Quick answer
A usable AI video camera prompt names one move, its direction, its speed, and what must stay stable. For MiniMax H3, the current official guide shows bracketed camera instructions such as [pan], [zoom], and [static] after a key scene description. Natural phrases such as "slow camera push forward," "track beside the cyclist," or "orbit clockwise around the product" can also be written as prompt phrases, but MiniMax does not publish them as guaranteed H3 controllers. Treat them as instructions to test, not buttons with deterministic outcomes. Start with a static shot, slow push-in, pan, or tilt. Wide orbits, tracking shots, crane moves, and dolly zooms reveal more unseen geometry, which gives the model more scene structure to infer. Use this order: scene, subject action, one camera move, direction and speed, subject lock, then the ending frame. This guide separates verified H3 evidence from general cinematography vocabulary.
Last checked: August 6, 2026.
Camera prompt formula
Use this order:
[Scene and subject]. [Subject action]. Camera: [one movement], [direction], [speed]. Keep [identity, shape, horizon, or framing] stable. End on [final composition].
For the official bracket style, place a supported camera cue after the key description:
A lone hiker crosses a pale salt flat at dawn
[pan]. The camera pans slowly from right to left while the hiker keeps the same clothing and body proportions. End on a wide profile frame.
The MiniMax guide gives [pan], [zoom], and [static] as examples. It does not publish a complete bracket vocabulary or promise that every natural-language film term will produce the same path on every run.
| Movement | Direction | Speed | Subject lock | Risk |
|---|---|---|---|---|
| Static | Fixed | None | Hold angle, distance, and composition | Lower |
| Pan | Left or right | Slow or gentle | Keep subject scale and horizon stable | Low to medium |
| Tilt | Up or down | Slow | Keep vertical lines straight | Low to medium |
| Push-in | Toward subject | Slow | Keep face or product shape stable | Medium |
| Pull-out | Away from subject | Slow | Preserve the starting subject while revealing context | Medium |
| Tracking | With or beside subject | Steady | Keep identity, stride, and distance consistent | Medium to high |
| Orbit | Clockwise or counterclockwise | Small and slow | Preserve unseen sides and surface details | Higher |
| Crane | Up or down through space | Controlled | Hold horizon and scene geometry | Higher |
Pan, tilt, push-in, and pull-out
A pan rotates the camera left or right from a fixed position. Use it to scan a wide setting or follow a subject without changing camera distance. A tilt rotates up or down from the same position. It fits a vertical reveal, such as moving from a building entrance to its roofline.
A push-in moves the camera closer. A pull-out moves it away and reveals more of the setting. These physical moves are different from a lens zoom, although everyday prompt libraries often mix the terms. If the distinction matters, name the physical move and describe the intended parallax or background reveal.
This minimaxh3.tv text-to-video example used one natural camera phrase:
A cinematic sunrise over a calm alpine lake, soft mist drifting above the water, slow camera push forward.
Dolly, tracking, orbit, and crane
A dolly move travels toward or away from the subject. A tracking shot travels with the subject, often from the side, front, or rear. Tracking adds coordination because the subject and camera both move.
An orbit or arc travels around the subject. It can reveal surfaces that were not visible in the opening view, so it asks the model to infer more geometry. Keep the arc small when identity or product shape matters. A crane move changes camera height while traveling through space. State the final view, such as "end on a high wide shot," so the move has a destination.
Do not stack all four into one short clip. If a result fails, a single move is much easier to revise.
Zoom, lens, and framing terms
Zoom changes the field of view without moving the camera. Dolly changes camera position and therefore changes parallax. A dolly zoom combines opposite camera and lens changes. It is a demanding prompt phrase and should be labeled as an experiment unless you have a verified result from the same workflow.
Shot size and angle tell the model how to compose the subject:
| Term | What it asks for | Useful lock |
|---|---|---|
| Wide shot | Subject plus substantial environment | Keep the subject fully visible |
| Medium shot | Subject from roughly waist or chest up | Keep hands and face in frame |
| Close-up | Face, object, or detail fills the frame | Preserve facial or product geometry |
| Low angle | Camera looks upward | Keep verticals and horizon coherent |
| High angle | Camera looks downward | Keep body proportions natural |
| Over-the-shoulder | Foreground shoulder frames another subject | Keep eyelines and identities consistent |
| Rack focus | Focus moves between depth planes | Keep camera position fixed |
Speed and direction modifiers
Use measurable direction words instead of "dynamic camera." Try "left to right," "toward the doorway," "rise from eye level to rooftop height," or "a small clockwise arc." For speed, "slow," "gentle," "steady," and "brief" are easier to interpret than several conflicting adjectives.
Timing can make a move easier to read:
- Hold the opening frame briefly, then begin a slow pan left.
- Track beside the subject at a constant distance for the middle of the shot.
- Ease to a stop and hold the final composition.
If the clip is busy, simplify subject motion before making the camera faster.
Stable subject constraints
Stability instructions should name the detail that matters. "Keep everything consistent" is vague. Use constraints such as:
- Keep the same face, hairstyle, clothing, and body proportions.
- Keep the product label, cap, silhouette, and material unchanged.
- Keep the horizon level and the vertical lines straight.
- Keep the subject centered with the full body inside the frame.
Avoid asking for a large orbit around a flat reference image while also demanding exact unseen details. The prompt can request both, but the opening frame may not contain enough information to support them.
The ladder is a planning aid, not a measured MiniMax H3 failure rate. Scene layout, subject motion, input references, and generation randomness also affect the result.
20 copyable prompts
These are prompt phrases and templates. Only the slow forward lake example above is paired with a traceable first-party H3 output.
1. Slow pan left across a landscape
A lone red cabin beside a frozen lake at blue hour. Smoke drifts lightly from the chimney. Camera: slow pan from right to left. Keep the horizon level and the cabin shape unchanged. End with the cabin on the right third of a wide frame.
2. Slow pan right through an interior
A quiet ceramics studio with shelves of unfinished bowls. Dust floats in warm window light. Camera: gentle pan from left to right. Keep shelf lines straight and object positions stable. End on the potter's wheel.
3. Tilt up for a vertical reveal
The entrance of a narrow glass tower in light rain. Camera: slow tilt upward from the doorway to the roofline. Keep the building edges straight and the camera position fixed. End on the illuminated top floor.
4. Tilt down to reveal a subject
Moonlit clouds above a mountain observatory. Camera: gentle tilt down from the sky to the telescope platform. Keep the horizon steady. End with the telescope centered.
5. Push in on a portrait
A violinist stands backstage under one soft work light and takes a quiet breath. Camera: slow push-in from a medium shot to a close-up. Keep the same face, hair, clothing, and eyeline. End before the chin leaves the frame.
6. Pull out to reveal context
A chef places one finished dish on a stainless-steel counter. Camera: slow pull-out from the plate to reveal the full kitchen. Keep the dish shape and garnish unchanged. End on a balanced wide shot.
7. Dolly in toward a product
An unbranded matte-black watch rests on a pale stone block. A narrow light moves across the bezel. Camera: smooth dolly in toward the watch. Keep the case, crown, strap, and proportions stable. End on a centered product close-up.
8. Dolly out from a detail
Start on the brush texture of an abstract painting. Camera: steady dolly out to reveal the full canvas and quiet gallery wall. Keep the painted marks unchanged. End with the entire frame visible.
9. Rear tracking shot
A cyclist rides along a wet city street before sunrise. Camera: stable rear tracking shot at a constant distance. Keep the rider, bicycle geometry, lane direction, and clothing consistent. End as the cyclist reaches the next intersection.
10. Side tracking shot
A runner moves along a concrete seawall in soft morning light. Camera: smooth side tracking shot from left to right, matching the runner's pace. Keep the full body in frame and the stride natural. End with the runner still centered.
11. Small clockwise orbit
An unbranded ceramic speaker sits on a wooden table. Camera: slow 20-degree clockwise orbit. Keep the grille pattern, controls, silhouette, and table edges stable. End at a slight three-quarter angle.
12. Small counterclockwise arc
A weathered stone statue stands in a museum alcove. Camera: gentle counterclockwise arc around the front quarter of the statue. Keep facial features and stone cracks unchanged. End on a three-quarter view.
13. Crane up over a market
A small night market opens under strings of warm lights. Camera: controlled crane up from eye level to a high wide view. Keep stall positions and the central walkway coherent. End with the market layout visible.
14. Crane down into a stage
An empty theater stage under a single spotlight. Camera: slow crane down from the lighting rig toward stage level. Keep the stage geometry and light cone stable. End facing the center mark.
15. Truck left across a desk
A tidy architect's desk holds a paper model, ruler, and pencil sketches. Camera: short truck move to the left. Keep object scale and desk edges consistent. End with the paper model centered.
16. Pedestal up beside a dancer
A dancer holds a still pose in a plain rehearsal room. Camera: slow pedestal move upward from waist height to eye level. Keep the camera distance, body proportions, and wall lines stable. End on a medium shot.
17. Slow zoom in
A lighthouse flashes across a dark sea. Camera remains fixed while the lens slowly zooms toward the lantern room
[zoom]. Keep the tower vertical and the coastline unchanged. End on a tight architectural frame.
18. Slow zoom out
Start on a hand-drawn island marked on an old map. Camera remains fixed while the lens slowly zooms out to reveal the full map. Keep all printed lines and coast shapes stable. End with the map centered.
19. Dolly zoom experiment
A person stands still at the end of a long empty corridor. Camera: experimental dolly zoom, moving backward while narrowing the field of view to keep the person the same size. Keep the face, body, and corridor lines stable. End on the same subject scale with compressed depth.
This is a high-complexity prompt phrase, not a documented guaranteed H3 control.
20. Locked-off shot
A cup of tea steams beside a rain-streaked window. Camera: locked-off static shot
[static]. Only steam and raindrops move. Keep the cup, table, framing, and background fixed for the full clip.
Movement risk and fix table
| Symptom | Likely cause | Suggested change | Evidence status |
|---|---|---|---|
| Subject shape changes during an orbit | The move reveals unseen surfaces | Reduce the arc, add a stronger reference, or switch to a push-in | General workflow guidance; not measured as an H3 rate |
| Background bends during a pan | Direction, subject motion, and scene motion compete | Slow the pan and simplify other motion | Prompt troubleshooting guidance |
| Face drifts during tracking | Camera and subject move while identity must remain fixed | Shorten the move and name the face, hair, clothing, and distance locks | Prompt troubleshooting guidance |
| Zoom feels like forward travel | Lens change and physical movement are mixed | Say whether the camera position stays fixed | Cinematography distinction |
| Dolly zoom does not read clearly | Opposite lens and camera changes are too complex | Test a simple dolly or zoom first, then add the combined move | Experimental prompt phrase |
| The camera ignores the move | The instruction is vague or buried | Put one camera cue near the key description; use an official bracket example where appropriate | Official guide supports [pan], [zoom], and [static] examples |
| Final framing cuts off the subject | No ending composition was specified | Add a final shot size and frame boundary | Prompt planning guidance |
Change one variable per retry. If you change camera path, subject action, lighting, and framing at the same time, you will not know which change fixed the result.
FAQ
Does MiniMax H3 support camera movement prompts?
The current MiniMax H3 video-generation guide shows camera instructions in prompts and gives [pan], [zoom], and [static] as examples for text-to-video. The guide does not publish an exhaustive controller list or guarantee repeatable paths.
Where should the camera instruction go?
The official guide says to place it directly after key descriptions. In longer natural-language prompts, keep the camera sentence close to the subject and action it affects.
Is a zoom the same as a dolly?
No. Zoom changes the lens field of view from a fixed camera position. A dolly moves the camera through space, which changes parallax.
Why does an orbit cause more distortion?
An orbit may reveal sides of a subject and background that were not visible in the opening view. That requires more scene inference. This is a general geometry constraint, not a published MiniMax H3 failure-rate claim.
Should I use several camera moves in one prompt?
Use one move first. Add another only when the sequence and timing are clear and you can verify each change.
Can I copy these prompts into MiniMax H3?
Yes. Replace the scene and subject details, then keep one move, one direction, one speed, a specific stability constraint, and a final composition. The natural-language terms are prompt phrases to test.
Create a camera-directed clip
Open the MiniMax H3 text-to-video generator, start with one of the lower-change moves, and compare the full result with your stated ending frame. For a broader prompt structure, use the MiniMax H3 prompt guide. The showcase provides current examples for visual reference.
Sources and methodology
- MiniMax API Docs: Video Generation, checked August 6, 2026. Used for the current H3 model context, text-to-video mode, and the documented
[pan],[zoom], and[static]examples. - minimaxh3.tv text-to-video generator, checked August 6, 2026. Used to verify the current product entry.
- MiniMax H3 prompt guide, checked August 6, 2026. Used to keep this camera-specific guide separate from the broader prompt page.
- First-party MiniMax H3 lake example, checked August 6, 2026. The prompt and output are shown together above. It is one example, not a benchmark.
The 20 templates use standard cinematography terms as prompt phrases. Except for the cited lake example and the bracket cues named by MiniMax, they were not run as new H3 tests for this article. No success rate is claimed. minimaxh3.tv is an independent product and is not affiliated with, authorized by, or operated by MiniMax.
