To tell a story in a 15 second AI ad, give every shot one piece of information the viewer can see, and arrange those shots into a beginning, a turn and an end. Four or five shots are enough. Let eyes, hands, a prop, light and sound carry the meaning instead of dialogue or text. If someone can watch your ad with the sound off and tell you what changed between the first frame and the last, you have a story. If they can only tell you what the product is, you have a product video.
Key takeaways
- A story is a change you can see. Something or someone is different at the end than at the start.
- In 15 seconds, every shot has to carry exactly one new piece of information.
- Four or five shots of roughly 2 to 5 seconds each is enough for a beginning, a turn and an end.
- Dialogue is the easy option. Eyelines, hands, props, light, posture, the camera and sound do the job faster and work in any language.
- Plan the story as a shot list before you generate anything. In AI the shot list is the script.
What counts as a story in a 15 second ad?
A story is a change. At the start, a person or a situation is one way. Something happens. At the end, it is a different way. That is all a story needs, and it fits in 15 seconds.
What it doesn't need is explanation. You don't need to say who the characters are, where they live or what they want. You need the viewer to see the before, see the moment that changes it, and see the after.
Most AI ads I see skip the change. A beautiful product, the product from another angle, then the logo. That is a mood board, not an ad.
Why does dialogue usually make short ads weaker?
Alfred Hitchcock put it simply in his interviews with François Truffaut: "When we tell a story in cinema we should resort to dialogue only when it's impossible to do otherwise."
Words are slow. A line of dialogue takes three seconds to say what a look says in one, and it only works for people who have the sound on and speak the language.
Directors have proved this at every length. Stanley Kubrick's 2001: A Space Odyssey has no dialogue in its first 25 minutes. Pixar's Up tells the whole marriage of Carl and Ellie, from their wedding to her death, in a sequence only minutes long that is almost completely silent. It works through images and music, and it is one of the most famous stretches of storytelling in modern animation.
This matters more now because AI video can talk. Seedance, for example, can generate synchronised dialogue together with the video. That makes it tempting to let a character explain the story out loud, which in 15 seconds is usually the weakest choice.
What does every shot in a short ad need to carry?
One piece of visible information. Not two, not a mood. One thing the viewer learns that they didn't know a second ago.
A useful test: describe each shot in a short sentence that starts with "we learn". We learn the bottle is old. We learn someone is watching him. We learn she approves. If you can't finish that sentence, the shot isn't carrying anything and it can go.
This works because the audience builds the meaning between shots. The classic demonstration is the Kuleshov experiment. Around 1920, Lev Kuleshov cut the same expressionless close-up of the actor Ivan Mosjoukine together with a bowl of soup, a girl in a coffin and a woman on a divan, and audiences reportedly read hunger, grief and desire in the same face. The original experiment's reliability is debated, and one recreation didn't find the effect, but later studies support the basic idea: context changes how we read a face.
For a short ad, that is great news. You don't need a perfect performance in every shot. You need the right shot next to it.
How do you fit a beginning, turn and end into 15 seconds?
Think in three beats and four or five shots.
The beginning (about 3 to 7 seconds). Show what is normal, and hook the viewer while you do it. The first shot has to stop the scroll, which I go into in depth in how to hook viewers in the first two seconds. Google's ABCD guidelines for YouTube ads say to open with people on screen if the ad features any, use tight framing, and get to more than two shots in the first five seconds. All of that fits a story opening.
The turn (about 4 to 6 seconds). Something arrives, breaks, is discovered or is decided. This is the shot that makes it a story. It is usually a face reacting, a hand doing something, or a reveal.
The end (about 3 to 5 seconds). Show the result. How things are different now. This is also where the product usually lands in its most natural place, because the change happened because of it, around it or with it.
Google's guidelines also say to introduce the product or brand in the first five seconds and to humanize the story with people. In practice, that means the product appears early as part of the world, not as a logo card, and the ending gives it its hero moment.
Which devices tell a story without words?
These are what I direct when there is no dialogue. Each one is something you can write into a prompt and a model can simulate.
Eyelines and off-screen space. If someone looks out of frame, the viewer wants to see what they see, and your next shot answers it. A look and what it lands on is the simplest story unit in film.
Hands. Hands show decisions: hesitating over a door handle, letting go of something, pushing an object across a table.
A storytelling prop. One object can carry a whole backstory. In Up, a jar of spare change for a dream trip keeps being broken open for life's emergencies, and that one prop tells years of a marriage. A ticket, a ring, a key, a letter, a bottle with handwriting on it.
Light change. A light switching off or a door opening onto daylight is a change in time and mood nobody has to explain. More in how to prompt light in AI images and video.
Posture. Shoulders dropping, a head lifting, arms unfolding. The body changes before the face does.
The camera. A slow push-in says something is happening inside this person. A rack focus moves attention from one thing to another without a cut, which is perfect for a reveal.
Sound. More on this below.
| Device | What it tells the viewer | Best beat | How to prompt it |
|---|---|---|---|
| Eyeline to off-screen | Something important is over there | Beginning or turn | "looks past camera towards the doorway on the right" |
| Hands | A decision is being made | Turn | One hand action in close-up, everything else still |
| Storytelling prop | Backstory, history, meaning | Beginning or end | Describe the object's wear and detail precisely |
| Light change | Time has passed, mood has shifted | Turn or end | Name the light source and how it changes |
| Posture | How a character feels about what just happened | Turn | Describe what the shoulders, head or arms physically do |
| Camera | Where to look and how close to feel | Any | Locked off, slow push-in, slow slide or rack focus |
| Sound | What is happening off screen, how it feels | Any | Generate or add specific sounds, not just music |
How do you use sound to tell the story?
Films only have two ways into your head: the eyes and the ears. Most AI ads use the second one for a stock music track and nothing else.
Sound can carry information the image doesn't show. A door closing off screen tells you someone left. Silence right before the turn makes the turn land.
You have a few ways to build it now. Some video models generate audio with the video. Seedance can output synchronised sound effects and ambient audio with the frames, so you can describe the sound you want in the prompt. For more control, ElevenLabs has a text to sound effects generator that makes clips up to 30 seconds, with a loop option for room tone and ambience. For music, Suno generates tracks from a text prompt, including instrumentals.
My rule: the story has to read with the sound off, because a lot of people will watch it that way. Then sound makes it land harder for everyone who has it on.
Worked example: an olive oil brand in four shots
Here is how I would turn a simple brief into a 15 second story.
The brief: a small family olive oil brand. The message is that the oil is made the way it always has been, and it is passed down.
The story in one sentence: a young man cooks for his grandmother for the first time, using her oil, and she hands the bottle on to him.
The change: at the start, it is her oil. At the end, it is his.
| Shot | Seconds | We learn | Device |
|---|---|---|---|
| 1 | 0 to 3 | There is an old bottle with handwriting on it, and it matters | Prop, one thing moves |
| 2 | 3 to 7 | A young man is cooking, and someone is watching him | Rack focus, eyeline |
| 3 | 7 to 11 | She tastes it, and she approves | Face, slow push-in |
| 4 | 11 to 15 | She passes the bottle to him | Hands, prop |
Four shots, one piece of information each, and the product is on screen in three of them.
Shot 1: the hook (0 to 3 seconds)
Extreme close-up of a dark green glass bottle of olive oil on a worn wooden kitchen table, a paper label with faded handwriting in blue ink. A single golden drop of oil runs slowly down the side of the bottle from the neck. Camera locked off, zero camera movement, nothing else in the frame moves. Warm late afternoon light from a window on the left, deep shadows, shallow depth of field, fine film grain.
Shot 2: the beginning (3 to 7 seconds)
Close-up from behind a young man's shoulder of his hands pouring olive oil from a dark green bottle into a hot iron pan on an old gas stove, a little too much oil, it spits. Camera locked off. Focus starts sharp on the pan and his hands, then slowly racks to the background, where an elderly woman in a grey cardigan stands in the kitchen doorway with her arms folded, watching him. Warm late afternoon light from a window on the left, small cluttered family kitchen, shallow depth of field.
Shot 3: the turn (7 to 11 seconds)
Close-up of an elderly woman with deep lines around her eyes, seated at a worn wooden kitchen table, lifting a piece of bread dipped in olive oil to her mouth. She chews slowly, her eyes close, her jaw slows, then she lets out a short breath through her nose and gives one small nod, the corners of her mouth lifting very slightly. Very slow, steady push-in from a medium close-up to a close-up on her face. Warm late afternoon light from a window on the left, deep shadow on the right side of her face, visible skin texture.
Shot 4: the end (11 to 15 seconds)
Top-down shot of a worn wooden kitchen table, an elderly woman's hand slowly slides a dark green glass bottle of olive oil across the table into a young man's open hand, his fingers close around it. Camera locked off directly above. Only the two hands and the bottle move. Warm late afternoon light from the left, two plates and a piece of bread at the edge of frame, fine film grain.
Sound: room tone and a clock for shot 1, the oil spitting too loudly in shot 2, near silence for the taste in shot 3, the bottle sliding on wood in shot 4, and a few notes of quiet guitar coming in only for the last four seconds.
Notice that shot 3 doesn't say she is happy or proud. It says what her eyes, jaw, breath and mouth do, in order. In my experience, models simulate a body much better than an emotion word. I break that method down fully in how to direct emotion in AI video.
Also notice the camera. Every shot is locked off, a slow push-in or a rack focus. In my experience those are the moves current models handle reliably, and they keep the attention on the story rather than on the camera.
How do you keep four shots feeling like one film?
A story only works if the audience believes it is the same people in the same place. Three things make the difference:
- Same characters. Generate the young man and the grandmother from the same reference images or character sheet in every shot. Faces that drift between shots break the story instantly. I cover the full workflow in how to keep a character consistent across AI video shots.
- Same light. Describe the same light source in every prompt. "Warm late afternoon light from a window on the left" appears in all four prompts above for a reason.
- Start from images. Make each first frame as a still until everything matches, then animate. Fixing a still is much cheaper than fixing a clip.
What mistakes kill a short AI story?
- No change. Four shots of the product from different angles, with no before and after.
- Explaining with text. A caption that says "She taught him everything" means the images didn't do their job.
- Too many shots. Eight shots in 15 seconds leaves under two seconds each, and the viewer can't read them.
- Ending on a logo. The end beat is the payoff. Give the product its moment inside the story, then add the logo.
- Inconsistent characters. If the grandmother looks different in shot 3, the audience is no longer watching a story, they're spotting differences.
- Music only. A stock track over everything is a missed chance to tell part of the story with sound.
Checklist before you generate
- Can you say what changes between the first frame and the last, in one sentence?
- Does every shot pass the "we learn" test?
- Is there a clear turn, a shot where something is decided, discovered or arrives?
- Does the first shot work as a hook on its own?
- Does the story read with the sound off?
- Are the characters and the light the same in every prompt?
- Is every camera move one a model can pull off cleanly?
FAQ
How many shots should a 15 second AI ad have? Four or five is a good range. That gives each shot roughly 2 to 5 seconds, which is long enough for the viewer to read one piece of information.
Should my AI ad have dialogue? Only when the image can't do the job. Dialogue is slower than an image, only works with the sound on, and only for people who speak the language.
How do I show emotion without dialogue in AI video? Write what the body physically does, in order: what the eyes, jaw, breath and mouth do. Avoid emotion words like happy or sad in the prompt.
Do I need sound if the story works with the sound off? Yes. The story should read silently, but sound makes it land harder for everyone who has it on, and it can carry information the image doesn't show.
How do I keep the same characters across the shots? Generate every shot from the same reference images or character sheet, describe the same light in every prompt, and build each first frame as a still before you animate it.
Sources
- Hitchcock/Truffaut, quote on dialogue, via Goodreads: "Hitchcock/Truffaut Quotes"
- 2001: A Space Odyssey trivia on dialogue, MovieMistakes: "2001: A Space Odyssey trivia"
- Up opening sequence, Wikipedia: "Up opening sequence"
- Kuleshov effect, Wikipedia: "Kuleshov effect"
- Google, "ABCDs of effective video ads" and the YouTube ABCD reference guide
- Adobe Firefly, "Seedance partner model" page, on synchronised dialogue, sound effects and ambient audio
- Scenario Help Center, "ElevenLabs Sound Effects (SFX): The Essentials"
- Soundverse, "How to Write Effective Prompts for Instrumental Music on Suno.ai"
Want to direct AI video like this?
AI Unchained is my community for marketers, brand owners and filmmakers who want to direct AI video, not just prompt it. Every week we break down a film technique and put it into practice, there are three live calls a week, members get our prompt GPTs, and I reply to every piece posted in the Cinema Room the same day.
Running a brand or a content team? I also run in-company AI video workshops, so your team learns to make ads and social content in house.
