Sutra

Paper before pixels: the stage you approve at is the stage every correction happens at

The stage you approve at is the stage every correction has to happen at. Here is what paper actually contains for a short ad, what a change costs at each stage, and the cheap gate we added after hours of compute produced broken basics.

What is in here
  1. What does paper actually mean for a fifteen-second ad?
  2. The week that scored two out of a hundred
  3. What does a change cost at each stage?
  4. What should you actually be shown at each gate?
  5. A frame that exists gets defended
  6. The cheap thing between approved paper and a full build
The short answer

Because the stage you approve at is the stage every correction has to happen at. Approve a finished cut and a change means a re-render, a re-grade, and every check that was attached to the old bytes. Approve paper and a change means editing a sentence. We inverted our own production order after a week of work scored two out of a hundred, and the diagnosis was sequence rather than effort: every decision was being made from the expensive end.

What you get out of this
  1. What paper actually contains for a fifteen-second ad, written out in full
  2. What a change costs at each stage, from a sentence to a re-render that voids every verdict
  3. The four things you should be handed, in order, and how long each takes to judge
  4. The cheap gate we added after hours of compute produced work with broken basics

Five things people believe about storyboards

Flip them
Four of these five were said to us by somebody paying for creative. The fifth we believed ourselves for about a week, which is how long it took to cost us one.

01What does paper actually mean for a fifteen-second ad?

A beat sheet with real durations, one named hook described as an event on screen, the soundtrack committed as an actual file rather than a genre word, and every shot mapped to a file that exists or to a named route for getting it. A story with an unsourced critical shot and no route does not go on the menu at all.

One page from a story menu, then the lock it becomes

Copy this shape
STORY 04  "the drawer that already has one"
  emotion     habit, not relief
  hook        f0-f12: a hand opens a drawer and the pouch is already in it
  answer      "thirty in a pouch, one a night" lands at 4.2s
  sound       music. first sound at f0. 96 BPM. drop inside beat 3, not on its first frame
  unseen      the product introduced as furniture rather than as a product
  brand mark  first appears at 20-30% of runtime, held 1.0s at full opacity

BEATS   dur    what happens                       source
  01    1.4s   drawer opens, low angle            corpus / kitchen-03.mp4
  02    0.8s   pouch already in the tray          corpus / still-11.jpg, card move
  03    2.1s   patch to forearm, one motion       to generate, prompt written
  04    1.2s   drawer closes, room tone only      corpus / kitchen-03.mp4

COVERAGE    every critical shot EXISTS or names a route. no route = not on the menu
DEVIATION   <what departs from precedent> / <why precedent is silent> / <risk accepted>
Written out here as an example rather than lifted from a live project, but this is the shape ours take. The two lines people skip are COVERAGE and DEVIATION. The first stops a story being picked that we cannot actually shoot. The second turns an argument about taste into a decision somebody made on purpose, in writing, before it cost anything.

A mood board is not this. A mood board says how the ad should feel and leaves every decision that costs money undecided. The page above decides them all while they are still free, which is also the argument for treating the reference video as the specification rather than as inspiration.

The durations are the part that surprises people. They look like a formality and they are load-bearing: once every beat has a real length, the soundtrack can be derived from the shot table instead of guessed at, and changing a shot by six frames moves its sound hit with it automatically.

The same story, as it exists on paper before anything renders

Second by second
0s1s2s3s4s5s6s7s8s9s10s11s12s13s14sHook: the…Pouch alrea…Patch to forearm, one…Second ho…The answer line, as typeEnd cardBed, bass ledMix yields while you readEnds loudPICTURESOUND
Read it as a list
AtChannelWhat happens
0.0sPictureHook: the drawer opens
1.4sPicturePouch already in the tray
3.0sPictureBrand mark, first appearance
4.2sPicturePatch to forearm, one motion
7.0sPictureSecond hook, different mechanism
8.4sPictureThe answer line, as type
11.6sPictureEnd card
0.0sSoundFirst sound at frame zero
0.5sSoundBed, bass led
8.4sSoundMix yields while you read
11.6sSoundEnds loud
Every marker here is decided on paper. The brand mark is placed at about a fifth of the way in because that is where our own measurements put it, the sound starts at frame zero because films that fade in lose the scroll decision, and the mix yields under the type because reading and listening compete for the same attention.

What people mean by storyboard, and what we ask you to approve

The honest split
What people mean by storyboard, and what we ask you to approve
DimensionA storyboard, as usually meantWhat we hand over
The artifactA grid of thumbnailsYes. A menu of six to twelve stories, then one locked page
The hookNo. Implied by frame oneYes. Named as an event: what changes on screen in the first twelve frames
SoundNo. A genre word, chosen after the cutYes. An actual file or a written script, committed before the first cut
DurationsApproximateYes. Real, to the frame, because every sound hit is derived from them
ShotsDrawnYes. Mapped to files on disk, or to a named route for getting them
GapsNo. Invisible, because a drawing of a shot looks like a shotYes. Escalated by name before you approve anything
The right column takes about four minutes to read and roughly the same to argue with. That is the entire cost of the most consequential approval in the project.

02The week that scored two out of a hundred

We built the first production wave the obvious way. Pick an idea. Crop usable scenes out of whatever was on disk. Cut the ad. Fit music to the cuts. The work came back scored two out of a hundred, and it had burned about a fifth of a week's generation credits on edits that made the films worse rather than better.

The failure was not effort. Every decision was being made in the wrong order, so every correction landed at the most expensive stage available. Music was the clearest case: a bed fitted to a finished cut can only ever be a bed that does not fight the cut.

The order we banned, the order that replaced it, and what actually moved

Switch between them
Idea first, corpus second

This is the order almost everyone runs, because it is the order a person naturally thinks in. It puts the cheapest decision last and the most expensive one first.

  • Have an idea
  • Hunt for shots that could serve it
  • Cut the ad from what was found
  • Fit music to the finished cuts
  • Discover the idea was wrong, at the render
The third panel is the whole thing. Two of those three changes cost nothing to make and neither of them is about a tool.

A shot list derived from an inventory is not a story.

written down the day the wave scored two

The sound is decided on paper too, and that is the change people resist

A reel captures two senses. On a working split of ours, picture is roughly sixty percent of it and sound the other forty, and the two multiply rather than add: an excellent picture under a bland bed is a fifty to sixty percent film, while ninety and ninety gives you ninety. That is why the bed is a direction decision rather than a finishing task.

Vote, then see

The last time an ad of yours turned out wrong, when did you find out?

03What does a change cost at each stage?

Minutes on paper, an hour at the direction lock, minutes of compute at the mockup, most of a day once a full build exists, and after delivery it costs the relationship. Those numbers are the entire architecture. Every checkpoint exists to drag a decision leftward into a stage where being wrong is affordable.

The same change, priced at five different moments

Click along it
The story menu, before anything exists

Six to twelve candidates, each with its hook, its sound direction and its coverage. You kill the ones you do not want and say why.

A change here is a sentence. You can reject every single candidate and lose an afternoon.

Cost of a change here: minutes
Read the cost line rather than the stage name. Ours, from our own production log. The step nobody expects to matter is the fourth one: a re-render does not only cost the render, it voids every check that was measured on the old bytes.

04What should you actually be shown at each gate?

Four artifacts, in this order, and none of them is a finished ad. If a studio's first delivery is a finished ad, every note you give will be a repair note. There is a longer version of this in how to brief and QA ad creative.

The four things that should land in your inbox

One screen at a time
Checkpoint A
The menu, on paper. Four minutes

Six to twelve stories, one paragraph each: what happens, why a stranger stops in second one, what the ears get, and which shots exist already.

Your job is to kill, not to rank. Kill reasons are the most useful thing a client produces, and nobody can reliably say what is good in advance.

Checkpoint B
A contact sheet and the track. Thirty seconds

Stills from every beat in order, plus the actual soundtrack file. Play the track once while scrolling the sheet.

If the track feels like it belongs to a different film, say so now. Sound is close to half of what a viewer receives, and it multiplies with the picture rather than adding to it.

Mockup Gate
Styleframes and one motion test. Two minutes

Two or three finished-looking frames, a short motion test of the single riskiest beat, and a rough bed. Minutes of compute.

This is where you find out whether the thing in your head and the thing in ours are the same thing, at a price where being wrong is survivable.

Checkpoint C
Three to five finished variants. Watch cold

Watch them on a phone, at arm's length, with sound, without reading the notes first. Then pick.

Rework at this stage means picking, not re-directing. If you find yourself writing repair notes here, a gate upstream was skipped.

1 / 4

Roughly nine minutes of your attention in total, spread across three or four days. Nothing on this list requires you to be a video person, which is the point of writing them this way.

Six questions to ask of the paper page before you approve it

Tick as you go - it remembers
0%
Roughly four minutes, and none of it requires you to know anything about video. If you cannot tick a box, that is the note. Send the box back rather than approving and hoping the render fixes it.

05A frame that exists gets defended

This is the part that is about people rather than pipelines. Once frames exist, the conversation stops being about whether the idea was right and starts being about whether the render can be saved. Two documented effects push the same way: escalation of commitment, where a course already paid for attracts more support rather than less, and the IKEA effect reported by Norton, Mochon and Ariely, where people value what they helped assemble above the identical thing assembled without them.

The rule this producesNothing renders before the first yes. Not a test, not a quick look, not a mood piece. The same logic governs bad takes once generation starts: a wrong shot is regenerated from a changed instruction rather than patched, which is the argument in re-roll, never repair.

And the person who built it is the worst judge of it

There is a second-order version of this that costs studios more than the first. A maker reads their own intention into a frame, so the person who built a shot is the worst available judge of whether it lands. On the same films, our build side's self-scores ran between thirty-six and seventy-six points above mine. Self-scoring measures compliance with the brief. It does not measure whether the thing works.

Four numbers this order runs on

Measured in this studio
3checkpoints - A paper, B direction, C picks - plus the Mockup Gate between B and CSutra Haus pipeline
2/100the score of the wave built in the old order, which is why the order changedSutra Haus verdict log
0renders permitted before the story is approved on paper. There is no exception for a quick testSutra Haus pipeline
72hfrom brief to a first finished cut, with every checkpoint and the Mockup Gate inside that windowSutra Haus
All four ours, counted off our own production and verdict logs. The second one is the one we would rather not publish, and it is the reason the other three exist.

06The cheap thing between approved paper and a full build

Paper approval alone was not enough, and we found that out the expensive way. A draft review came back with a film scored one out of a hundred for having no visual hook, and eight static creatives carrying mechanical breakage: text running outside the plates it belonged to, an image chip almost entirely hidden behind other elements, labels too small to read. All of it after hours of compute.

The verdict was to fix the system before the outputs. One of the two rules that came out of it was the Mockup Gate: between approved paper and any full-cost build sits a cheap mockup that has to be ratified. Styleframes, a motion test of only the riskiest beat, a bed sketch. Minutes of compute, never hours. A full build before that yes is not permitted.

A hand tips a glass carafe of dark coffee toward a stoneware cup balanced on a stack of books, lit by hard low light, with the shadow of a bare branch thrown across the wall behind
The pour is the beat that gets the motion test. Pick the shot most likely to break, prove that one, and let the rest of the build follow it.

Choosing which beat to test is the skill. It is always the one where something physically happens, because that is where generated footage still gets caught. The other half of the same architecture, the checks that run in code so that no human has to catch them, is in deterministic pipelines plus AI generation.

Questions people actually ask

Open what you need
Do I need a storyboard for a video ad?

For anything over about six seconds, yes, though not in the drawn-thumbnails sense. What you need is a page that names the hook as an event, gives every beat a real duration, commits the sound, and says where each shot is coming from. That page is what makes the rest of the project cheap to change.

Why should I approve a storyboard before the video is made?

Because approving it later does not remove the decision, it only moves it to a stage where reversing it costs a re-render instead of a sentence. It also protects you from your own instincts: once frames exist, everyone in the room starts defending them, including you.

What should a storyboard for a fifteen-second ad include?

A named hook described as something that changes on screen in the first half second, a beat list with durations to the frame, the soundtrack as an actual file or script, and a source for every shot.

Add one line naming any departure from how the reference material does it. An undeclared departure is the single most common way a good plan turns into a bad film.

How long does the creative approval process take?

On our side, three checkpoints - A paper, B direction, C picks - plus the Mockup Gate between B and C, all inside seventy-two hours. That is about four minutes to read the menu, thirty seconds on the contact sheet and track, two minutes on the mockup, and a cold watch of the finished variants. Roughly nine minutes of your attention, spread over three or four days.

Can I skip the storyboard if the ad is AI-generated?

That is the case where skipping hurts most. Generation makes files cheap and decisions no cheaper, so a pipeline with no paper stage produces a large pile of finished things nobody chose. The volume argument only works if selection is real, and selection needs something to have been decided first.

What if I approve the paper and still dislike the finished ad?

That happens, and it should be planned for rather than defended against. Fewer than half the films that clear every automated check here survive the first human viewing. The gates do not make a film good. They make being wrong affordable, so the argument left at the end is about whether the ad works.

A hit rate is roughly the same for everybody, as Motion found across more than 550,000 ads, so the only thing you really control is how many genuinely different swings you get for your money. Every hour spent repairing a render is a swing you did not take.

Where the numbers came from

  1. Harvard Business School. The IKEA Effect: When Labor Leads to Love (Norton, Mochon and Ariely) - why people overvalue the thing they helped assemble, including a render
  2. Wikipedia. Escalation of commitment - the effect that makes an existing frame harder to kill than an idea
  3. Motion. Creative Benchmarks 2026: winners are rare - 550,000+ ads: the hit rate is roughly the same everywhere, so throughput is the lever

Every figure above links to the place it was published. Numbers marked as ours are measured inside this studio and we say so where they appear. We do not print a statistic we cannot point at.

Badal Kariwal

Runs Sutra Haus, a one-person ad studio that has shipped over a thousand finished creatives - film and stills - for DTC brands and hotels. Writes here about what the work actually taught him, including the parts that failed. The person who reads your brief is the person who builds the work. Send him something to make.

See the paper stage before you pay for anything

Send a link. Get the story menu, then the ad.

One finished cut built from your own product inside three days, free and yours to run whether or not we ever work together. You see the story on paper first, and you can kill every candidate on the page without it costing either of us anything.

Replies within a day. Ad within three.
Read next