A reference image can explain appearance without explaining where a video starts, when an action changes, or where it ends. First and last frames, timed keyframes and loop images guide these different creative decisions in Grok Imagine Video 1.5. Explore these video-generation approaches on the model page.
How do first and last frames differ from references and timed keyframes?
Guide appearance: Clear reference images; their order is not a storyline
Establish the opening: A first frame before the action begins
Define the ending: A last frame with the desired closing composition
Guide a particular moment: A middle image with a time
Share an opening and ending composition: One loop image
Either endpoint can be used on its own. A loop image replaces separate first and last frames, while references and timed keyframes may remain. Ordinary reference images are not an ordered endpoint pair.
What to prepare before starting
This model generates video from images: provide a prompt and at least one image source. Choose 1–15 whole seconds, with a prompt of up to 5,000 characters. Give one image a clear purpose before adding more controls.
How to plan endpoints and timed keyframes
Step 1: Choose first and last frames
Prepare a closed package, a partly opened package and the revealed product. Keep product shape, lighting and camera angle compatible. Then arrange the opening, a middle milestone at three seconds, and the ending.
Step 2: Place a middle milestone and describe the action
Open the package slowly, use the partly opened composition around the third second, reveal the watch, and hold the final view with the dial visible. Keep the camera steady.
Example note This is an untested creative brief. Timing guides generation; it does not guarantee exact frame-level action.
Review continuity between milestones.
Use at most four middle keyframes. Decimal seconds are supported; times must fall strictly inside the clip and be at least one third of a second apart. Shortening a video means revisiting those times so a former closing milestone does not sit outside the clip.
Input-planning concept: top to bottom, opening, a partly open pose guided around 3 seconds, and the ending of an 8-second clip. Prepare separate input images; this is not a Grok result or evidence of timing accuracy.
Step 3: Choose 720p or 1080p for the required controls
When the shot needs an opening, ending and timed middle images together, prefer 720p. Use a single first-frame image at 1080p; combined first, last and timed middle frames are currently unavailable at that resolution.
If higher resolution matters more, describe motion from one starting image. If the scene depends on several compositions, keep a compatible combined-control plan. Changing resolution alone does not preserve the same controls or establish the same result.
Ordinary references allow up to seven images at 480p or 720p and one at 1080p. That reference count does not establish compatibility for every frame combination.
How does the plan change for a loop video?
Try a gentle sway, a lighting change or movement in fabric. Choose a shared opening and closing image, then describe a repeatable action:
Let the fabric sway gently and gradually return to its opening pose. Keep the camera fixed and the lighting steady.
Example note This is an untested loop concept. A shared endpoint image does not guarantee a seamless loop.
Prefer 720p for loops and begin without elaborate camera travel or large actions. Play the result repeatedly and inspect the join for changes in position, brightness, object size or motion direction.
Top to bottom: the opening pose, a sway to the other side, and a return toward the opening composition. Conceptual loop storyboard, not a tested Grok video or a verified seamless result.
After generation, check the action and loop seam
After generation completes, inspect hands, product details, occlusion and the final pose throughout the action. Keep materials, prompt, settings and the chosen video together. Change one major creative decision in the next attempt. For an ordered-image storyboard, explore Kling 4.0.
How should you revise timing or a visible loop seam?
Keyframe times no longer fit a shorter clip: Move every time strictly inside the new duration
The loop jumps at the join: Reduce movement, keep the camera fixed and compare endpoint position and brightness
A combined-control 1080p plan fails: Use 720p combined controls or rebuild the shot around one 1080p first frame
Reference images do not play in order: References guide appearance; use dedicated frames for endpoints and middle milestones
Choose the role of each image before the generation plan
Start with endpoint images when the opening and ending matter, and add timed keyframes only when a middle action needs guidance. For a repeating shot, build restrained movement around a loop image. Then choose resolution around the controls: prefer 720p for combinations and one first frame for 1080p.
