An AI video prompt should describe what the camera records over time. Terms such as “cinematic” or “dynamic” do not define a shot. A usable prompt identifies the subject's blocking, camera position, camera movement, focal priority, timing, and end state.
The simplest rule is: one shot, one dominant subject action, one dominant camera move. Complexity can be added after the basic motion is stable.
Distinguish subject motion from camera motion
These are separate instructions:
Subject motion: the cyclist starts still, looks left, then pushes away from the curb.
Camera motion: a slow lateral track keeps pace from the opposite side of the street.If you only say “the cyclist moves dynamically,” the model may move the person, the camera, the background, or all three. Separating motion layers reduces unintended zooms and drifting viewpoints.
Essential camera movements
Locked-off shot
The camera does not move. Use it for precise blocking, product demonstrations, visual effects plates, or scenes where the environment provides the motion.
Locked-off waist-high camera. A red umbrella rolls from left to right across the
empty platform. The frame and focal length remain unchanged for the whole shot.Pan and tilt
A pan rotates left or right from a fixed position; a tilt rotates up or down. Name the subject the camera follows and where the move ends.
Slow pan right following the runner from the tunnel entrance to the finish marker.
The camera stays at chest height and stops when the runner reaches the center.Dolly or push-in
The camera physically moves closer. It changes perspective, unlike a digital zoom.
Very slow dolly forward from a medium-wide shot to a medium close-up. Keep the
chef's eyes centered while the kitchen background gains gentle parallax.Pull-back or reveal
Move away to disclose context. State what is hidden at the beginning and visible at the end.
Begin close on a weathered brass compass. Pull back steadily to reveal it resting
in the hand of a sailor at the bow of a small boat in fog.Tracking shot
Move beside, ahead of, or behind a subject. Define distance and relative speed.
Low side-tracking shot, camera two meters from the skateboarder and matching speed.
The wheels remain sharp while storefront lights streak softly in the background.Orbit
The camera circles a subject. Short generations usually work better with a partial orbit than an unspecified 360-degree move.
Smooth 45-degree clockwise orbit around the seated musician, from frontal
three-quarter view to profile. Keep her face and guitar as the focus.Crane or pedestal move
Move vertically to change scale or reveal geography.
Rise slowly from eye level to three meters above the table, ending in a clean
overhead view of the map and all four players' hands.Handheld movement
Describe intensity and purpose. “Handheld” without limits can produce chaotic motion.
Restrained documentary handheld camera with subtle breathing motion, no sudden
rolls or whip pans. The operator walks backward as the speaker approaches.Write the shot in time beats
For a six-second clip, divide the action into three readable beats:
0–2 s: locked medium shot; the elevator doors remain closed.
2–4 s: doors open and the subject looks up.
4–6 s: slow push-in as warm light reaches the subject's face; end on a still gaze.The beats should connect. Do not demand five locations, three costumes, and several cuts in a clip that has time for one clear action.
Add lens language only when it changes the story
Lens cues are useful when they describe perspective:
- wide viewpoint: stronger depth, more environment, visible parallax;
- normal viewpoint: balanced perspective and natural spatial relationships;
- telephoto feel: compressed background and isolated subject;
- macro: very close detail and shallow focus;
- deep focus: foreground and background both readable.
Avoid contradictory combinations such as “extreme wide-angle compression.” If exact optics are not supported by the selected model, describe the visual effect you want.
A complete shot template
[Duration and framing]
A six-second continuous medium-wide shot.
[Subject blocking]
An adult florist enters from frame left, crosses to the center table, and places
one yellow flower in a blue vase. She ends with both hands resting on the table.
[Camera]
Slow diagonal dolly from the doorway toward the table, moving from eye level to a
slightly lower angle. No cuts, zooms, orbit, or camera roll.
[Focus and environment]
Focus begins on the florist and shifts gently to the yellow flower in the final
second. Early morning shop light, soft rain visible through the window.
[End state]
Finish on a stable frame with the vase centered and the florist behind it.Troubleshooting unwanted motion
The camera moves too much
Remove competing movement terms. Use “locked-off,” “single continuous shot,” or “one slow move only,” and explicitly exclude zooms, rolls, and cuts.
The subject changes direction
Anchor movement to the frame: “enters from frame left, travels in a straight line, exits frame right.” Define the endpoint.
The shot loses the face or product
State the tracking relationship: “keep the subject's face unobstructed and within the central third throughout.” Reduce the speed or amount of orbit.
The ending feels accidental
Write a final composition, not just an action. First-and-last-frame generation can help when the ending position is critical.
Motion looks physically implausible
Simplify the action, specify contact points, and remove simultaneous camera effects. A hand picking up a cup needs a stable cup, a visible grip, and a clear destination.
Model-aware use
Camera syntax varies by provider. MiniMax's official documentation, for example, describes text-to-video, image-to-video, first-and-last-frame generation, and director-style camera instructions on supported models. Treat those documented modes as capabilities, then apply the shot-design framework above as your own production layer. See the official MiniMax video generation guide.
Final shot check
- Can the action be understood without reading the prompt?
- Is there one dominant camera move?
- Are start and end compositions defined?
- Does the movement fit the clip duration?
- Is subject motion separated from camera motion?
- Are focus, lighting, and sound supporting rather than competing?
- Does the final frame give the editor a usable cut point?
Plan the shot first, then create it in the Seedream AI Video Generator. For transition-heavy scenes, continue with the first-and-last-frame workflow.
