Short 15-Second Promo Videos: An AI Production Workflow

Plan, prompt, and edit 15-second AI promo videos that convert, with shot maps, prompting habits, placement tips, and a launch checklist.

Fifteen seconds is not a trimmed commercial. It is its own format with its own job: earn the next three seconds, then the next ten, then the tap. The viewer is holding a phone in one hand, sound is probably off, and the decision happens before the first frame finishes moving. AI video generation has compressed the production timeline for this format from weeks to an afternoon, but speed is not the real advantage — iteration is. A team that can produce eight honest attempts at an opening learns something a single polished cut never teaches.

This guide walks the whole loop: why the format behaves differently, how to design a promise before touching a tool, how to build and prompt a shot list, when generation beats a camera, how to edit, where placement changes the cut, the mistakes that quietly sink results, and how to measure whether the video actually did its job.

Why Fifteen Seconds Rewards a Different Kind of Thinking

Longer ads can build. Fifteen seconds cannot. Every second must carry either new information or a new emotional beat, ideally both, and the constraint exposes weak ideas fast. If the concept is vague, there is nowhere to hide it.

Four structural realities shape the work:

The first frame is the thumbnail. Most feeds show a still before motion begins. An opening frame that is a logo on white spends the only free impression on nothing.
Sound is optional, reading is not. A large share of views start muted, so text, motion, and expression must carry the message before anyone taps for audio.
There is room for exactly one idea. A promo that sells a product, explains a mission, and announces a discount usually sells none of them.
Replays are a feature. Short loops get watched twice, which means a clean loop point can double effective watch time.

The practical rule that follows: one promo, one promise. Write the promise as a sentence a viewer could repeat after a single viewing. If you cannot write that sentence, the script is not ready.

The iteration advantage

Traditional production makes variation expensive, so teams test two or three cuts and call it a campaign. Generation flips that ratio. When a new shot costs minutes rather than a shoot day, the sensible move is six to eight genuinely different openings for the same offer, run against a small budget, and let the data choose. The winner is rarely the internal favourite — which is exactly why the test is worth running.

Designing the Promise Before You Open Any Tool

The most common waste in AI production is generating before deciding. Beautiful clips that do not serve a message are expensive even when they are cheap.

The one-sentence test

Write the sentence a viewer should be able to repeat. Examples: “Cold brew in ninety seconds, no equipment.” “The CRM that closes the loop between sales and support.” “A mattress that survives a move without sagging.” Every downstream decision — shot list, pacing, captions, call to action — is judged against that sentence.

Choose the metric before the script

Different jobs need different promos. Decide up front what the asset is for:

Job	Primary signal	What the cut must do
Awareness	three-second view rate	stop the scroll with one strong image
Consideration	hold rate or completion	explain the mechanism in one beat
Retargeting	click-through rate	repeat the offer and remove friction
Retention	reply sentiment	show a use case the viewer recognizes

A promo optimized for awareness and a promo optimized for retargeting look different even for the same product. Producing one asset that tries to do both is the fastest way to make neither work.

Write the voiceover line first, even if you never record it

A single spoken sentence — roughly thirty-five words maximum — forces discipline. If the line does not fit into fifteen seconds read aloud at a natural pace, the concept is too complicated for the format.

Building a Shot List That Fits the Runtime

A fifteen-second promo usually needs five to eight shots held for roughly 1.5 to 3 seconds each. Sketch beats as text before opening any generator:

Hook image (0:00–0:02)
Product in action (0:02–0:05)
Detail or proof shot (0:05–0:08)
Human reaction or outcome (0:08–0:11)
Differentiator or offer (0:11–0:13)
Call to action (0:13–0:15)

This is not a law, but it blocks the most common failure: three gorgeous product shots and no reason to care.

Three timing examples

Software. 0:00 a cursor stuck in a slow approval loop; 0:03 the interface resolving it in one click; 0:07 a calm dashboard with the queue empty; 0:10 one line of text naming the outcome; 0:13 the trial offer. The product screen must appear within two seconds — viewers are evaluating the interface, not the story.

A drinks brand. 0:00 macro liquid pour with condensation on glass; 0:04 hands lifting the bottle from a fridge shelf; 0:08 a person drinking, eyes closed; 0:11 pack shot with the flavour name; 0:13 offer. Texture sells here, and the human beat prevents the cut from feeling like a stock montage.

A local service. 0:00 the problem in one frame — a leaking pipe, a cluttered garage, a flat tyre; 0:04 the technician arriving with a branded van in frame; 0:08 the fix in progress, hands and tools; 0:11 before and after split; 0:13 service area plus booking prompt. Local audiences respond to recognizable streets and uniforms far more than to polish.

Storyboard in text, not in frames

Write each beat as one line with a verb: “coffee pours into glass,” “cursor selects approve,” “hand turns the key.” Verbs make prompts easier and expose beats that are really stills rather than motion. A beat list of six verbs is already half the production.

Prompting Shots That Survive the Edit

Generation quality is high enough that the bottleneck is usually the brief, not the model. Treat each prompt as a shot brief with six components: subject, action, camera, lens and framing, light, and mood. Duration and aspect ratio are technical parameters, not creative ones.

Weak prompt: “coffee shop, cinematic.”

Strong prompt: “Close-up of dark coffee pouring into a glass cup on a walnut counter, slow motion, 50mm lens, shallow depth of field, warm morning window light from the left, visible steam, calm and premium mood, vertical 9:16, five seconds.”

The second version gives the model decisions to make instead of a category to interpret.

Lock the look with reference frames

Generate a still of your hero shot first and reuse it as a visual reference for every related clip. Colour, texture, and set dressing stay consistent, and the edit needs far less correction later. An AI image generator is the cheapest way to build those anchors, and three reference stills are usually enough for an entire campaign.

Describe camera motion explicitly

“Slow push in,” “handheld follow,” “static tripod,” and “dolly left” prevent the drifting, weightless camera that makes generated footage feel synthetic. If a shot does not need motion, say so — stillness reads as confidence.

Generate longer than you need

A five-second clip gives room to trim into the strongest 1.8 seconds. Cutting into the middle of a generation usually hides artefacts at the head and tail, where models tend to warm up and fall apart.

Failures worth checking before you fall in love with a shot

Hands. Extra fingers, fused knuckles, objects passing through palms. Crop tighter or replace the shot.
Text. Signage, labels, and screens generate as almost-letters. Keep real text out of generated frames and add it in the edit.
Morphing. Objects changing shape mid-move, liquids that become solids. Shorten the clip or pick another take.
Crowd faces. Background people smear at small sizes. Blur or reframe.

Keep a prompt log. When a shot works, save the exact wording, the reference frame, and the parameters. Reusable prompts are what turn occasional luck into a production system, and a prompt library habit pays for itself within two campaigns.

When to Generate and When to Shoot

Situation	Better choice
Conceptual or impossible scenes	Generation
Real employees or customers speaking	Live action
Packaging accuracy and product texture	Live action, or generation with a real photo reference
Rapid variant testing of openings	Generation
Regulated claims with exact on-screen text	Live action or motion graphics
B-roll of places you cannot travel to	Generation

The honest answer is often hybrid: generation for atmosphere, transitions, and impossible angles; real footage for the product and the people. Audiences forgive stylized scenes but never a wrong logo.

Editing Rules That Decide Whether It Looks Professional

Generated footage does not edit itself, and the gap between polished and synthetic usually lives in the timeline rather than the model.

Cut on motion

Trim each clip so the cut lands while something is already moving. Cuts during static frames read as accidents. A simple trick: place the cut two frames after the peak of a movement.

Rhythm over rule

Aim for roughly 1.5 to 2.5 seconds per shot, with one deliberately longer beat for the hero product. Anything under a second reads as a flash; anything over three starts to sag unless it carries a strong human moment.

Typography: one idea per card

Seven words maximum, high contrast, positioned inside the safe zone so platform interface does not cover it. Use one typeface and two weights across the whole cut — inconsistent type is the fastest way to look amateur.

Captions are not optional

Burned-in captions improve comprehension in noisy rooms and are an accessibility requirement. Keep them to two lines, place them above the bottom interface, and check legibility at phone size with screen brightness turned down.

Sound design in three layers

A music bed that matches brand energy, two or three tactile effects (pour, click, whoosh, door), and only if needed a voiceover. The highest-impact moment is often silence right before the call to action.

Grade once, over everything

A single adjustment layer across the timeline for contrast and white balance does more for perceived quality than any individual shot. Clips generated from different prompts rarely match out of the box.

Placement Changes the Cut

The same fifteen seconds should not ship identically everywhere. Placement determines framing, the first frame, and where the call to action sits.

Paid social (9:16, 4:5). Hook inside 1.5 seconds, captions on by default, call to action in the final two seconds and repeated as on-screen text.
Landing page hero (16:9, 1:1). A slower opening is fine because the visitor already clicked. Lead with clarity, not disruption.
Email. A silent autoplay loop beats a full audio promo. Design the first two seconds to work as a still.
Retail or trade show loop. No sound, no captions, seamless repeat. The loop point matters more than the ending.
App store preview. Interface within two seconds, no branding preamble.

Build the vertical cut first because it is the most constrained, then reframe outward. Doing it in reverse usually produces a vertical version that looks cropped rather than composed.

Mistakes That Quietly Sink Performance

Too many ideas. Features, mission, and price in one cut means the viewer retains none of them.
Logo-first openings. Brand recall is worthless before interest exists.
Generic AI aesthetics. Floating shapes, glowing gradients, and drifting cameras signal template to experienced viewers. Ground every shot in a physical place or object.
Unreadable text. Small type, low contrast, or text hidden under platform interface.
A late call to action. The last frame should still be readable if someone stops at second twelve.
No sound-off test. Watch the finished cut muted at phone size. If the message collapses, the edit is unfinished.
One version only. A single cut teaches nothing about which hook the audience actually wants.
Skipping review. Claims, prices, and disclaimers need sign-off before launch, not after the first thousand impressions.

How to Measure a Fifteen-Second Asset

Short-form performance is decided early, so measurement should reflect that.

Hook rate: three-second views divided by impressions. This grades your opening frame and first cut.
Hold rate: completion or midpoint views divided by starts. A high hook rate with a low hold rate means the promise and the payoff disagree.
Click-through rate: the clearest signal that the offer and the call to action line up.
Cost per action: the only number that ties creative to revenue.
Comment sentiment: on organic posts, replies reveal whether the hook was understood or merely noticed.

Change one variable at a time — opening frame, first line of text, music genre, or call-to-action wording. Eight unrelated variants produce noise, not insight. And scale in steps: a winning hook in a small test usually survives a bigger budget, but not always.

FAQ

Five to eight for most offers. Fewer feels static; more becomes a strobe. The exception is a deliberate fast montage for a sale or event, where eight to twelve very short cuts work if the audio carries continuity.

Can one prompt produce a finished fifteen-second video?

Not with the specificity marketing needs. Generation works best shot by shot, assembled in an editor. Single-prompt results drift in style and cannot guarantee product accuracy or a readable call to action.

Do I need a voiceover?

Only if the product needs explaining. Many strong promos run on music, effects, and captions. If you do use one, write for the ear, keep it near thirty-five words, and record after picture lock so the pacing matches the cuts.

How do I keep the look consistent across dozens of generations?

Build a small style kit: three reference frames, a fixed colour description, a preferred lens and lighting phrase, and a saved prompt template. Feed the same references into every generation and grade the finished timeline as one piece.

Is fifteen seconds always better than a longer ad?

No. Fifteen seconds wins for awareness, offers, and retargeting. Longer formats still win for complex products and considered purchases. Use the short cut as the entry point and route interested viewers to a longer asset.

Which aspect ratios should I export?

At minimum 9:16, 4:5, and 16:9. If the team is stretched, prioritize 9:16 and 1:1, then crop deliberately instead of letting the platform decide.

Once the script and shot list exist, an afternoon for a single cut. Budget a day for generation and selection, half a day for editing and sound, plus time for claims review and variant exports.

Turn the Brief Into Motion

Fifteen seconds is unforgiving, but it is also the cheapest place to learn what your audience responds to. Write the promise, sketch the beats, generate more options than feels comfortable, and edit with the discipline of a thirty-second spot compressed into half the time.

Orelon is built for exactly this loop: cinematic ideas in motion, generated fast enough to test and refined enough to ship. Start with the AI video generator, borrow a starting structure from the video templates, and keep the Orelon blog open for prompt patterns and workflow refinements. Your next promo does not need a bigger budget — it needs more honest attempts at the first three seconds.