Start with a reference image. Describe what should move.
Use a photo, illustration or AI-created image as a visual reference for a short video. Describe the movement you want rather than rewriting every detail in the picture. Start with one image and one clear action, then inspect how the result differs from your reference.
Start with a promptPick an image with a visible main subject and a composition that suits your intended shot. Avoid distracting overlays and use material you have permission to upload. With one reference, the image determines the video’s aspect ratio.
Choose subject motion, environmental movement or a camera move. A walking subject, swaying curtains and a forward camera push are different instructions. Begin with one main change so you can judge the result.
Mention the appearance or layout you want to retain, then check it throughout the clip. These are written requests, not locked regions or a fidelity guarantee. More reference images do not automatically improve consistency.
For a room image, decide whether the camera moves through the space or a detail such as a curtain moves within it. Ask for the furniture arrangement and room layout to remain stable. Review straight lines, object positions and newly revealed areas; a camera move can introduce details not visible in the original image.
Interactions between animals can involve overlapping bodies and fast changes in pose. Start with a simpler movement before attempting several simultaneous actions. Review limbs, contact and each subject’s appearance from beginning to end; an attractive frame alone does not establish consistency.
For a character like a kitten knight, specify a small gesture or a clear direction of travel. Check the face, costume and props across the whole result. If important details drift, simplify the action and camera before trying again. A reference guides appearance; it does not lock identity or guarantee an exact match.
“Make this room move” leaves the subject and camera unclear. Explain which part should move, what should remain stable and how the scene is viewed. The room example below teaches a prompt structure; it is not an original generation record for a displayed asset.
Start with a promptStart by identifying the visual source: “Use the uploaded room image as the visual reference.” Let the picture guide the scene’s appearance. Add only the details needed to explain your intention; describing a completely different room can conflict with the reference.
Choose one specific change: “Let the sheer curtains sway gently in a light breeze.” Name the moving element, direction and pace. A subtle action is a useful first experiment, not a guarantee of better results. Avoid combining several character actions and camera shifts in your first attempt.
Explain what you want to retain: “Keep the furniture arrangement and room layout stable.” This helps express your priorities, but does not create a mask or lock objects in place. Check those elements in the result instead of assuming that written preservation requests were followed.
Distinguish camera motion from scene motion: “The camera stays fixed in a wide view, with the existing warm sunlight.” A fixed view lets you assess the curtains separately. If you choose a push or orbit, review newly revealed areas as well as the original composition.
Complete teaching prompt: “Use the uploaded room image as the visual reference. Let the sheer curtains sway gently in a light breeze. Keep the furniture arrangement and room layout stable. The camera stays fixed in a wide view, with the existing warm sunlight and a continuous shot.” Compare the result with the source. If the room shifts, simplify the movement or clarify the fixed camera; change one detail per attempt.
These static concepts illustrate motion directions to explore, not playable video results or verified input-output records. Prepare a clean source image rather than uploading a teaching graphic with captions or an inset reference.
For a shopfront reference, describe a slow forward camera move while the building remains still. Review window frames, flowers and reflections as the viewpoint changes. Keep background activity minimal, and check that the movement does not crop out the intended subject.
For a room reference, try gently swaying curtains with a fixed camera. Ask for the furniture and room layout to stay stable. Compare those details with the original image; if the entire scene moves, clarify that the curtains—not the camera—carry the action.
For a drink-can reference, choose a focused water or ice movement rather than several competing effects. Describe which product features should remain recognizable. Check the can’s shape, surface details and any lettering before commercial use; image guidance does not guarantee packaging accuracy.
Start your idea here, then continue in the workspace to sign in, confirm the reference and settings, and submit. On mobile, add your reference images and choose your settings in the workspace.
Choose a clear JPEG, PNG or WebP image, up to 20 MiB. Prepare the composition before upload: a single reference determines the video’s ratio. If you need a new starting visual, create one with the AI Image Generator.
Write what should move, what should stay stable and how the camera behaves. In the workspace, select 3–12 seconds and an output setting, then review the credit cost. See the AI Video Generator for the broader input and specification overview.
Watch the full clip and compare key details with the original reference, including the first and last frames. If needed, simplify one motion instruction before generating again. Download the completed MP4 within 72 hours of completion and save it locally.
Use JPEG, PNG or WebP images, up to 20 MiB each: at most seven references at 480p or 720p, or one at 1080p. A written prompt is required when submitting in the workspace. Choose an image with a clear main subject and use material you have permission to upload. The shared generator also works without an image; for that approach, explore AI Text to Video.
With one image, the reference determines the aspect ratio, so prepare its composition before uploading. At 480p or 720p, with two to seven references you can choose a supported ratio and explain what each image should guide. At 1080p, only one reference image is supported. Multiple references are not dedicated first-frame and last-frame controls, a slideshow, or a guarantee of consistent identity. Start with one clear reference when that is enough for your idea.
You can describe the movement you want and the appearance or layout you would like to retain. The reference and prompt guide generation, but exact preservation is not guaranteed. There are no masking, locked-region or motion-strength controls in this workflow. If details drift, try a simpler action or fixed camera and compare again. Higher output resolution does not necessarily fix changed shapes, labels or faces.
Select 3–12 seconds in whole-second increments and 480p, 720p or 1080p output settings. Grok Imagine 1.5 generates at the selected setting without a separate upscaling step. Text-only and multiple-reference inputs support 1:1, 16:9, 9:16, 3:2 and 2:3, with no automatic ratio option; a single image determines its own ratio. The default is 5 seconds, 480p and 16:9 when no single image determines the ratio. New accounts receive 50 credits valid for 30 days, enough if unused for one default 5-second 480p video for 40 credits, with 10 credits remaining. At 480p, 3, 5 and 12 seconds cost 20, 40 and 90 credits. Cost depends on duration and output setting, not reference count; confirm it in the workspace. This is not a daily or unlimited free allowance.
Compare the whole clip with the original reference, especially important product details, faces, straight lines and object positions. Check the first and last frames as well as the movement between them. Clarify or simplify one instruction if you try again. Completed videos are available as MP4 downloads in the workspace for 72 hours after completion. Keep a local copy before expiry; generation time is not guaranteed.
Free and paid generated outputs may be used commercially under our Terms of Service. You remain responsible for rights and permissions covering your references and intended use; this does not guarantee exclusive copyright. Uploads and outputs are private by default, and prompts, uploads and outputs are not used to train AI models under our platform policy. Read the Privacy Policy for details.