ONCE Advertising Cinematography Research · Advertising Cinematography Research 03/10
Research article / work analysis · ONCE Editorial Team

One of the easiest ways to make a commercial feel “premium” is to set the camera in motion: glide through a doorway, pass a performer’s shoulder, skim a product and arrive at a close-up as the voice-over lands. Movement can make a shot feel expensive, continuous and controlled. It can also manufacture activity for an idea that has no action logic of its own.

This article asks: why does the camera move, and who is actually driving the frame? Our central claim is that meaningful camera movement is not merely a frame changing position. It is a describable change in the viewing relationship. It should reveal information that was previously hidden, follow an action with causal importance, redistribute the distance between person and product, or make the space participate in the argument. If performance, product action or production design already performs those tasks, a fixed camera may be clearer. If the camera moves without adding information, it may redirect attention from the product to the technique used to film it.

To keep preference separate from mechanism, we propose three falsifiable questions: Who is the source of movement? What information does the move add? What spatial cost does it create? We test the framework against verified production material for Honda’s Cog, B&Q’s Flip, and Apple AirPods’ Bounce. They are not equivalent films and they do not prove that a technique always works. Their value lies in showing three different allocations of kinetic authority.

Materials and method: separating the work, the disclosure and our interpretation

We reviewed Honda UK’s official retrospective on Cog, a production interview with director Antoine Bardou-Jacquet and Wieden+Kennedy creatives, and an industry project record; D&AD’s production account of B&Q’s Flip and its making-of page; and The One Club’s project description and credits for AirPods’ Bounce. We also conducted current searches across advertising video, creative marketing, digital marketing, AIGC and video production to test how the problem changes in short-form, multi-format and generated images.

These materials allow us to confirm the mechanical chain, two live-action passes and track-based camerawork in Cog; the back-to-back houses, offset rotating rig, practical tilt and controlled falls in Flip; and the city set raised by six feet, choreographed performer and in-camera work in Bounce. We did not perform an authorised frame-by-frame reconstruction of the three films, and we do not infer undisclosed focal lengths, stabiliser models, speed curves, lighting diagrams or post-production parameters. The analysis below keeps four categories separate: documented production fact, limited observation of available material, ONCE’s interpretation, and a proposed test that has not been run.

Test one: who is the source of movement?

Camera movement is often written as a list of verbs: push, orbit, travel, follow. A useful storyboard needs to identify the subject of those verbs. Change in the image may come from four sources: the camera changes position; a performer or product moves inside a fixed frame; the set, platform or light changes; or post-production and generative systems alter the viewpoint. All can make an image feel mobile, but they carry different causal meanings and production risks.

When a camera tracks a rolling object, the object appears to pull the viewpoint through the scene. When the camera remains a stable reference while a room turns over, the spatial abnormality becomes easier to read. When a performer bounces while the camera preserves a relative distance, the emphasis may fall on the relationship between body and product. Compressing all three into “a moving shot” erases the creative proposition before the shoot begins.

This is also why equipment cannot stand in for movement design. A dolly can create an almost imperceptible shift or a rapid lateral displacement. A stabiliser can enter a performer’s personal space or produce a viewpoint with no position at all. A locked camera may feel detached, or it may become more tense than any follow shot because the action inside the frame is so strong. Equipment defines a feasible range. The movement source determines who holds agency in the image.

Cog: the camera is not the engine but the witness to causality

Honda UK’s official retrospective describes Cog as a 120-second chain reaction beginning with a transmission bearing and proceeding through parts from the Accord. The official account says the final film joined two live-action takes and was shot over five days. A production interview adds that parts of dramatically different sizes required the camera to keep moving and refocusing; insufficient studio length was also why the sequence was divided into two passes. The director recalls roughly 60 takes. The camera had to dance up and down a track with the mechanical reaction, and a failure in either the chain or the camera required the entire setup to be reset.

The important point is not simply that a long take was difficult. The documented fact is that the camera tracked the car parts and finally moved in on the completed vehicle; the mechanical event happened first, and the viewpoint changed with it. Our interpretation is that the movement logic depends on a strict hierarchy: the camera cannot reveal the next device too early, and it cannot fall so far behind that the causal link breaks. Every lateral move, height adjustment and focus change must answer the question, “What is triggering what now?”

Movement in Cog therefore has a legible information gain. One moment shows a part; the following move reveals the next object that it will strike. The changing frame completes the next term in the causal chain. The camera does not manufacture the event; it hands the sequence of transmission to the viewer. A smooth, uniform move independent of the moving parts might be technically cleaner, but it would weaken the idea of an engineering system operating by itself.

The cost is equally clear. Track position, focus, component speed and studio length become mutually dependent. A late change to one product action affects the entire take. Continuity magnifies small errors and removes much of the editor’s freedom to repair them. A long take is not proof of truthfulness. It is a decision to bind the brand proposition to production risk. Cog works not because the shot is long, but because reliable operation is the very subject it asks us to watch.

Test two: what information does the movement add?

A path can look beautiful on a plan and still have no informational value. Compare the frame before and after the move. Does a new object appear? Does the distance between person and product change? Is an occlusion removed? Is spatial continuity established? Does the cause or consequence of an action become visible? If every answer is no, the move may add only parallax, dizziness or awareness of the rig.

“More emotional” also needs to be decomposed. A slow push can let the background recede while a face or product detail takes more screen area. A lateral move can use foreground and background parallax to reveal depth. An orbit can turn a flat package into a volume. These are conditional mechanisms, not a dictionary of emotions. A push can lose information when product focus drifts. An orbit may expose an unwanted reverse side, a lighting reflection or a compositing gap. Continuous stabilised flow can make every location speak in the same tone.

Advertising time is short, so information gain must be compared with comprehension cost. If the viewer has to understand what the camera is doing before they can understand what the product is doing, the hierarchy may have been inverted. A fixed camera does not mean an absence of rhythm. Entrance, falling props, rotating scenery, changing pools of light and sound can all carry time.

Flip: when the house turns, a stable reference can be stronger than a follow

D&AD’s production description of B&Q’s Flip states that the team built two back-to-back houses and gardens on a giant offset rotating rig, enabling a one-shot transition from an old space to a new one. The protagonist had to move between the two in real time. The tilting house and falling objects were achieved practically, while selected elements in the old house and garden used controlled releases to time their falls.

The value of the case is that it changes our intuition about what is moving. The finished idea is full of rotation, falling and inverted space, but the primary movement sources are first the set and the performer. If the camera compensated for every tilt and worked hard to keep the house visually level, it could make the event harder to understand. Some relatively stable visual reference is necessary for gravity, performance and falling objects to explain one another.

We do not invent a precise camera path from the written material. We make a conditional argument: when the set supplies the dominant movement, the camera’s task may not be to add a second layer of motion but to choose a viewpoint from which the transformation can be read. It may make necessary corrections to keep the performer visible without chasing every change into a smoothly centred composition. The new information is a rewritten spatial rule, not constant renewal of the viewpoint.

Flip also gives us a counterexample. Fixed does not mean passive, and following does not guarantee clarity. If a fully locked camera leaves the protagonist obscured during the rotation, the action line or product information disappears. If the camera continuously corrects the composition, the inversion can be cancelled out. The useful choice lies between those failures: let the viewer lose some stability while preserving enough reference to understand why stability was lost.

Test three: what is the spatial cost?

As soon as the camera moves, at least four things change: relative displacement of foreground and background, subject size, occlusion, and the angle at which light and reflections meet surfaces. Production costs change too. Lighting must stay outside a wider field of view. Sets need to hold up from more directions. Focus and wireless monitoring become harder. Performer blocking must share time with the camera. Visual-effects tracking and alternative aspect ratios inherit more constraints.

Movement is therefore not a free increase in energy. Close following can place the viewer inside an action while making a handheld product grow and shrink unpredictably. A low orbit can strengthen a product’s volume while exposing floors, ceilings and weaknesses in reflection control. A fast move can connect two spaces while removing an essential performer from a vertical crop. Every path exchanges one kind of information for another.

For a commercial, the useful record is not merely which device was used. It is what the move promises spatially: when the subject enters and leaves; when the product face must remain legible; how person-to-product distance changes; which background references must survive; and whether 16:9, 9:16 and square versions can share the trajectory. If these are absent from the technical scout, a perfectly smooth move can still fail delivery.

Bounce: following the body without losing the product relationship

The One Club’s project description for Apple AirPods’ Bounce says that the city becomes a trampoline and a young man’s commute becomes a jumping performance. To make the product’s promise of greater freedom feel physically credible, the team raised an entire city set six feet above the ground, allowing the performer’s choreographed movement to be captured in camera rather than created through visual effects. The page credits director Oscar Hudson and production companies Pulse Films and Framestore.

That disclosure lets us separate three kinetic layers: the performer changes height and speed on an elastic surface; the set creates a real physical condition; and the camera has to choose a relationship between readable performance and readable product. Our interpretation is that a follow shot here should not merely keep the performer centred. It needs to make take-off, force, suspension and landing spatially coherent without allowing AirPods to disappear as the product cue.

If the camera mechanically preserves the performer at the same size at every moment, the body’s changing relationship to the city can be flattened, leaving only expression and pose. If the lens movement over-emphasises environmental parallax, the connection between earbuds and experience becomes weaker. The value of following lies in deciding when to synchronise with the performer, when to let the body move within the frame, and when to lag briefly so the environment can react. Camera and performer are not master and servant; they divide the work of describing one action.

The case also exposes a risk in vertical cut-downs. A horizontal version can contain body trajectory, street reference and landing point. A vertical frame favours upward motion but may remove the horizontal object that explains cause or consequence. Multi-format adaptation should not place a tracking box on a face and call the problem solved. It must identify the “necessary second object” in each action: the ground, product, landing point or another person’s response.

Fixed, track and stabilised follow: three promises, not a hierarchy

The cases support a more useful comparison than equipment rankings.

A fixed camera promises a stable reference. It lets blocking, product action and production design reveal their own changes and makes before-and-after relationships easy to compare. It is vulnerable to occlusion and off-frame action; once the subject leaves the effective area, the camera offers no correction.

A constrained track or mechanical path promises repeatable spatial relationships. It suits causal chains, product travel, compositing alignment and reveals that depend on controlled parallax. The value of Cog is not the track itself but the shared timing between track and parts. The cost is that path, focus, lighting and art direction become more tightly locked.

A higher-degree-of-freedom stabilised follow promises responsiveness to performance. It works when a person leads and distance is being continually rewritten, and it can enter places where track is impractical. But the ability to follow does not mean the camera should compensate continuously. Excessive correction removes speed differences inside the frame, while floating movement can make every emotion feel alike.

Handheld does not automatically mean real, and robotics does not automatically mean a precise aesthetic. A movement system tells us how a camera can arrive somewhere, not why it should go there. The criteria remain movement source, information gain and spatial cost.

AIGC and virtual cinematography: separate camera motion from object motion first

Generated video makes this problem more explicit. The CVPR 2026 paper SymphoMotion treats camera motion and object dynamics as separate controls because parallax in a 2D image can entangle them. The ICLR 2026 work 3D Scene Prompting and MoCa likewise frame scene consistency and controllable camera trajectories as central problems. These are bounded research tasks, not evidence that commercial generation systems have solved long-take consistency.

For an advertising team, the useful response is not to stack “dolly, orbit, handheld” in a prompt. Write two independent trajectories first: how the camera moves in three-dimensional space, and how the person or product moves relative to the scene. Then define occlusion, scale, material and brand identifiers that must remain stable. If a generated background expands, product geometry drifts or a performer jumps position, the result should not be excused as stylised cinematography. The spatial relationship has failed.

The same separation improves live-action previs. A storyboard, animatic or technical previs should mark camera path, subject path, their relative speed and each information reveal. That can expose, before set construction and shooting, that a beautiful travelling move adds no new information—or that a fixed shot already does the job.

A minimum test not yet performed: shoot one action four ways

The following is a proposed experiment. It has not been performed for this article and has no results.

Choose a simple advertising action: a performer lifts a product from a table, walks to a window to use it, then turns toward another person. Keep performer, product, set, lighting, dialogue and total duration constant. Shoot four versions:

  1. Lock the camera and let the performer complete all movement.
  2. Move the camera on a single straight track, beginning only after the product is lifted.
  3. Use a stabilised follow that holds the performer at approximately constant size.
  4. Move only when the product first becomes obscured or when the second person is revealed; remain fixed otherwise.

Record three classes of observation. First, movement source: when camera, person, product and set begin to move. Second, information gain: what newly legible object or relationship appears after each camera move. Third, spatial cost: duration of legible product face, number of occlusions, focus failures, exposed lighting or set limitations, and losses in alternative crops. People who did not participate in production can also describe, in order, who did what, when the product changed state and how the distance between the two performers changed. That can test narrative legibility; it cannot prove sales, memory or a universal psychological effect.

The experiment can fail. Inconsistent performance speed contaminates the comparison. A different focal length in the stabilised version changes angle of view as well as movement. Unequal editing duration mixes rhythm with camera motion. Familiarity with the product may alter comprehension. Those conditions should be stored with the observations rather than hidden in favour of the smoothest result.

Write movement as a sentence that can be disproved

In a director’s treatment, “the camera slowly pushes in to create a premium feeling” is nearly impossible to test. A more useful instruction would be:

As the performer raises the product from the table to eye level, the camera travels half a body-width laterally over three seconds, revealing the product face previously hidden by the shoulder and the usage setting outside the window in the same frame. It then stops while the performer completes the action.

This sentence defines movement source, start condition, information gain and stop condition. The crew can test whether it works and propose a cheaper or clearer alternative. If the camera has no reason to stop, it often has not yet found a reason to start.

The first study on a director’s visual thesis examined how image rules grow from a brand proposition. The second study on lens and viewpoint separated perspective, field of view and spatial distance. This article adds time to the model: when does a spatial relationship change, who changes it, and what does the viewer know afterward?

To apply this logic to a live project, see ONCE’s TVC production and campaign collaboration scope. With a brand proposition, product action, primary delivery ratios and one spatial change that must be understood, send ONCE a project brief and begin by asking whether movement is necessary rather than which device to hire.

Sources and research note

Sources checked on 4 October 2026.

This is an original ONCE Editorial Team research article and work analysis, not a peer-reviewed academic paper. The cover is an original commercial concept image and does not depict the cited campaigns, an ONCE client project, a real set or an experiment. No third-party campaign frames are reproduced in this article.

← Focal Length Is Not Perspective: Designing Distance in Commercial Images The Product Reflects Its World: How Advertising Builds Material Value →