Anyone who has handed a storyboard to a crew knows two kinds of collaborator. One shoots the boards. The other reads the boards as a suggestion and comes back with something you did not ask for. AI video models turn out to split along the same line, and a recent test makes the difference unusually easy to see.

One brief, two models, fifteen pairs

The team at Everypixel put two current video models, Seedance 2.0 and MiniMax H3, through the same set of briefs and published a shot-by-shot comparison of two AI video models in August 2026. The five scenarios read like a small slate of commercial and genre work: a handheld UGC selfie on a city street, a night car chase on wet neon roads, a stylized 3D cartoon about a small robot and a cookie jar, a cel-shaded 2D anime fight, and a 3D broadcast news opener with chrome typography. Each one was broken down beat by beat over ten seconds, with every camera position described.

Each scenario was then fed to the models in three ways: text only, text plus a pre-generated opening frame that locks characters, palette and framing, and a numbered six-panel storyboard sheet the model was asked to shoot in order. That makes fifteen head-to-head pairs; four are discussed in detail, and the authors say the other eleven were closer together.

They did not judge on pretty frames. Their criteria were the ones a 1st AD would care about: did the model shoot what the brief said, did anything appear that nobody asked for, did objects hold their shape in motion, did the shots read as one continuous event, and how many takes it took before a clip could go to a client.

The titan with two heads

The headline example is real and comes straight from the storyboard test. The six-panel anime sequence showed a lone hero on a broken rooftop, a stone titan as tall as the surrounding towers, and a single white slash through the titan’s wrist. According to the authors, MiniMax H3 shot the panels in the order they were drawn. Seedance 2.0 treated the first panel as a starting point, developed it, and returned a clip in which the titan had two heads, plus other elements that appear nowhere in the reference.

This was not a one-off. Across the storyboard runs, Seedance treated panels as artistic direction rather than specification, reordering beats and adding shots.

What each model got right and wrong

The chase is the useful counterweight. Both models needed roughly three attempts before either produced something usable, and the authors call it Seedance’s clearest advantage in the whole test. Their explanation: a ten-second action scene holds far more information than any prompt encodes, so a model that invents plausibly gets further on a thin brief.

On cost, the authors report that, at the API settings they used in production, ten seconds of Seedance output cost them three times as much as ten seconds on MiniMax. They flag that the Seedance figure includes a markup from the route they call it through, so it is not a clean rate-card comparison.

What this means for indie previs

The article’s key point is that improvisation and execution are a trade-off, not a quality gap. If you are using AI video to explore, to find a look or rough out an action beat you have not designed yet, a model that improvises is doing you a favour. Once you have boarded a scene and the creative decisions are made, the same habit turns every generation into an inspection pass: you have to find what changed before you can approve anything.

For small crews, pick the model by project stage, not by showreel. Exploration and concept work can tolerate invention. Previs that will be shown to a DP, a stunt coordinator or an investor needs a model that shoots the plan.

How to brief an AI video model like a DP

Give it a timed shot list, not a mood. Break the clip into beats with seconds attached, as the test did.

Specify every camera setup: size, angle, movement and where it starts and ends.

Lock the look with an opening frame before you ask for motion, so characters, palette and framing are fixed.

Number your storyboard panels and state the order explicitly.

Name what must not appear, especially if your model tends to add props or extra figures.

Budget for takes. On complex action, plan on roughly three generations per usable clip.

Review the periphery of the frame, not just the subject, before anything leaves the room.

What this comparison does not tell you

It is one set of briefs from one team, not a benchmark. Fifteen pairs, with four examined in detail.

The cost ratio reflects the specific endpoint the authors use, including a markup; they did not separate model price from route premium.

Seedance 2.5, which accepts up to thirty reference images and generates up to thirty seconds in one pass, was deliberately left out, so the results say nothing about long-form or reference-heavy work.

Self-hosting MiniMax H3 might change the economics further, but the authors say they have no credible estimate yet.

Before you choose a model, decide whether you want a collaborator who improvises or an operator who follows the boards. In this test, those were two different models.

Before Sunrise came out in 1995 and gave over almost its entire running time to two strangers walking around Vienna talking. No plot mechanics, no third act villain. Critics called it a gimmick. Psychologists who study attraction would recognize the structure immediately, because the film puts its characters through nearly every condition the research says produces closeness between strangers, and it does so in roughly the order the studies suggest.

The First Ten Seconds

Film grammar handles the opening of a date with a single reaction shot. Someone looks up, the camera holds for a beat, and the audience is told everything about how the evening will go. That compression matches the evidence. In a 2006 experiment, participants were shown photographs of unfamiliar faces for 100 milliseconds, then asked to rate attractiveness, likeability, competence, aggressiveness, and trustworthiness. Their snap ratings correlated highly with the ratings people gave when allowed unlimited time, and trustworthiness produced the strongest correlation of the set. Extending the exposure to a full second changed those ratings very little. On this point the reaction shot is a fair representation of what happens across a restaurant table.

Adrenaline and the Chase Scene

The convention where two characters fall for each other while being pursued or nearly killed has a documented basis. In 1974, researchers stationed an attractive interviewer on two bridges over the Capilano Canyon in British Columbia, one a solid structure and one a swaying footbridge 230 feet above the ground. Men approached on the high bridge were more likely to use sexual imagery in the writing task they were given, and more likely to call the interviewer afterward.

The accepted reading is that physical arousal from fear was misread as attraction to the person standing in front of them. Films used the effect for decades before the research named it, which is why so many first meetings on screen happen during a crisis and so few happen in a quiet room.

Relationship Types on Screen

One thing cinema handles better than most first-date advice is the range of what a relationship can look like. The long-distance couple, the age-gap pairing, the arranged introduction, the two people who met through a family friend, the older partners that audiences label sugar daddies when a film puts them on screen. All of them show up regularly, and audiences follow them without needing the setup explained.

That range is closer to life than the single template romantic comedies are accused of pushing. People choose different structures for different reasons, and films have been comfortable with that for decades.

Mutual Gaze and the Long Look

The held look between two characters is the most overused shot in the genre and one of the best supported. A 1989 study put 96 participants into pairs and assigned them to gaze at a stranger’s hands or eyes, or to count the other person’s blinks, for two minutes. Those who looked into their partner’s eyes while being looked at reported markedly higher affection than any other group, and pairs of opposite-sex strangers who held unbroken eye contact for the full two minutes reported increased feelings of passionate love.

Two minutes is a long time on screen. A director who holds a two-shot for eight seconds is already testing the audience’s patience, and the compression is one of the few places where film shortens something the research says needs duration.

Shared Laughter in Early Courtship

Screenwriters give their leads a running joke by the second scene, and the audience treats it as proof the pairing works. Research on humor in early courtship supports the instinct with one correction. A study of strangers meeting for the first time found that the number of times a man attempted humor and the number of times a woman laughed at those attempts both predicted her interest, and that the strongest signal of all was the two of them laughing at the same moment. Researchers describe the split in what each side does with a joke as the humor gap, with men more often producing the material and women more often selecting on it.

That pattern is old enough to be a screenwriting convention in its own right, and the reason the funny character is almost always the one pursuing.

The same researcher later pooled 43 samples covering more than 15,000 people and found every positive style of humor associated with higher relationship satisfaction, and most negative styles associated with lower. On screen, the running joke establishes that shared timing in about 40 seconds.

Escalating Disclosure

Good first-date dialogue starts with something trivial and arrives at something the character has told almost nobody. That progression is the exact design of a 1997 study at Stony Brook, where pairs of strangers worked through 36 questions arranged in three sets of rising intimacy, ending with prompts about death, family, and personal failure. After about 45 minutes, some participants reported a level of closeness they compared to their oldest friendships. One pair later married.

The study measured closeness rather than romantic love, and Arthur Aron, who designed it, spent years correcting people on that point. What it demonstrates is that mutual, escalating self-disclosure produces connection reliably enough to be run in a lab. A screenwriter who moves a conversation from parking tickets to dead parents in 20 minutes is following the same design.

Physical Proximity and the Shrinking Table

Watch where characters are placed. Booths instead of tables across a wide room, benches, a shared taxi, a walk where the two people keep drifting into each other. Directors close the physical distance steadily through a scene because the audience treats proximity as intimacy without being told to. The zones are well documented, from the roughly 18 inches of intimate distance out to the several feet of personal space people hold with strangers.

The staging matches what happens on real dates, where the distance between two people narrows in small increments and neither person can identify the moment it changed. Films speed this up, but the direction of travel is accurate.

Common Exaggerations in the Genre

The research support runs out in three places. First, the effortless verbal rhythm, where nobody talks over anyone and no question lands flat. Second, the absence of practical friction, since almost no film first date involves a delayed train, a phone that keeps buzzing, or a 20-minute silence while both people eat.

Third, and most misleading, the certainty. Characters know by the end of the night. Most people leave a genuinely promising first date unsure, and the uncertainty is not a sign that anything went wrong.

The Timing Distortion

Screen romance treats attraction as something settled in one evening, while the studies point at repeated contact over weeks as the stronger predictor of continued interest.

One Last Look at the Vienna Scene

Before Sunrise works because it gives its two strangers what the research calls for and then gets out of the way. Fourteen hours in one another’s company, sustained eye contact, escalating questions, steadily closing physical distance, and a deadline that supplies the arousal the Capilano bridge supplied. The plot is a fantasy, since almost nobody meets a stranger on a train and spends the night walking a foreign city. The mechanics underneath it are ordinary, and anyone who wants a better first date can copy them without the train.

Late in Rounders, Mike McDermott sits across from Teddy KGB through a long night of heads-up play, spots the Russian’s tell, and folds a key hand. KGB, irate at missing the chance to bust him, starts playing on tilt. The scene has almost no physical action, and its tension comes from decisions made behind the camera.

Information Given to the Audience

Alfred Hitchcock explained the principle to François Truffaut with a bomb under a table. If 2 people talk and the bomb explodes without warning, the audience gets 15 seconds of surprise. If the audience sees the bomb placed and a clock reading 12:45 when it is set for 1:00, the same conversation produces 15 minutes of suspense. His conclusion was that the public should be informed whenever possible.

At a card table, a hand where the viewer knows nothing until the cards are turned over produces a single moment of surprise. A hand where the viewer knows one player’s cards, or suspects a bluff, turns every bet and every pause into a question the audience is already trying to answer.

The director’s first decision is therefore how much to reveal and when. Revealing too little leaves the viewer watching chips move without meaning. Revealing everything at once removes the doubt that makes a hand worth watching.

Faces at the Table

A poker scene depends on actors who can do very little convincingly. Players are supposed to hide the strength of their cards. A performance that shows too much breaks the premise, and one that shows nothing gives the camera nothing to read.

Rounders solves the problem with contrast. John Malkovich plays Teddy KGB loudly, with a heavy Russian accent and a compulsive habit of eating Oreo cookies at the table, while Matt Damon plays Mike quietly. The viewer watches the loud player for leaks and the quiet one for nerve, and the difference in style tells the audience who is in control of himself before any cards are shown.

Rounders also seats a real player, Jake Swartout of West Palm Beach, at an Atlantic City table, and the faces around the leads look like those of a working card room. In The Cincinnati Kid, Edward G. Robinson approached the veteran stud champion Lancey Howard so personally that he later wrote, in his autobiography, that he did not so much identify with Lancey as become him.

Hole Cards and Point of View

In texas holdem, each player is dealt 2 private cards and shares 5 community cards with the table, so the screen can show the shared board to everyone while keeping each player’s own cards hidden.

A shot of one player lifting the corners of those 2 cards puts the viewer in that player’s position. From then on, the audience reads the other faces at the table the way the character does, looking for confidence or fear. When a film never shows a character’s cards, the viewer moves to the other side and studies that character from the outside.

Rounders uses both positions. The narration aligns the viewer with Mike, while Teddy KGB stays opaque until Mike spots his tell. The audience feels the relief of that discovery along with the character.

Shots Built in the Editing Room

Much of the intensity in a poker scene is assembled after filming. Aaron Sorkin, who directed Molly’s Game in 2017, has called televised poker the worst spectator sport ever. For the film he set aside 6 or 7 days for shooting only poker scenes, collected a large number of inserts such as chips being pushed forward and cards being dealt, and hired real players for the background. In the editing room, those short shots became fast sequences built around Molly watching the game. Sorkin also wanted the long games to look improvised, with no visible choreography.

The approach is older than Sorkin’s film. The Cincinnati Kid was directed in 1965 by Norman Jewison, who took over after Sam Peckinpah was fired 4 days into shooting, and was photographed by Philip H. Lathrop and edited by Hal Ashby. In its heads-up finale, the film moves around the table through close-ups of the cast, and the cuts come faster as the final hand approaches its end. The faster cutting raises the tension before anyone knows the cards.

The principle behind these cuts was demonstrated by Lev Kuleshov in the 1910s and 1920s. He alternated one shot of actor Ivan Mozzhukhin’s expressionless face with a bowl of soup and with a girl in a coffin, and audiences saw hunger in one case and grief in the other, although the face never changed. In a poker scene, the shot placed next to a still face decides if the viewer reads fear or calm in it.

Close-ups also show information that a wide shot loses. A hand resting on a chip stack, a glance at the clock or a swallow before a call can tell the audience more about a player’s position than any line of dialogue, and an editor can stay on that detail for exactly as long as the scene needs.

Soundtrack Choices

Robert Altman’s California Split, released in 1974, pioneered his use of an 8-track sound recording system, which allowed the overlapping dialogue he is best known for. Altman also cast real gamblers, including former world champion Amarillo Slim, and had the background players talk at the same time. The card rooms in the film sound crowded and distracted.

Other films go the opposite way and strip the soundtrack down to the click of chips and the snap of cards. Against that near silence, a single bet becomes loud. The card-playing scenes in Rounders take a third approach, with Matt Damon’s voiceover guiding the viewer and explaining terms an outsider would not know, so the tension does not depend on the audience already knowing the game.

Rooms, Light and Smoke

John Dahl shot Rounders in about 45 days, beginning in December 1997, mostly in New York, and gave the games the look of a film noir drama, with dimly lit rooms and cigarette smoke. The dark rooms keep the frame on the players and the felt.

Jewison and Lathrop lit The Cincinnati Kid in Metrocolor with strong reds, yellows and greens, which makes the game’s world vivid and gives the final hand the feel of a public event.

The Fold Revisited

The final game in Rounders works because the film has already placed the viewer behind Mike’s eyes. By the time KGB’s tell appears, the audience has learned the vocabulary through the narration and knows what Mike stands to lose. When Mike baits KGB into going all-in on the last hand and wins with the nut straight, the payoff belongs to the viewer as much as to the character.

She has a very particular set of skills. Skills she has acquired over a very long career…

We’ll soon be seeing action legend Michelle Yeoh (Everything Everywhere All at Once, Royal Warriors) in The Surgeon, an upcoming action thriller Continue reading →