A useful camera prompt tells the model where the camera starts, how it travels, and what the viewer sees when it stops. Naming a “cinematic shot” is not enough. Write the move as a physical path, keep it separate from subject motion, and test one change at a time.
Open the right workspace first
Use a route that matches the source material you already have:
- Open C Dance AI to keep image and video work in the same workspace.
- Open Seedance 2.5 image-to-video when an approved image should define the first frame.
- Open Seedance 2.5 text-to-video when the model should establish the composition from the prompt.
- Visit the Seedance 2.5 model page before starting if you need the current generator entry point.
Image-to-video is the easier place to study camera motion. The source image already establishes the room, subject, lighting, and composition, so the text can concentrate on movement. Text-to-video asks the model to solve all of those jobs together.
The camera prompt formula
Build the instruction from six parts:
[starting frame] + [primary camera move] + [direction and speed]
+ [subject action] + [reveal or visual purpose] + [ending frame]
Here is the formula filled in:
Medium-wide eye-level shot of a violinist alone on a rehearsal stage. The camera makes a slow dolly-in along a straight path while she raises the bow and begins the first note. The movement reveals the worn texture of the instrument and ends in a steady chest-up frame. One continuous take, smooth controlled motion.
Each clause has a job. The opening clause fixes composition. The second names a physical camera path. The subject clause prevents the performer from becoming a statue. The reveal explains what the move is for, and the ending clause gives the shot somewhere specific to land.
Do not add every cinematic term you know. “Dolly-in, zoom, orbit, handheld, shallow focus, drone shot” describes several incompatible rigs. A shorter prompt with one readable path is easier to generate and much easier to repair.
Separate movement, angle, framing, and lens
These four controls are often mixed together, but they describe different properties.
| Control | What it changes | Examples |
|---|---|---|
| Camera movement | The camera's position or orientation over time | pan, tilt, dolly, track, crane, orbit |
| Camera angle | Where the camera looks from | eye level, low angle, overhead, bird's-eye |
| Framing | How much of the subject is visible | wide shot, medium shot, close-up |
| Lens or focus | Field of view and focus behavior | wide-angle, telephoto, shallow depth of field, rack focus |
A low-angle close-up is not a camera move. Neither is a wide-angle lens. You can combine them with a move, but keep the syntax explicit:
Low-angle medium shot with a wide field of view. The camera tracks left at walking speed, parallel to the cyclist. The distance between camera and cyclist remains constant.
That wording gives the reviewer three separate checkpoints: the low camera position, the medium framing, and the lateral path. If the result pushes forward instead, you know the path failed rather than the angle.
The diagram is an original editorial illustration, not output evidence from a named video model. It keeps the subject and set fixed so the path differences remain easy to compare.
Twelve camera moves and when to use them
1. Locked or static camera
The camera stays in one position and keeps the same framing. Use it for dialogue, product demonstrations, precise body movement, or any test where camera motion would hide another failure.
Locked camera on a tripod for the entire shot. Medium shot of a ceramic artist turning a bowl by hand. Only the hands, wheel, loose hair, and apron fabric move. The frame, horizon, focal length, and distance remain constant.
Describe what does move inside the frame. A static instruction with no subject or environmental action may still produce a vague floating motion because video models are built to create change over time.
2. Pan left or right
A pan rotates the camera horizontally from one fixed position. Use it to reveal a second subject, follow lateral action from a distance, or move the viewer's attention across a wide environment.
The shot begins on an empty train platform. From a fixed tripod position, the camera pans slowly right as the arriving train enters frame. The move ends with the conductor centered beside the open door. No forward camera travel.
The final subject gives the pan a stopping point. Without one, the camera may keep rotating or stop before the intended reveal.
3. Tilt up or down
A tilt rotates the camera vertically without changing its position. It works for height reveals, a transition from an object to a face, or a view that moves from the ground to the skyline.
Start on the climber's chalked hands gripping the rock. The camera tilts upward from the same fixed position, following the rope to her focused face, then stops in a close chest-up frame. Slow, steady motion.
Use “tilt” when the camera pivots. Use “rise” or “crane” when the entire camera moves upward.
4. Dolly-in or push-in
A dolly-in moves the camera closer. Foreground objects shift against the background, and perspective becomes more intimate. Use it to direct attention, increase tension, or move from context toward a detail.
Wide shot of a researcher alone in a dark archive. The camera makes a slow straight dolly-in from the doorway, passing between two shelves and ending two meters from the desk in a medium shot. The researcher looks up only during the final second.
Name the clear corridor for the camera. If foreground geometry blocks the path, the model may bend shelves, pass through objects, or substitute a zoom.
5. Dolly-out or pull-back
A dolly-out moves away from the subject and expands the visible world. Use it for scale, isolation, a reveal behind the camera's starting area, or an ending that releases tension.
Begin in a close-up of a chess player's eyes. The camera slowly pulls straight back while keeping her centered, revealing the full board, her silent opponent, and the empty tournament hall. End in a symmetrical wide shot.
List reveals in the order they should enter the frame. That sequence gives the move a simple temporal structure.
6. Lateral tracking or truck shot
The whole camera travels sideways. Unlike a pan, a tracking shot creates parallax as foreground and background layers slide at different speeds. Use it for walking, cycling, running, or showing a long row of objects.
Side-on medium-wide shot. The camera tracks right at the same speed as a courier riding through a covered market. Keep a constant four-meter distance and constant chest-height camera position. Foreground pillars pass close to camera while the distant stalls move more slowly.
The constant-distance clause stops a lateral move from turning into a push-in. Foreground landmarks make the intended parallax visible.
7. Follow shot
A follow shot travels with the subject from behind or at a three-quarter angle. Use it to place the viewer inside an action rather than observing from the side.
The camera follows two steps behind the hiker at shoulder height as she walks along the narrow ridge. It maintains the same distance and three-quarter rear framing. Her backpack moves naturally; the valley opens gradually on camera left.
Keep the relationship between camera and subject stable. If the subject also turns, accelerates, and crosses obstacles, test those actions after the basic follow path works.
8. Orbit or arc shot
An orbit moves along a curved path around the subject. It changes the visible background and exposes new sides of the subject. Use it for a hero reveal, a relationship between two people, or a product with important depth.
The camera travels through a slow 60-degree clockwise arc around the standing sculptor, keeping a three-meter radius and eye-level height. She remains still and centered. The move begins at a front three-quarter view and ends at a side profile with the unfinished statue visible behind her.
Start with 45 to 90 degrees, not a full revolution. Large arcs demand unseen geometry from both the subject and the environment. If identity matters, pair the shot with the character consistency workflow.
9. Crane or vertical rise
A crane shot moves the camera vertically and may add a gentle arc. Use it to change scale, reveal a crowd, or connect a ground-level detail to a larger setting.
Begin at waist height behind a street drummer. The camera rises smoothly to six meters while tilting slightly downward, revealing the surrounding dancers and the full circular plaza. End in a stable high wide shot.
State both the change in height and the new viewing angle. A rise without a tilt instruction can leave the camera looking horizontally into empty space.
10. Aerial or drone movement
An aerial path implies height and broad environmental coverage. Use it for geography, scale, route reveals, or transitions between locations.
Low aerial shot moving forward above a narrow river at sunrise. The camera gradually climbs above the tree line, then reveals the mountain village in the final wide frame. Smooth flight, constant forward direction, no sudden banking.
Give the movement a route rather than writing only “epic drone shot.” Rivers, roads, rooflines, and coastlines are useful paths because the scene provides a visible direction.
11. Handheld movement
Handheld describes a motion texture more than a path. It adds small human corrections and can make documentary, action, or backstage footage feel immediate. Too much shake can hide faces and object details.
Handheld medium shot following a cook through a busy kitchen, with restrained shoulder-level sway and small natural operator corrections. The camera stays behind her right side and does not overtake her. Readable face and stable horizon between steps.
Specify restrained, moderate, or aggressive movement. “Handheld” alone leaves the amount of shake open.
12. Zoom and dolly zoom
A zoom changes focal length while the camera stays still. A dolly zoom combines physical camera travel with an opposite zoom so the subject stays similar in size while background perspective changes. The second move is harder because two controls must stay synchronized.
Begin in a medium close-up of the guard at the end of a long corridor. The camera dollies backward while the lens zooms in at the matching rate, keeping the guard the same size in frame as the corridor appears to stretch. One continuous centered shot, slow and controlled.
Test an ordinary dolly and an ordinary zoom first. If both controls are already unreliable in the scene, combining them will not make the result easier to diagnose.
Put the ending frame in the prompt
Camera prompts often describe the start and the journey but omit the destination. A final-frame clause reduces ambiguity:
Starts: overhead wide shot of the workshop table.
Move: the camera descends in a slow vertical crane move while rotating 30 degrees clockwise.
Subject: the watchmaker places one gear into the open mechanism.
Ends: eye-level close-up with the gear centered and both hands visible at the frame edges.
The end does not need to be a separate reference image, but it needs a visible composition. If the exact endpoint matters for an edit, use the first-and-last-frame prompt workflow and approve both endpoints before generation.
Control speed, easing, and timing
“Slow” is useful, but it is still relative. Add a temporal cue when the move has an editorial beat:
- steady speed for the full shot;
- slow start, then a gentle acceleration;
- complete the push during the first three seconds, then hold;
- remain locked until the subject turns, then pan to follow;
- stop moving before the final action.
Write the sequence in chronological order. Avoid asking for several overlapping timing systems. A clean instruction such as “hold for one second, push forward for three seconds, then hold the final close-up” gives the model a readable progression.
The short motion graphic below is an owned editorial diagram. It animates the planning board to show why paths should be reviewed as separate choices; it is not a benchmark or a generated-camera demonstration.
Run a controlled camera test
Use the same starting image for every attempt. Copy the subject and scene clauses without changing them. Change only the camera block.
| Test | Camera block | What it isolates |
|---|---|---|
| A | Locked eye-level camera | Baseline subject and scene motion |
| B | Slow straight dolly-in | Forward camera travel and perspective |
| C | Slow lens zoom-in from fixed position | Focal-length change without travel |
| D | Slow 60-degree clockwise orbit | Curved path and newly exposed geometry |
Compare the frame at the start, midpoint, and end. Check subject scale, horizon, path direction, reveal order, and the final composition. Do not choose a winner based only on atmosphere. The test is meant to show which physical instruction the model followed.
Save the successful prompt beside the input frame and output file. A plain naming scheme is enough:
market-walk-source-v01.png
market-walk-locked-v01.mp4
market-walk-dolly-in-v02.mp4
market-walk-track-right-v03.mp4
market-walk-camera-notes.md
For longer sequences, plan each shot separately with the multi-shot storyboard guide. Camera continuity becomes easier to inspect when each clip has one purpose.
Diagnose camera prompt failures
The prompt says pan, but the camera moves sideways
The model interpreted the intent as tracking. Reinforce the fixed pivot: “camera rotates right from one stationary tripod position.” Remove moving-subject language during the test and add a clear object for the pan to land on.
The prompt says dolly-in, but the result looks like a zoom
Add foreground objects and ask for visible parallax. State that the camera travels physically forward along a straight path. A blank background offers little evidence that the camera changed position.
The subject changes size during a tracking shot
Specify constant distance, matching speed, and constant focal length. Simplify subject acceleration. If the camera catches up or falls behind, the apparent size will change even when the lateral direction is correct.
An orbit bends the face, body, or product
Reduce the arc and subject movement. Use a source image or reference pack that defines the side the camera will reveal. A single frontal image does not contain complete information about a profile, a backpack, or the back of a product.
The camera never stops
Name the final composition and add a hold: “the move ends in a stable chest-up frame and the camera remains locked for the final second.” A movement verb without a destination can behave like an open-ended instruction.
The result contains a cut
Ask for one continuous take and describe the reveal as a physical journey through the same space. Remove a second location or a sudden impossible viewpoint. Those cues may encourage the model to solve the prompt with an edit.
A static shot still drifts
Use a positive lock instruction, then give motion to the subject or environment. If slight drift remains, select the cleanest generation and stabilize it during editing. Do not make an entire prompt out of repeated negative phrases.
Common prompt-writing mistakes
The most expensive mistake is changing several variables after one weak result. A new lens, longer prompt, different source image, faster subject action, and extra camera move create a new experiment. They do not explain the old failure.
Also watch for these problems:
- Calling a frame size a movement, such as “camera moves into a close-up,” without saying whether it dollies or zooms.
- Giving the camera no physical route through a scene with obstacles.
- Combining subject rotation with a large orbit before either action works alone.
- Describing mood instead of motion: “dynamic cinematic energy” has no direction, speed, or endpoint.
- Using a reference image whose perspective contradicts the requested movement.
- Packing a short clip with several reveals that need separate shots.
The Seedance 2.5 prompt guide covers the wider prompt structure. Use this page as the camera layer inside that structure.
A practical shot card
Before generating, fill one card per shot:
Shot purpose: reveal the empty seats behind the speaker
Source: approved medium-wide stage image
Start frame: eye-level, speaker centered, first row visible
Camera move: slow dolly-out on a straight path
Subject motion: speaker lowers the microphone and stays still
Environment motion: none
Reveal order: front row, middle rows, full empty hall
End frame: symmetrical wide shot, speaker small but centered
Duration plan: move for four seconds, hold for the final second
Continuity locks: wardrobe, stage lighting, chair layout, horizon
Acceptance check: correct path, three reveals in order, stable identity, no cut
This card is short enough to review before every generation. It also records what the shot is supposed to communicate. A technically correct orbit that reveals nothing useful is still the wrong camera choice.
Start a camera movement test
Prepare one clear source image, then open Seedance 2.5 image-to-video. Generate a locked baseline first. Duplicate the prompt and change only the camera clause for the next attempt.
Keep the starting frame, prompts, and three review frames together. After one move passes, add timing or subject action in a separate test. That sequence turns camera prompting into a controlled shot-building process instead of a vocabulary guessing game.