plain camera language well, and slow, smooth moves hold up better than frantic ones.

Time words change the whole look of a clip: a time-lapse compresses hours, slow motion stretches a second, and an extreme slow-motion macro turns a splash into a sculpture.

WriteWhat it does
slow zoom inThe view narrows gradually toward the subject.
slow smooth push-inThe camera glides forward toward the subject.
the camera dollies from … to …The camera travels along a path you name.
The camera never moves.A locked-off shot: only the subject moves.
Only the camera moves.The scene stays frozen while the camera travels, good for products and still lifes.
high-angle wide shotLooking down on a whole scene from above.
action POV with extreme camera motionA first-person view that rushes, swerves, and shakes.
slow time-lapse feelTime runs fast: crowds stream, clouds race, light shifts.
high-quality slow motionMovement stretched out, every detail visible.
extreme slow-motion macro shotA tiny subject, very close, nearly frozen in time.

References and Characters

Upload images and mention them as @image1, @image2; mention saved Characters and References by their @handle. Mage numbers them in order (your uploads, then Characters, then References) and rewrites each mention to match, so @mara becomes "character from image 3" and a saved outfit @linen-shirt becomes "outfit from image 4".

Give each reference one role and a fixed place in the scene: "the model from @image1 walks toward the camera wearing the shirt from @image2". Describe what the people and things do in the shot, not what the pictures show.

  • Keep the cast small, two or three at most, and keep props still. Props that move, pass from hand to hand, or multiply are where references break.
  • For a product, name what must not change: "the bottle, cap and label text stay unchanged while only the camera moves".
  • xAI's own documentation tags references as <IMAGE_0>, <IMAGE_1>. On Mage, use @image1 and @handles instead, and never mix the two in one prompt.
  • An @ followed by a letter is always read as a saved handle, so an email address or an unsaved handle stops the generation before anything is charged.

A product that must not change

Slow smooth push-in on a white marble bathroom counter at sunrise. The perfume bottle from @image1 stands in the center beside a folded linen towel and a small sprig of eucalyptus, while warm light slowly drifts across the counter and catches the facets of the glass. The bottle, cap and label text stay unchanged while only the camera moves. Clean, premium commercial look. Soft room tone, a faint drip from the tap, a single low piano chord.

Why it works: The product has one role and one place, the scene around it holds still, and the continuity line protects the label.

A story in several shots

Each generation is one shot. For a sequence, write one prompt per shot and use the same @handles or the same approved still in every one, so the character carries across.

For the closest match, make the keyframes first: one image of the character, then edits of it for each new scene and angle (on Grok Image or Mango, for example). Pass each keyframe as the reference for its shot, or animate it on a model that starts from a first frame, such as Lemon.

Editing a clip

Upload the clip as the Video and write one change, then what stays: "Change the color of the car to matte black. Keep everything else exactly as in the original, including the driver, the street, and the camera movement." A full description of the scene is not needed: the clip already has it.

  • To compare options, run several single-change edits of the same clip at once (a red dress, a black hat, a different jacket), each with its own keep sentence, and pick the best.
  • An edit makes new sound. To keep the original clip's audio, put it back over the result in an editor.

Sound and music

Every Grok Video clip comes with a generated soundtrack, and there is no switch for a silent clip. Write the sound into the prompt, last: "gentle music box melody", "rain on asphalt, distant traffic, a passing tram bell", "crowd murmur and footsteps".

Grok Video on Mage has no lip-synced speech, no voice references, and no Character voices. Leave quoted lines out; for a character who talks, use Cherry 2 Pro or Lemon. For a silent clip, mute it in an editor.

Text on screen

Generated text is not a strength. Keep on-screen text to what already exists, such as a label in a reference image, and say that it stays unchanged. Add titles and captions in an editor afterwards.

Common mistakes

Avoid this, do this instead

  • Avoid: A description of what each reference image shows.

    @image1 is a woman with long red hair in a green dress. @image2 is a beach at sunset.

    Instead: What the referenced people and things do in the shot.

    Slow push-in. The woman from @image1 walks barefoot along the beach from @image2 at sunset, then stops and lets the wind lift her hair. Waves wash in, gulls call.

  • Avoid: Five references with overlapping roles.

    Instead: Two or three, each with one role.

  • Avoid: Props that move, change hands, or multiply.

    Instead: Props that stay put, plus the continuity sentence.

  • Avoid: A full scene description for an edit.

    Instead: The one change, then "Keep everything else exactly as in the original, including …".

  • Avoid: Dialogue in quotes.

    The barista smiles and says "Your usual?"

    Instead: The sounds and the music of the scene.

    The barista smiles and slides a cup across the counter. Espresso machine hiss, cups clinking, low café chatter.

  • Avoid: <IMAGE_0> tags beside @ mentions.

    Instead: @image1 and @handles only.

  • Avoid: A whole story in one short clip.

    Instead: One or two beats per clip, and one clip per shot.

FAQ

Grok Video prompting questions

How long should a Grok Video prompt be?
Three to six sentences for one shot: the angle, the action in order, the place, the light, a style line, and the sound. A short clip holds one or two beats; more than that gets rushed or skipped.
Is Grok Video the same as Grok Imagine?
Yes. Grok Video is xAI's Grok Imagine video model, run by xAI and reached through Mage.
Can Grok Video make characters talk?
Not on Mage: it makes a soundtrack, but no lip-synced speech. For a character who speaks, use Cherry 2 Pro or Lemon.
Can Grok Video start from my image?
No. On Mage its images are references for people, products, clothing, or places, not a first frame. To open a clip on an exact still, use Lemon.
How do I keep a character consistent in Grok Video?
Save the person as a Character and mention the same @handle in every shot, with one role and a fixed place each time. For a sequence, make keyframe images of them first and use each as the reference for its shot.
Does Grok Video support negative prompts?
No. Write what must stay in positive sentences instead, such as the continuity sentence for a scene or "Keep everything else exactly as in the original" for an edit.
Can I make a silent clip with Grok Video?
Every clip comes with sound. Mute it in an editor, or describe only quiet room tone in the prompt.

Make your first video with Grok Imagine

New accounts get 300 free Gems, enough to try Grok Video. No install, no credit card.

Try Grok Video