How to copy camera movement from a reference video

A generated scene cannot be assumed to reproduce a reference camera trajectory exactly. If the path, timing, blocking or legal continuity must match, keep the original footage as the motion source and change only the visual treatment.

Choose the workflow before writing a prompt

• Preserve the original camera and action: upload the clip to a video-to-video restyle workflow. Magic Hour documents this mode as changing the visual style while preserving the source motion and structure.
• Transfer a person's performance to another character: use Motion Control with a character image and reference performance. This transfers subject movement; the output background comes from the image, so it is not a camera-path-copy tool.
• Recreate a camera move around a new subject: use image-to-video or text-to-video and describe the camera path, speed, framing and end state. Treat the result as an approximation that requires review.

Step 1: isolate the camera move

Trim a reference to one continuous shot. Three to eight seconds is a useful starting range for diagnosis, but use the duration your tool supports. Avoid cuts, speed ramps, zooms combined with pans, or a subject whose movement hides what the camera is doing.

Write down four observations before generating:

• Path: forward, backward, left, right, up, down, around the subject, or irregular handheld drift.
• Framing: wide, medium, close-up, over-the-shoulder, centered, or intentionally off-center.
• Speed: constant, accelerating, decelerating, or easing into the stop.
• Depth: whether foreground and background objects move at visibly different rates and create parallax.

Separate camera motion from subject motion. A person walking toward a locked camera is not a dolly-in. A static subject growing larger while the background perspective changes is closer to a dolly-in. A digital zoom changes framing without creating the same parallax as a camera moving through space.

Step 2: prepare a reference the model can read

• Use one shot with no edit or transition.
• Keep the main subject visible and avoid heavy occlusion during the key move.
• Prefer a stable source unless handheld movement is the feature you want.
• Match the new scene's starting composition to the reference when possible.
• Use footage and performances you have permission to process.

When preserving the source camera path through video-to-video, choose the exact section first and keep the restyle prompt focused on appearance. When generating from a still, leave enough space around the subject for the requested move.

Step 3: write the camera instruction first

Use this compact structure: [camera path] + [speed] + [framing] + [subject action] + [end state]. Add lighting and style after the movement is unambiguous.

Dolly-in prompt

Workflow recipe

A cinematic, ultra-realistic lakeside garden path surrounded by blooming flowers (white daisies and soft pink roses) leading toward a calm blue lake with majestic mountains in the background. Warm sunlight shines through tree branches, creating soft sun rays and gentle lens flare. The camera is handheld at chest level, simulating a person slowly walking forward along the stone pathway, producing a natural subtle sway. Camera framing remains static (no panning or tilting), only slow forward movement with slight handheld motion. Flowers and leaves gently sway in a light breeze, and the lake surface has soft, subtle ripples. Sunlight flickers naturally through the leaves as the perspective advances slightly. Rich depth of field with foreground flowers shifting softly as the viewer moves forward. Seamless loop animation, smooth continuous motion with no visible cuts or jumps. Cinematic lighting, ultra-detailed, immersive atmosphere. Aspect ratio 16:9.

Model

veo3.1-lite

Format

9:16

Resolution

720p

Duration

8 seconds

Copy settingsTry in Text-to-Video

Side tracking prompt

Prompt: “Camera tracks left to right beside the cyclist at matching speed, medium side profile, background moves faster than the distant skyline, stable distance from the subject, one continuous shot.”

Orbit prompt

Prompt: “Camera makes a slow 45-degree clockwise orbit around the stationary chair, constant radius and height, three-quarter front view ending at a side view, smooth stop, room geometry remains stable.”

Handheld prompt

Prompt: “Subtle shoulder-height handheld drift while following the speaker, restrained vertical movement, subject remains in the center third, natural corrections rather than rapid shake, documentary feel.”

Crane-up prompt

Prompt: “Camera rises slowly from table height to above the subject while tilting down to maintain framing, smooth vertical path, increasing view of the room, gentle ease-out at the top.”

Step 4: make the first test cheap to diagnose

Generate a short version with one movement and one subject action. Keep the same input, model and settings for the first comparison. Change only one variable between attempts, such as speed or end framing.

Do not automatically generate three to five versions. Make one diagnostic attempt, inspect the failure, and generate the next only when you know what instruction or input should change. This controls cost and produces a clearer learning record.

Step 5: evaluate motion, not a preview frame

• Direction: does the camera travel the intended way from beginning to end?
• Speed: is acceleration deliberate, or does the shot lurch and stall?
• Framing: does the subject keep the intended size and position?
• Parallax: do near and far objects move consistently with the claimed camera path?
• Geometry: do walls, furniture, products and faces keep their shape while the view changes?
• Continuity: does the end frame make sense for the next edit?

Save the input, full output, selected model, prompt, settings, credit quote and acceptance decision. A still preview cannot prove that the camera move or subject stays consistent across the clip.

Common failures and the first fix to try

• The subject moves instead of the camera: state that the subject is stationary and name the physical camera path.
• A dolly looks like a zoom: request visible foreground parallax and include near, middle and far scene elements.
• The shot changes direction: shorten the reference and remove competing motion instructions.
• The camera is right but the subject warps: reduce movement, simplify the scene or use the original footage as the video-to-video base.
• The background changes unexpectedly: confirm whether you chose character Motion Control, image-to-video or source-preserving video-to-video; these workflows use the background differently.
• Handheld motion becomes jitter: ask for restrained drift and natural camera corrections rather than generic “camera shake.”

What exact copying requires

For a commercial shot that must match a precise camera track, use the original plate, motion-control data from production, or conventional compositing and 3D camera tracking. Generative prompting is useful for visual approximation and exploration; it is not a substitute for a measured camera path when continuity is contractual.

If you only need a new visual style on the same movement, start with Magic Hour Video-to-Video. If you want a different character to perform the reference action, use Motion Control. If the new subject begins as a still image and approximate camera direction is acceptable, use Image-to-Video.

Sora examples in older tutorials are historical

OpenAI discontinued Sora's app and web experience on April 26, 2026 and says the Sora API had a published discontinuation date of September 24, 2026. Do not choose Sora for a new long-term workflow. If you have existing Sora content, follow OpenAI's current discontinuation and export guidance.

Official sources checked

Workflow boundaries were checked September 13, 2026 using Magic Hour's Video-to-Video product documentation and Motion Control documentation. Sora dates come from OpenAI's discontinuation notice. Product capabilities and access can change; verify the selected mode before a production job.