Show a marketing team a generated product video and someone often says it looks “too clean.” Skin comes out poreless, surfaces read like plastic, and every frame is equally sharp. Viewers rarely name the cause, but they feel it.

Cinema had this argument when digital cameras replaced film, and some productions now pay to put texture back: grain, halation, imperfect lenses. That history explains why the finish of an AI video matters, and how texture helps generated shots sit next to each other and next to filmed footage.

REC SAME FRAME, TWO FINISHES HALATION GRAIN AS RENDERED ONE TEXTURE PASS
The same frame before and after a texture pass: grain, a halo around the highlight, softer edges.

What grain actually is

Grain is part of how film forms an image. Netflix’s engineers describe it as tiny particles formed during development, and they credit it with a role in storytelling, adding depth and realism. The particles fall in a different random pattern on every frame, so grain keeps moving even when the picture is still.

Netflix takes grain seriously enough to rebuild it. In a 2025 post, its engineers explain that heavily grained titles are cleaned before encoding, a description of the grain is sent with the video, and the grain is synthesized again on the viewer’s screen. On one grainy feature, They Cloned Tyrone, the bitrate fell from 8,274 to 2,804 kbps. Across about 300 titles, the average saving at 1080p and above was 36%.

Generated video also begins with noise: most video models start from random noise and remove it step by step until an image appears. Film keeps its grain, while a model is trained to clear noise away, and what it leaves on skin and fabric is a smoothed guess at texture.

Why some directors still pay for celluloid

In January 2020, The Hollywood Reporter wrote that Kodak had signed new agreements with Disney, NBCUniversal, Paramount, Sony and Warner Bros., each committing to buy undisclosed amounts of motion picture film. The article listed directors then shooting on it, among them Quentin Tarantino, Greta Gerwig, Martin Scorsese and Christopher Nolan. Earlier pacts, signed in 2015, were believed to cover two years; the new ones were thought to run longer.

For Nolan’s Oppenheimer, Kodak manufactured its Double-X black-and-white negative in 65mm, which the company described as a first, since that stock had never gone through an IMAX workflow. IndieWire reported that the IMAX and Panavision cameras had to be adapted to the new emulsion. Cinematographer Hoyte van Hoytema shot the color scenes on Kodak’s 250D and 500T negatives, which he told Kodak “still had enough texture for me.” To IndieWire, he described the film as “a three-hour-long movie about faces.” The appetite has lasted: Kodak counted 20 productions shot on its film at Cannes in 2026, seven of them in competition, twelve on 35mm and eight on 16mm.

Faces are also where generated video most often slips. A generated close-up tends toward even, poreless skin, closer to a retouched beauty shot than to film, which matters in any testimonial. A reference photo of real skin helps; a texture pass after generation holds the look.

Halation and imperfect glass

Halation is the glow that rings bright highlights on film. Light passes through the emulsion, bounces off the back of the film and returns through the layers; the red-sensitive layer, closest to that surface, is hit first, which is why the halo looks red or orange. Kodak has limited it since 1934 with a carbon backing called remjet, which complicated lab processing. In 2025 it announced an anti-halation undercoat to replace that backing across its Vision3 stocks, promising the same image quality.

Some productions want those effects badly enough to manufacture them. Greig Fraser shot Dune on a digital Arri Alexa LF, then the graded image was laser-recorded onto Kodak film and scanned back for a final grade. American Cinematographer describes the aim as a grounded look for a fantastical world, and lists what the round trip added: halation, gate weave, breathing, softness and a grain suited to IMAX. “The process added a depth and substance to the image that’s a bit intangible,” Fraser said.

On The Batman, Fraser used anamorphic lenses that Arri Rental detuned for him, with astigmatism, coma and edges that fall off softly. Wide open they flared and lost contrast; closing the aperture slightly brought contrast back, so he could set it shot by shot. For a car chase he added rehoused Soviet still lenses, praising their “lack of scientific perfection.”

A model can paint a halo or a lens flare on request, but with no glass behind it, a flare may drift from its lamp or vanish in the next generation. Optical character holds better when added after generation, on every shot at once.

Grain as the glue between shots

Netflix’s engineers call grain “notoriously difficult to compress,” because it is random. That is why their pipeline separates it from the picture and rebuilds it under control, and why finishing teams treat texture as its own layer rather than something left to chance.

Generated video needs that separation more than film does. Many generators work on a compressed version of the picture: in the 2021 paper that set out latent diffusion, Robin Rombach and colleagues describe a “perceptual compression” stage that removes high-frequency details. Pores, fabric weave and grain are that kind of detail, so the decoder has to reinvent them. Each clip is also rendered alone, one crisp, the next softer, so cut together they look like footage from different cameras.

One texture pass for the whole edit

Where grain belongs in a project that mixes generated and filmed shots

1. Generate and upscale: Every shot first, texture held back1Generate and upscaleEvery shot first, texture held back2. Even out the base: Remove stray noise and oversharpening2Even out the baseRemove stray noise and oversharpening3. Grade together: Generated and filmed shots side by side3Grade togetherGenerated and filmed shots side by side4. One texture layer: Same grain and halation on every cut4One texture layerSame grain and halation on every cut5. Check the final file: After platform compression, at real size5Check the final fileAfter platform compression, at real size
A common finishing order for mixed projects. Netflix’s grain pipeline (2025) follows the same logic: take the noise out, then put it back under control.

One grain and halation treatment over every cut, after grading, gives generated and filmed shots a shared surface. Keep it light: heavy grain reads as a filter, can blur a label and compresses badly. It also fixes nothing structural; a warped logo stays warped.

What to check before you sign off

Ask who handles texture, and when: grain should go on after all generation and upscaling, across the whole cut. Ask to see a generated shot next to a filmed one at full screen, and look at skin in the close-ups. Check that halos and flares follow real light sources and stay consistent from shot to shot. Read your product label after grain. Then judge the export on the platforms where it will run, at the size people will watch it.

Sources

Gabriel Brien

Gabriel Brien

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

More about Gabriel →