Grok Imagine Prompts: 14 Examples

Grok Imagine prompts for image-to-video should specify what moves and what stays fixed. The model already has the still. Use the prompt to direct movement, camera behavior and pace; keep descriptions of the subject for details that need preserving. These examples are starting points, not tested success-rate claims.

Below are two fill-in templates and 14 prompts for Grok Imagine on SynthPulse, where the workflow starts with a still image.

Where to use these prompts

On SynthPulse, Grok Imagine 1.5 is offered as an image-to-video model. You bring the still, and Grok animates it. This is a SynthPulse workflow limit, not a limit of Grok itself: xAI also documents text-to-video generation.

Start with a still image:

1. Make or upload the still. The Image studio has Nano Banana Pro (45 credits), Seedream 5.0 Pro (18 credits at 1K), Nano Banana (10 credits), and GPT Image 2 (15 credits). Check lettering in the result rather than assuming it will remain exact.
2. Take it to the Video studio, pick Grok Imagine 1.5, attach the still, and write a motion prompt.

Before uploading, inspect the crop. Leave room for the requested move: a push-in crops the edges, while a sideways move or orbit asks the model to invent surfaces behind the subject that your photograph never showed, including any lettering on the hidden side of a package.

If that unseen detail matters, choose a smaller movement or prepare a different starting view in the Image studio. Settle the composition before buying motion tests.

What a motion test costs

Grok Imagine 1.5, in credits:

• 480p / 4s: 24
• 480p / 8s: 48
• 720p / 4s: 45
• 720p / 8s: 90
• 720p / 12s: 135

Prices and available settings in this guide reflect the repository configuration as of 2026-10-23; check the current studio quote before generating. For comparison, the same 4 seconds at 720p costs 140 credits on Kling 3.0 and 410 on Seedance 2.0. At 24 credits a try, you can afford to be wrong ten times and still have spent less than one Seedance 2.0 clip.

Grok Imagine 1.5 tops out at 720p here; 1080p is not one of the options in this studio, although xAI supports it in its own API.

Grok Imagine prompts: four things to specify

Give the model four pieces of direction:

1. What moves: the specific thing in the frame that animates.
2. How the camera behaves: including "does not move," which is a real and frequently correct answer.
3. Pace: slow, steady, gentle, brisk. Choose a pace the action can finish at within your selected duration.
4. What must not change: the anchor you want preserved. Inspect that detail in the result.

The still supplies the subject. Keep extra description focused on details you need preserved, rather than asking the model to redesign the scene.

Grok Imagine templates to copy and fill in

Replace the brackets with details from your still. Pick one camera behavior; keep the action short enough for a single clip.

Still camera

[Visible element] [one slow, specific movement] for the full clip. Camera stays completely still. Keep [subject or composition anchor] and [lighting or background] unchanged.

Moving camera

[Visible element] [one slow, specific movement]. Camera [one gentle movement] throughout the clip. Keep [subject or composition anchor] in frame and preserve [lighting or visual style].

For example, with a still of a steaming cup on a table: "Steam rises slowly from the cup throughout the clip. Camera stays completely still. Keep the cup, table and warm lighting unchanged."

14 Grok Imagine prompts to adapt

Portraits

Her hair and the collar of her jacket move slightly in the wind. She blinks once and her gaze stays on camera. Camera is completely still. Background does not change.

He turns his head slowly to the left, stopping in profile. Nothing else in the frame moves. Locked-off camera, no zoom.

Subtle breathing motion in the shoulders, one slow blink. Camera pushes in very slightly over the full clip. Expression unchanged throughout.

Product

The bottle stays exactly where it is. Camera orbits slowly to the right around it, about 30 degrees total, staying level. Lighting and shadows stay consistent.

Steam rises steadily from the cup. The cup does not move. Camera holds still. No other motion in frame.

Camera pushes in slowly toward the watch face. The watch stays still and in focus. Background falls further out of focus as the camera moves.

Food

Sauce drips slowly off the edge of the spoon. Everything else on the plate is still. Camera holds a static overhead shot.

Gentle rising steam from the bowl, slight shimmer on the broth surface. No camera movement. Nothing else changes.

Landscape and weather

Clouds drift slowly left to right. The water surface ripples gently. The shoreline and rocks stay completely fixed. Camera is locked off.

Camera rises slowly and steadily. The treeline stays sharp. No tilt, no rotation, continuous upward movement for the whole clip.

Snow falls lightly and evenly through the frame. The cabin and the trees do not move at all. Static camera.

Character and atmosphere

The candle flame flickers. Shadows on the wall shift with it. The figure at the table stays still. Camera does not move.

Neon sign flickers once, halfway through. Rain falls steadily. The street and the parked car stay fixed. Locked-off camera.

Dust motes drift slowly through the shaft of light. Everything solid in the frame is motionless. Camera holds completely still.

Each example names the moving element, camera and anchor. Three sentences are a useful starting point; add detail only to resolve a specific ambiguity.

Five problems to check in a motion prompt

You re-described the subject. "A red vintage car on a coastal road at sunset, cinematic, driving fast" mixes a new scene description with an action request. Say "the car pulls away from camera, everything else holds still" instead.

You asked for a cut. "Then it cuts to a wide shot" asks for an edit within the generation and may produce an unwanted transition. One continuous action per clip; assemble cuts in an editor.

You left the camera unspecified. Without a camera instruction, the result may include drift you did not ask for. "Camera does not move" is a legitimate and often ideal instruction.

Your still had small text in it. Lettering can warp under motion. Keep text large and near the center, or keep it out of the animated region entirely.

You asked for too much at once. Start with one moving element. If several actions compete, split the shot into separate clips or simplify the movement before adding detail.

Plan the sound separately

This SynthPulse option does not expose an audio-generation control, so check the actual clip and plan to add sound in your edit if needed. A track from the Song studio draws from the same credit balance. For shots that need sound, Google documents native audio for Veo 3.1. SynthPulse does not disable audio in its Veo request, but check the finished clip before relying on it; audio is not guaranteed for every generation.

Budget for rejected takes

At 24 credits per 480p attempt, the sensible loop is:

1. Settle the still first. Composition mistakes are far cheaper to fix in the Image studio.
2. Run the motion prompt at 480p until the movement is right. Budget three or four tries: that is under 100 credits.
3. Re-run the exact same prompt at 720p for the final. 45 credits.
4. Everything lands in your Library.

For example, a 15-credit still, four 24-credit motion tests and one 45-credit final attempt total 156 credits. That is a planned budget, not a guarantee of a finished clip; each new attempt may vary.

A prompt generator worksheet

You can build a Grok prompt generator from a simple worksheet. Write four short answers before joining them into sentences: which visible element moves, what it does, how the camera behaves, and what remains unchanged. Keep the result to one continuous shot. Save the sequence planning for your edit.

For a product photograph, fill the worksheet this way: moving element, steam; action, rise slowly; camera, locked overhead; anchors, cup position and printed logo. The resulting prompt is: “Steam rises slowly from the cup. Hold a locked overhead view. Preserve the cup position, printed logo and existing light.” If the logo warps, remove camera motion before adding more descriptive words. Check the label again.

Save the image, exact prompt, model, resolution and duration together. Without those settings, a promising result is difficult to compare with the next attempt. A worksheet also helps you decide when to stop: once the requested motion reads clearly and the important anchors survive, more adjectives are unlikely to solve a different production problem.

Grok Imagine tags: style, camera and motion cheat sheet

Treat Grok Imagine tags as ordinary descriptive language, not secret commands or a guaranteed syntax. Use the table to describe a shot in plain language. Choose a small compatible set and remove words that contradict the source image.

For a flat illustration, “preserve flat colors” is more coherent than “photorealistic cinematic render.” For a photograph, a camera phrase can describe movement without asking for a different lens or lighting setup. Words such as “orbit” also request unseen surfaces; begin with a smaller move if the model invents the back of a product.

Templates for three repeatable jobs

These are copyable starting points. Replace the brackets and test them on a still you have permission to use. Start with one that matches your image.

Product reveal: keep the label readable

Keep [product] stationary. Move the camera very slowly toward [detail] while keeping the full label in frame. Preserve the existing lettering, silhouette and light direction. No rotation or scene change.

Use a small approach to bring attention to a product detail. If the lettering changes, a static shot with only environmental movement is a more useful next test than a larger camera move.

Illustrated character: preserve the drawing

[Scarf or hair] sways gently once. The character keeps the same pose and expression. Locked-off camera. Preserve the original linework, flat colors and background. No new objects.

For an illustration, begin with one secondary element rather than a full body turn. A turn requires the model to invent views that are absent from the still, making identity preservation harder to assess.

Landscape: keep the horizon stable

[Clouds or mist] drifts slowly toward [direction]. Keep the horizon, buildings and foreground fixed. Static wide view. Preserve the existing time of day and color palette throughout the clip.

Let the weather supply the movement while the camera stays still. Add a camera move only after checking whether the terrain itself remains stable.

Grok image editing prompt examples: fix the still first

An image-editing prompt changes the still before you animate it. On SynthPulse, do that preparatory work in the Image studio with a compatible image-to-image model. Grok Imagine 1.5 in the Video studio produces a clip; it is not a still-image editor. Use the following examples at the image stage.

Simplify a product background

Keep the bottle shape, cap and label layout from the reference. Replace the busy background with a plain warm-grey backdrop. Keep a soft contact shadow. Do not add lettering, props or another bottle.

Change a jacket color

Change only the jacket fabric from blue to muted burgundy. Preserve the face, pose, seams, background and light direction. Keep the original framing.

Prepare an illustration for animation

Preserve the character and original drawing style. Remove the lettering behind the head and extend the plain background around the hair. Keep the expression and pose unchanged.

Inspect edits before animating: a generative edit can change details even when told to preserve them. Use a conventional editor for changes that must preserve pixels exactly. Then write motion instructions for the accepted still, rather than asking the video model to repair the image and animate it simultaneously.

Frequently asked questions

Can I reuse Grok templates across different images?

Yes. Keep the structure and replace the image-specific details: a scarf can sway, but a cup needs a different action, and a full camera orbit around a flat drawing requires the model to invent a view the artist never drew. Check the source before submitting.

Are “Grok prompts” and “imagine prompts” the same task?

Both are broader terms. Grok prompts can mean chat requests; imagine prompts can mean instructions for a new image. This guide covers motion paired with a still. Need a different composition? Make that image first.

What is Grok Imagine agent mode, or a Grok Imagine agent?

SynthPulse does not offer that mode. Its Grok video option accepts a still and a motion instruction, without an agent to plan shots or run repeated generations. This guide does not define agent features in xAI's own interface. Plan each shot manually and review the quote before submitting.

What does Grok remix mean here?

There is no Grok remix control in this workflow. To make a variation, change the source or motion instruction and submit a new generation. Change one variable at a time so you can see what helped; a new attempt can also vary on its own.

Can a Grok prompt generator run this automatically?

The worksheet above is manual. Fill in an action, camera behavior and two anchors, check the sentence against your picture, then submit it with the settings you selected. Each generation uses its displayed credits.

Should I reuse the same prompt at a higher resolution?

Reuse it to keep the setup comparable, then review the new take from start to finish. Higher resolution does not reproduce an earlier generation exactly. Pause on the opening, middle and final frames to check identity, product shape, lettering and background drift.

---

More on the model: Grok Imagine on SynthPulse. Start with the still in the Image studio, or go straight to the Video studio.