Grok Imagine 1.5 prompt guide

This guide works from xAI’s video generation guide and the example prompts in its video overview. For the model, see the Grok Imagine 1.5 page.

xAI’s docs say 1.5 turns text into video in two hidden steps: it renders a first frame from the prompt, then animates that frame. So a prompt’s first sentence is a picture. xAI’s advice is to describe the shot, including camera movement, lighting and sound; its own examples add a line on what must not change.

1. 1Open with the shot and what the first frame shows: the angle or shot size, the subject, the place.
2. 2Say what happens, in order, with a trigger where it helps: “When the lights change…”.
3. 3Add one camera move, or write “The camera never moves.”
4. 4Add a continuity line for what must stay: “every object stays exactly where it is”.
5. 5Set the light and the look.
6. 6End with the sound: each line in quotes after its speaker, then “Sound:” for the rest, and whether there is music.

xAI’s example prompts write the camera in plain film words, and often say what doesn’t move as well as what does. Pick one move per shot and give it a speed.

When the camera should be the only thing moving, say so, as xAI’s perfume ad does: “Only the camera moves.” For a rough, physical feel, describe what causes the shake, as its dragon-flight example does: wingbeats, turbulence, sudden turns.

xAI says 1.5 makes effects, ambience and dialogue in the same pass, landing on the action, with lip-synced speech that is clearer and better synced than its previous model. Its examples write sound as directions you can hear. See native audio.

• Name the speaker, give the delivery, then the line in quotes: Quietly, cautious, voice tight: “We? You didn’t even ask me.”
• Put an acting beat inside a line in square brackets, as xAI’s café scene does: “I am. [she looks down, pauses] I just thought…”
• Write the room under the dialogue on a “Sound:” line, and say if it should never go silent between lines.
• Write “No music” when you want none; xAI’s talking-character recipe asks for no other speech and no music.
• Fit the words to the seconds: the same recipe sets the clip’s length from the number of words in its lines.

xAI’s docs don’t list the languages 1.5 speaks, so check each take of a line in another language. xAI’s API can give up to 3 speakers a preset voice; on DreamBench the voice comes from your description. See lip sync.

xAI’s API takes 1 to 15 seconds and defaults to 8; DreamBench’s Video page offers 4, 5, 6, 8, 10, 12 or 15. xAI’s examples run from one line to several hundred words; on DreamBench a Grok prompt can run to 2,500 characters.

In xAI’s API, keyframes pin pictures at chosen seconds; in a prompt, order does that job: “then”, “when”, “as he talks”. On DreamBench, Grok renders 16:9 for YouTube, 9:16 for Reels, TikTok and Shorts, 1:1 and 4:3; xAI’s API also offers 21:9, 3:2 and others.

In image to video, the still becomes the first frame, the prompt describes what happens next, and the clip keeps the still’s shape. xAI’s launch post puts it in one line: give it a starting image, describe the motion. See image to video.

1. 1Don’t describe the picture again: it is already the first frame.
2. 2Say what moves, in order: “Make her smile, wave and then turn around again.”
3. 3Give direction and speed where they matter: “from left to right, accelerating hard”.
4. 4Add the camera, the sound, and what must stay as it is.

xAI’s image-to-video runs at up to 1080p on 1.5, and its API can also pin a last frame, or up to 4 keyframes inside the clip. DreamBench offers neither, and its composer has no start or end slot: in a film, it draws each clip’s opening frame from your references and Grok animates it. See first and last frame.

xAI’s reference-to-video takes up to 14 images on 1.5, at up to 720p, to bring in people, products and places without fixing the first frame. Its advice for many references: give each one a single role and a fixed place in the scene, keep the cast small and keep props still. In a prompt:

• Describe each character in the same words every time, and say where they stand: “Sam stays on the left, Lena on the right, in both shots.”
• Count the props once and hold them: “one plate with exactly two croissants; nobody touches them”.
• Keep the clothes, the time of day and the light.

xAI’s storyboard recipe keeps a character by editing one keyframe into each new scene. DreamBench draws each character, place and prop once as a sheet, and each clip’s opening frame from the sheets it needs. See character consistency.

xAI shows two ways past one shot. Its café example puts two shots in one 15-second clip, “joined by a single hard cut, no cross-dissolve, no fade”. Its storyboard recipe plans four shots, each its own eight-second clip, and joins them. See multi-shot video.

• In one clip: say how many shots and how they join, describe the second shot from where the camera stands, and keep everyone on the same side of the frame.
• Across clips: one moment per clip, each with its own setting or angle, and the same words for every character.

On DreamBench, describe the film and pick Grok Imagine 1.5: DreamBench writes each shot as one continuous Grok clip of up to 12 seconds, draws its opening frame and cuts the clips together; redo any one with a note. On the Video page, a cut you write reaches Grok as written.

Paste these into DreamBench’s Video page or any tool that runs Grok Imagine 1.5. Each opens on its first frame.

Action: a mountain bike jump

8 seconds, 16:9. Low wide shot of a muddy forest trail, a rider in a red helmet cresting the rise. She launches off a jump, lands in a spray of mud and skids round the bend. The camera tracks low beside her. Overcast light. Sound: tires tearing through mud. No music.

Product: a perfume bottle

8 seconds, 1:1. Close shot of a frosted glass bottle labeled “SALT & FERN” on wet black slate. A slow, smooth push-in, ending with the label sharp. Light ripples across the glass as if through water. Only the camera moves; the bottle and its lettering never change. Nothing new appears. Sound: soft ambient music.

Dialogue: two speakers at a kitchen table

15 seconds, 16:9. Two shots joined by a single hard cut. Shot 1, a wide two-shot at a kitchen table in morning light: a man in a gray sweater slides a ticket across and says, grinning: “Two weeks in Lisbon. Say yes.” Hard cut to an over-the-shoulder shot from behind him, held to the end: she picks up the ticket, then, laughing, surprised: “You booked it already?” Continuity: the table, two mugs and the ticket match in both shots. Sound: a kettle, birdsong outside. No music.

A crowd: the night market

10 seconds, 16:9. High-angle wide shot of a night market in the rain, lanterns between stalls. When a gong sounds, hundreds of people with umbrellas pour into the lane and weave past each other without colliding. Slow time-lapse feel. Sound: rain on canvas, sizzling woks.

Time-lapse: a garden through a day

12 seconds, 16:9. A locked-off time-lapse of a cottage garden from dawn to night: shadows sweep across the lawn, sunflowers turn with the sun, the sky goes gold, then deep blue, and the windows light up. Clouds race overhead. The camera never moves. Sound: birdsong, then crickets.

Food: syrup on pancakes

6 seconds, 9:16. Extreme slow-motion macro shot: dark maple syrup pours onto a stack of pancakes, folds over itself and drips down the sides as a pat of butter slides. Warm window light, steam rising, shallow depth of field. Sound: a soft drip, a faint sizzle.

Animation: a clay hedgehog

10 seconds, 16:9. Stop-motion claymation, visible fingerprints in the clay, a slightly choppy frame rate. A small brown clay hedgehog in a green scarf rolls an apple across a mossy table, stops it with one paw and takes a proud bite. Continuity: every object stays exactly where it is; nothing appears or changes shape. Warm miniature-set light. Sound: a gentle music-box melody.

POV: a sledge run

8 seconds, 16:9. POV from a wooden sledge racing down a snowy forest slope, mittened hands on the rope at the bottom of the frame. It swerves between pines and bumps over a ridge, like handheld action-camera footage: shake from every bump, snow spraying past the lens. Sound: runners hissing, a whoop. No music.

Vertical social clip: a hook to camera

6 seconds, 9:16. Selfie framing on a busy station platform at rush hour, a young woman in a denim jacket holding the phone at arm’s length. She leans in and says, quick and conspiratorial: “Nobody tells you this about the night train.” A train pulls in behind her. Sound: platform announcements, brakes squealing.

A short line: the bakery counter

5 seconds, 4:3. Medium shot of a baker in a flour-dusted apron behind a counter of loaves, warm light from the window on the left. She holds out a paper bag and says, cheerfully: “Still warm. Mind the crust.” The camera is locked off. Sound: a shop bell, a till drawer.

Weather: a wave on the sea wall

8 seconds, 16:9. Low shot from behind a bollard on a stone harbor wall under a dark sky. A storm wave explodes over the wall and spray crashes across fish crates, then drains back. The camera stays locked off. Sound: booming surf, gulls. No music.

Two clips: the lighthouse keeper

Clip 1. 6 seconds, 16:9. Wide shot of an old lighthouse keeper with a white beard and a navy wool sweater climbing the last steps of an iron staircase, lantern in hand. The camera tilts up with him. Cold blue dawn light. Sound: boots on iron, wind.

Clip 2. 6 seconds, 16:9. Close shot of the old lighthouse keeper with a white beard and a navy wool sweater striking a match beside the great lamp; warm light floods his face. The camera is locked off. Sound: the match, the lamp’s hiss.

The two clips describe the keeper in the same words and keep the dawn light; the framing changes at the cut.

Most weak Grok prompts skip the first frame.

• One prompt run twice gives two clips.
• Every DreamBench clip has sound; xAI’s API can make a silent one, DreamBench can’t, so mute it in an editor.
• A real person or product can’t be known from words; it takes a picture you have the right to use.
• One generation stops at 15 seconds, and on DreamBench’s Video page a prompt past 2,500 characters is refused.

• One clip: on the Video page, pick Grok Imagine 1.5, paste your prompt, choose 4 to 15 seconds, the frame and 720p or 1080p. It goes to xAI’s grok-imagine-video-1.5 as written.
• A prompt: the AI video prompt generator writes one for any model in shot-list order; it has no Grok preset and writes the prompt only.
• A film: describe it, add up to 8 reference pictures and pick Grok Imagine 1.5; DreamBench writes, draws and cuts every shot, and you can redo any one.
• Not offered: 1.5 Lite, reference images or preset voices, first, last or key frames, silent clips, or 21:9.

Grok Imagine 1.5 is on every plan. The composer shows the price in credits before you generate; the first time, create a free account and choose a plan. For another model, see the MiniMax H3, Veo 3.1 or Kling 3.0 prompt guide; for prompts from words alone, text to video.