Focus Tells the Audience Where to Look

Every frame of a video ad answers a question before the viewer asks it: where should I look? Much of the answer comes from focus. On a camera, what is sharp and what goes soft follows from the lens, the aperture and the distance to the subject. In AI video there is no lens. The model paints blur as a look it learned from millions of images, and it can paint it in the wrong place.

For a brand, that is a practical matter. The product has to be sharp when it matters, the shots of one scene have to share a look, and the blur has to behave like optics or the image starts to feel fake. Three films show what focus can do for a story, and each holds a lesson for generated work.

Citizen Kane keeps everything sharp

Writing in American Cinematographer in February 1941, Gregg Toland explained the look he built with Orson Welles for Citizen Kane. He wanted the audience to feel it was “looking at reality, rather than merely at a movie,” and argued that the human eye, for practical purposes, works like a lens with everything in focus. He shot interiors at f/8 to f/16, on fast Super-XX film with coated lenses and powerful arc lights. In one scene, he wrote, a face less than 16 inches from the lens stayed sharp along with a group of men 12 to 18 feet away. George E. Turner later listed the wide-angle lenses, down to 24mm, and noted that because the sets had ceilings, nearly everything was lit from the floor.

That is the standard recipe for deep focus: relatively wide-angle lenses and small apertures, so near and distant planes are sharp at once. It moves the job of guiding the eye from the lens to the staging. Roger Ebert wrote that with everything sharp, “composition and movement determined where the eye looked first.” Optics alone did not get there. Yale University Library’s film notes record that optical printing helped create many of the film’s deep-focus compositions, and that effects pioneer Linwood Dunn estimated about half the shots involved visual effects.

A generated shot can be sharp front to back at no cost in light, and that is often why it looks wrong. A model asked for everything in focus tends to give every surface the same crisp detail, and the image loses its hierarchy. Kane’s deep focus was staged. A brand that wants deep focus needs the prompt and first frame to carry the staging: who is near, who is far, what the eye finds first.

Barry Lyndon shoots by candlelight

Stanley Kubrick wanted the night interiors of Barry Lyndon (1975) lit by candles alone. According to ZEISS, only ten of its Planar 50mm f/0.7 lenses were made, six of them for NASA, and Kubrick bought three. The lens let in about four times the light of an f/1.4, and the candles were given three wicks to burn a little brighter. Cinematographer John Alcott described 70-candle chandeliers, a key light of roughly three foot-candles, and the whole picture pushed one stop in development. For wider framings, an adapter turned the 50mm into a 36.5mm lens that still opened to f/0.7.

The price was focus. Alcott told American Cinematographer in 1976 that there was “hardly any depth of field” with that lens. The crew set a closed-circuit video camera at 90 degrees to the film camera, taped the actors’ positions onto a grid over its monitor and read distances from it, which gave the actors some freedom to move and still stay sharp. Faces hold while the room falls away into soft light.

Generated video has no light budget, so its blur is a style the model reproduces, and it fails in its own ways. Through a lens, blur grows with distance from the plane of focus. In a generation, a hand reaching toward the camera can stay crisp, a background can be smeared evenly whatever its depth, and focus can drift mid-clip with nobody pulling it.

The Graduate points with blur

Shallow focus can also point. On The Graduate (1967), Robert Surtees, already a three-time Oscar winner, chose to “emphasize selective focusing,” keeping the subject sharp and letting foregrounds and backgrounds go soft, American Cinematographer recounts. Some shots used a 500mm lens, and Panavision lent its new reflex camera to handle the critical focus.

A rack focus moves attention inside a single shot by shifting the plane of focus from one subject to another. On set, the focus puller turns the focus ring between two marked distances, timed to a line or a look. In the scene where Elaine realizes Benjamin’s affair was with her mother, Tim Dirks’ Filmsite describes her out-of-focus face slowly coming into focus. The realization lands without a cut.

Video models know the term, but a requested rack focus is fragile. It may start early, land on the wrong subject or stop halfway, and the soft background can change shape mid-pull. A focus puller’s discipline helps: one move per shot, a still camera, start and end subjects named plainly, several takes to choose from.

Why all-sharp and all-creamy both look fake

Real optics follow rules viewers have absorbed without knowing them. Columbia University’s film glossary defines depth of field as the range of distances that appears sharp. Everything at the same distance shares the same sharpness, blur builds gradually away from that plane, and a longer lens or a wider aperture narrows the range. Bright points in the background open into soft discs shaped by the aperture.

Generated images break these rules both ways: sharp everywhere, or everything but the face drowned in the same creamy blur. Research on image models shows why. A 2025 paper found that prompting for blur gives “crude approximations” and can alter the scene itself; a 2024 study found that asking for a 24mm versus a 70mm lens did not yield consistent images of one scene. Across separate shots, that means a creamy background in the wide and a crisp one in the close-up. Pick one depth of field per scene and repeat the same wording and references in every shot.

Who decides focus, then and now

Focus decisions on a film set and their counterparts in generated video

Focus decisions on set and their counterparts in generated video

On set	In generated video
Aperture and lens chosen per scene	Depth of field written the same way in every shot
Actors’ marks taped for the focus puller	Sharp subject fixed in the first frame
Focus puller racks between subjects	One rack focus per shot, start and end subjects named
Optical printer joins separate exposures	Filmed product composited onto a plate matched for blur
Dailies checked for sharpness	Full-size review of product edges

A filmed product in a generated room

Joining planes that one lens could not hold together is an old trick. Kane’s optical printer combined separately photographed images into one frame, and the seam had to stay invisible.

Product work often follows that logic now: the product is filmed for real so the label stays exact, then composited into a generated setting. Focus is often where these composites give themselves away. A crisp bottle on a heavily blurred background looks cut out, and so does a soft product in a razor-sharp room. Record lens, aperture and distance at the product shoot so the background can match, or generate it sharp and defocus it in compositing with a depth matte.

What to check before you approve

Ask which subject should be sharp in each shot, and check it at full resolution, down to the logo and label. Ask what depth of field each scene uses, and watch consecutive shots for a background that sharpens or softens without reason. Check that any rack focus lands on the right subject and that nothing in the blur changes shape. For a filmed product in a generated setting, ask for the product shoot’s lens, aperture and distance, and study the edges. Toland and Kubrick chose their focus for the story; in generated video, someone still has to choose it.