A brand that buys AI video is buying motion as much as images, and motion is where generated footage most often gives itself away. Almost every film in a theatre still runs at 24 frames per second, a number set for practical reasons in the late 1920s that audiences now read as cinema. Each step in that history, from hand-cranked cameras to motion smoothing on TVs, has a counterpart in how a model draws movement today.

REC 24 FRAMES = ONE SECOND SHUTTER OPEN 1/48 S PER FRAME
One second of film, and the small blur in every frame that makes it feel fluid.

Before sound, speed was a choice

Silent films had no fixed speed. Cameras were cranked by hand, and according to The Bioscope, a blog devoted to early film, silent-era speeds generally ranged from 14 to 24 frames per second and got faster up to the late 1920s. Exhibitors often ran films faster, simply to squeeze more into a show. A film shot at 18 frames and played at 24 runs a third too fast, which is why silent films can look sped up today.

Generated video has its own hand crank. A model has no clock tied to the real world, so the speed of a movement inside a clip is whatever it learned to draw. A walk can drift as if underwater, a dropped object can fall at the wrong rate. A Google DeepMind team that tested video models on physical scenarios concluded that “visual realism does not imply physical understanding.” The fix starts with watching every clip at real speed. Tempo can go in the prompt, a filmed motion reference can set the timing in a video-to-video pass, and a near miss can be retimed in the edit, though slowing a clip means inventing frames.

Sound locked it at 24

Synchronized sound ended that freedom. Film Independent notes that a 14% change in speed is tolerable in the picture, but in sound it is “far more noticeable and annoying.” Every projector had to run at the same rate as every camera. Encyclopaedia Britannica dates the standard to 1927: 24 frames per second, or 90 feet per minute of 35 mm film.

Why 24 and not 30? The likeliest answer is thrift. The Bioscope suggests it was probably the minimum speed that gave the first optical soundtracks decent fidelity, and Film Independent points to slow film stocks and the high cost of film. Nobody picked 24 for its beauty. Audiences grew up with it anyway.

Sound still exposes timing, and generated dialogue shows it first. A mouth a couple of frames ahead of the voice reads as a bad dub, however realistic the face. The same kind of technology can also repair footage: Variety reported that the makers of the 2022 thriller Fall used AI to alter actors’ mouth movements so that more than 30 F-words became milder lines, taking the film from R to PG-13 without reshoots. On a brand project, lock the voice track first, sync picture to it at the master frame rate, and check again after the final conform, since any retime can pull mouth and voice apart.

The other half of the look: the shutter

Frame rate is only half the recipe. The other half is how long each frame is exposed. The classic setting is a 180-degree shutter, open for half of each frame’s time, which RED Digital Cinema puts at close to 1/48 of a second at 24 frames per second. Anything that moves during that exposure leaves a slight smear, and that smear is what makes 24 frames feel fluid. RED notes that a narrower angle leaves gaps between the blurs, so frames start to read like separate stills.

A generated clip has no shutter. The model draws every frame, and the blur in it is paint, its learned idea of how that movement should smear. That is why a generation that looks great as a still can feel off in motion, too clean in one shot and smeared in the next. It bites hardest in compositing: put a product filmed with a 180-degree shutter in front of a razor-sharp generated background, and the product looks cut out. Shoot inserts at the master rate and shutter, match blur in the composite, and recheck after upscaling, which can sharpen edges that should stay soft.

When audiences saw more frames

Peter Jackson shot The Hobbit: An Unexpected Journey at 48 frames per second, double the standard. When it opened in December 2012, most theatres showed it at 24. Reviewing the high-frame-rate version for TheWrap, Steve Pond described “the disconcerting look of a high-def video production.”

Ang Lee went further with Gemini Man in 2019, shot at 120 frames per second, in 3D and 4K. SlashFilm reported that only 14 screens in the United States could show it at 120, and its critic found the action “videogame-like.”

How many frames per second?

Frame rates in film and television, in frames per second

Silent era, speeds variedSilent era, speeds varied: 14 to 2414 to 24Sound standard, 1927Sound standard, 1927: 2424The Hobbit, high frame rate, 2012The Hobbit, high frame rate, 2012: 4848US broadcast video, motion as seenUS broadcast video, motion as seen: 6060Gemini Man, 2019Gemini Man, 2019: 120120
Frame rates compared
FormatFrames per second
Silent era (speeds varied)14 to 24
Sound film standard, 192724
The Hobbit: An Unexpected Journey, high frame rate version, 201248
US broadcast video, motion as seen by viewers60
Gemini Man, 2019120
Sources: The Bioscope (silent era), Encyclopaedia Britannica (1927 standard), TheWrap (The Hobbit), What Hi-Fi? (broadcast video), SlashFilm (Gemini Man).

James Cameron took another route with Avatar: The Way of Water in 2022. Engadget reported that the film uses 48 frames for major action and doubles frames to mimic 24 in calmer dialogue scenes. Cameron has long called high frame rate “a tool, not a format.”

Pond’s complaint was about association: extra clarity made a fantasy epic look like video. Generated motion triggers the same reading when it is too even, with a camera that glides without weight and faces that never blur. For a brand, that reads as cheap before anyone spots a flaw. Cameron’s logic carries over: keep 24 for the cinematic master, and choose crisper motion before generation, only where the content earns it.

The soap opera effect, at home and in generated clips

Many TVs offer motion smoothing, which invents frames between the real ones. What Hi-Fi? gives the example of a 24 fps movie raised to 60. The result recalls soap operas, shot on video so that viewers saw motion sampled at 50 or 60 per second. In December 2018, Tom Cruise and director Christopher McQuarrie recorded a public service announcement asking viewers to turn it off. Cruise said it makes most movies “look like they were shot on high-speed video.”

The same trick is common in AI video work. Frame interpolation, which invents in-between frames, is used to stretch a short generation, slow it down or raise its frame rate. It is also how VBench, a research benchmark from Nanyang Technological University and Shanghai AI Laboratory, scores “motion smoothness”: it drops every other frame and checks how well an interpolation model rebuilds them. The same paper observed that models strong in temporal consistency often show less motion overall. Smooth can simply mean that little happens.

So a clip that looks silky in a preview may be the soap opera effect in disguise. Interpolated frames are guesses, and fast hands, thin objects and anything passing behind something else are where the guesses smear. When a shot is too short, generating a longer take, or chaining clips from a shared first or last frame, often holds up better than stretching one clip.

What to ask before production

A brand film usually mixes generated shots, stock footage and a filmed product insert, each possibly at a different rate. Conforming them to one timeline means repeating, dropping or inventing frames, and left to the export it shows up as stutter or rubbery smoothness. Motion belongs to a film’s look as much as its colour palette, so decide it early.

Which frame rate is the master? For a brand film meant to feel cinematic, 24 is the safe answer. Anything higher should be a deliberate choice.

Is every shot conformed to it before the edit? Generated, filmed and stock footage alike.

Which clips were interpolated or retimed? Ask to see them at full speed, watching hands and fast edges first.

Does motion blur match from shot to shot? Check the composites first.

Is lip sync checked after the final conform? And are the social versions cut from the master and checked on a phone?

Sources

Gabriel Brien

Gabriel Brien

Founder of Crimson Spark Agency. AI filmmaker and creative technologist, writing from running this work daily.

More about Gabriel →