Orelon logoOrelon
Pricing

Large Format vs Super 35mm: Cinematic Depth of Field

Nov 15, 2026 · By Orelon Team

Explore AI video templates

Browse a few community creations for inspiration, then open any template to continue creating in Orelon.

Compare large format and Super 35mm depth of field, see how AI video models estimate focus falloff, and prompt cinematic separation that holds up on screen.

Depth of field is the fastest way an audience reads scale. Before anyone can name a sensor or a lens, they register whether the room behind a face is a place worth studying or a soft wash of colour they are meant to ignore. That judgment arrives in under a second, and it is carried almost entirely by how focus falls away from the subject.

Two formats dominate the conversation. Large format, built around sensors closer to a full-frame stills chip, gives a thin, isolating slice of focus with a fast, dramatic decay behind the subject. Super 35mm, the gate that carried film production for decades and still anchors most digital cinema cameras, keeps more of the mid-ground legible and stays forgiving when people move. Choosing between them used to be settled at the rental counter. Increasingly it is settled at the prompt, or in post, where AI-assisted tools estimate depth and reshape falloff frame by frame.

That shift is worth taking seriously. A convincing large format look is not "more blur." It is a specific relationship between sensor area, focal length, subject distance, focus plane, aperture behaviour, and the shape of out-of-focus highlights. Get those relationships wrong and the image reads as a phone photo with a filter on it. Get them right and viewers simply accept the world you built.

Why format is a narrative decision, not a hardware flex

Directors talk about format as if it were a technical badge. It is really an editorial choice about how much of the world the audience is allowed to read at any moment.

A shallow-focus large format image says: this face is the entire story, everything else is texture. A deeper Super 35mm image says: this person exists inside a place, and the place is talking too. Neither is more cinematic. A film where every shot is at maximum aperture feels airless, because contrast is what gives separation its power. The moment you open up has to mean something relative to the shots around it.

This is why AI generation gets interesting. When the format decision lives in software, you can vary it scene by scene without changing camera bodies mid-shoot. You can also, unfortunately, vary it by accident, because a model will happily give you a hybrid look that matches neither format. The rest of this guide is about avoiding that hybrid.

What large format and Super 35mm actually change

Field of view and perspective compression

Sensor area changes what counts as a normal lens. A 50mm on large format frames roughly what a 35mm frames on Super 35mm. To match framing you either move the camera or change the lens, and both alter spatial relationships.

If you stand close with a longer lens, faces round out and backgrounds compress gently. If you step back with a wider lens, the environment stretches, corners pull, and the space between subject and background feels larger. Same subject, same frame, different world. Prompts that specify "large format, shallow depth of field" without implying a focal length and a camera distance tend to produce a wide-lens perspective with an artificially thin focus band, which is exactly the combination that looks wrong.

Falloff speed and the shape of the gradient

Large format gives a thinner slice of focus at a given aperture largely because you are using a longer lens at the same framing, and longer lenses magnify blur circles. Super 35mm keeps more readable mid-ground, which is why it remains the workhorse for ensemble scenes, action, and anything where the set is doing narrative work.

The detail that separates a real image from a synthetic one is not the amount of blur but the rate of change. Real defocus accelerates. The first thirty centimetres behind the subject still hold detail, then the falloff steepens, then backgrounds dissolve into shape and colour. AI blur frequently applies a more linear ramp, which produces a cutout quality: sharp subject, uniform mush, nothing in between.

The emotional grammar of blur

Ask one question before writing any prompt: should the viewer be able to read the room behind the actor? If yes, you want Super 35mm grammar even if the rest of the sequence leans large format. If no, and the background is emotionally irrelevant, thin focus buys you intimacy and perceived production value without complex lighting.

How AI models reconstruct depth and lens character

Depth estimation is a guess, and it fails in predictable places

Modern pipelines do not apply a flat gaussian blur. They estimate a depth map per frame, then blur each region according to estimated distance. That is a real improvement, and it has characteristic failure points you can learn to spot in a single review pass.

Look at hair against a distant background, glasses frames, thin foreground elements like grass, cables, or window mullions, and any edge where a subject overlaps something far behind. When a halo of softness eats into a sharp ear, the depth map is wrong, not the blur radius. Fixing the map with a rough mask before defocusing is faster than trying to paint the blur away afterwards.

Where lens character lives

Real lenses do not produce uniform blur. Out-of-focus highlights take the shape of the aperture: round when wide open, polygonal when stopped down, and stretched into a cat's-eye shape toward the corners because of mechanical vignetting. Some glass renders nervous, busy bokeh; some renders creamy, near-liquid falloff. Anamorphic designs stretch highlights horizontally and smear point lights into lines.

Ask a model for "bokeh" and you get something plausible but generic. Ask for "large format, oval highlights stretching toward frame edges, streetlights blooming into horizontal flares, focus plane on the collar" and you get something that reads as a specific lens. Vocabulary is what buys specificity, and it is worth keeping a personal file of phrases that worked.

A prompt vocabulary that buys specificity

Vague adjectives get vague results. The table below is a translation layer you can lift into your own prompts.

Intent What to write
Large format portrait "large format sensor, 75mm equivalent, subject 1.2m from camera, focus on the near eye, background falling off within a metre, creamy round highlights"
Super 35mm ensemble "super 35mm, 28mm equivalent, three subjects at mixed distances, deep focus through the mid-ground, far wall readable with light falloff"
Anamorphic night exterior "large format with anamorphic glass, oval highlights stretching to frame edges, streetlights flaring horizontally, focus plane on the jacket collar"
Documentary realism "super 35mm, 35mm equivalent, handheld, focus hunting slightly on a moving subject, background soft but still legible"
Product beauty "large format, 90mm equivalent, tabletop distance, focus on the front edge of the label, rear surface dissolving, no visible depth edge"

Every line specifies sensor character, focal length equivalent, subject distance, and where the focus plane sits. Those four pieces of information do more work than any mood adjective you can stack on top.

Three levers you can actually pull

Distance, focal length, and focus plane are the controls that matter, and all three are expressible in language.

Distance decides how much space exists between subject and background, and therefore how much falloff you can see. A person one metre from a wall will never look like large format at maximum aperture, no matter what the prompt says, because there is nothing behind them to blur.

Focal length decides perspective and blur magnification together. Longer lenses flatter faces and compress backgrounds; wider lenses open space and stretch corners.

Focus plane decides what the shot is about. "Focus on the near eye, background three metres behind" is worth ten adjectives, because it tells the model where sharpness begins and how far it has to travel.

If you are generating stills first to lock a look, the AI image generator is a cheap place to iterate on framing and falloff before committing to motion, where mistakes are more expensive to review.

Worked example: reshaping a Super 35mm plate into a large format look

Say you have a medium shot of a woman at a kitchen table, shot on Super 35mm at a 35mm equivalent, and the background is legible enough that the cut to your large format close-ups feels jarring. Here is a three-pass approach that keeps the result believable.

Pass one: isolate and estimate

Run depth estimation on the plate, then inspect the map before touching the image. Check hair edges, the gap between her arm and her torso, the chair back behind her, and any glass or steam in frame. Correct the map manually where it breaks. Blurring a bad map only produces a bad shot faster and with more confidence.

Pass two: defocus in zones, not with a slider

Pick a focus plane in metres. Decide how quickly sharpness decays in front of it and behind it, and make the two different, because real optics are asymmetric: depth of field usually extends further behind the subject than in front. Add a highlight-weighted bloom in the defocused areas so point lights glow more than flat surfaces do. That one adjustment separates a photographic result from a filtered one.

Pass three: match grain, contrast, and grade

A thin depth of field concentrates attention, so the sharp zone will look sharper and more contrasty than everything else. Add a touch of local contrast on the subject plane, then match grain across the full frame so the defocused region does not look digitally clean. Grain also hides depth-map artefacts, which is a practical bonus rather than a cheat.

Finally, check the grade at full size. Shallow focus removes a lot of colour information from the frame, so a grade built for a deep-focus shot often looks oversaturated once half the image goes soft.

Decision criteria: which look for which shot

Reach for large format character when the scene is intimate and the background is emotionally irrelevant; when you want production value without elaborate lighting; when the subject is static or moving slowly toward camera; when you are generating portraits, product beauty shots, or close dialogue.

Reach for Super 35mm character when the environment carries story information; when several actors share the frame at different distances; when the camera moves significantly; when you need to intercut with documentary, broadcast, or archival material; when you are working fast and need forgiving focus errors.

A blunt rule of thumb: if you would be annoyed that a viewer cannot read the background, use Super 35mm. If you would be annoyed that they can, use large format.

Six mistakes that expose synthetic depth of field

Blurring everything except the face. Real falloff is gradual. The first metre behind a subject still carries detail, and losing that middle zone is the most common giveaway.

Forgetting the foreground. An out-of-focus shoulder, doorway, or glass on a table sells depth more effectively than a soft background. Most defaults ignore it entirely.

Uniform highlight shape. Evenly blurred point lights are the clearest tell of a synthetic effect. Specify aperture shape and bloom.

Overusing the effect. If every shot is maximum separation, the format stops meaning anything and the piece starts to feel airless.

Mismatching format and motion. A sweeping camera move inside an extreme shallow-focus look leaves the viewer with too little spatial information to track. Wider depth is usually the better partner for movement.

Ignoring frame size. Depth reads differently in vertical crops. A composition built for widescreen can lose the exact gradient that made it feel cinematic once reframed, so check both aspect ratios before delivery.

Keeping a mixed timeline coherent

Combining generated shots with camera footage is the hardest scenario, because audiences forgive a lot but they do not forgive a cut where the background snaps from creamy to crunchy.

Build a small reference sheet before generating anything: two or three frames from the real footage, annotated with focal length equivalent, distance to subject, focus plane, falloff behaviour, and highlight shape. Paste the same descriptive sentence into every prompt in the sequence. Consistency of language produces consistency of look far more reliably than consistency of seed.

If you are building an entire piece from scratch, decide the format once and write it down. A short reference note such as "large format character, 60mm equivalent, focus on the near eye, background 3m behind, soft round highlights, fine grain" pasted into each prompt keeps a sequence from drifting. The prompt library is a good place to start building that sheet, and the video templates give you framing and motion starting points that you can then layer your own optical recipe over.

When the look works, save it as a written recipe rather than a vibe: sensor character, focal length equivalent, subject distance, focus plane, falloff description, highlight shape, grain. That recipe is the difference between a visual identity and a folder of unrelated nice-looking clips. Our AI video generator carries lens and format language through to motion, which matters once subjects start walking through the focus plane.

FAQ

Can AI really turn Super 35mm footage into a convincing large format look?

Within limits, yes. It can estimate depth, apply physically plausible falloff, and add lens character. It cannot invent spatial information that was never captured, so a deep-focus wide shot will not become a true large format close-up. What it can do is shift the perceived format enough that a sequence intercuts smoothly.

Does shallower depth of field always look more cinematic?

No. It looks more isolating. Cinema uses both deep and shallow focus, and the contrast between them is what gives each one impact. A piece where every shot is wide open feels flat in a different way than a piece that is all deep focus.

What is the single most useful prompt detail?

The focus plane, described in distance terms relative to the subject. "Focus on the near eye, background three metres behind" gives a model far more to work with than "shallow depth of field."

Why does AI bokeh look fake?

Usually because it is uniform. Real out-of-focus highlights have aperture shape, edge stretching through the frame, and brightness variation. Asking explicitly for aperture shape, corner stretching, and highlight bloom fixes most of it.

How do I keep depth consistent across a long sequence?

Write the optical recipe down and reuse the exact same phrasing in every prompt. Consistency of language beats consistency of settings, and it survives model updates better than a saved preset.

Is large format better for vertical video?

The narrower frame already compresses the background, so moderate separation often reads as more than it is. Be careful with extreme falloff in vertical, because there is less horizontal room for the gradient to breathe.

How much of this can I fix after generation?

Most of it, if you have a usable depth pass. Correct the map first, then defocus in zones, then match grain and contrast. Trying to fix a broken map with more blur never works, because the error is structural rather than cosmetic.

Make the format a decision, not an accident

Depth of field works on viewers before they have a chance to think. Choosing between large format and Super 35mm is choosing how much of the world your audience is allowed to see, and AI has moved that choice from the rental house to the prompting stage.

The reliable path is not a magic phrase. It is a small, disciplined vocabulary: sensor character, focal length equivalent, subject distance, focus plane, falloff, highlight shape, grain. Say those things precisely and the results stop looking generated. You can read more workflow breakdowns on the Orelon blog if you want to build that vocabulary systematically.

Orelon is an AI video generator for cinematic ideas in motion, built for exactly this kind of control, where lens language and format choices carry through from prompt to finished shot. Describe the optics, not just the subject, and watch how far the look travels. Create your first shot and treat depth of field as a creative decision from frame one.