Why AI Video Belongs in Your Promo Stack
Promo video has always been the most persuasive format on the internet, and also the most expensive one to produce. A single polished 30-second spot traditionally meant a scriptwriter, a crew, a location, an actor, a voice artist, an editor, a colorist, and a composer. That pipeline still produces the best results at the top end, but it prices out almost everything else: product updates, regional variants, A/B test cuts, localized versions, and the long tail of small campaigns that never justified a full shoot.
Generative video models changed the economics of that long tail. You can now take a written concept and turn it into usable footage in an afternoon, then iterate on it five more times before lunch tomorrow. The catch is that raw generation is not a workflow. Dropping a sentence into a text-to-video tool produces a clip, not a campaign. What separates teams that ship consistent, on-brand promos from teams that produce a folder of unrelated clips is process: a brief that survives generation, a storyboard that locks continuity, a sound pass, an edit built for multiple aspect ratios, and a measurement loop that tells you what to do next time.
This guide walks through that full process. It is tool-agnostic on purpose, because the models you use will change faster than the workflow will. The structure below works whether you are producing a SaaS launch film, a real-estate walkthrough, a fitness app promo, or a retail offer for paid social.
The End-to-End Workflow at a Glance
The seven stages
- Brief: the business goal, audience, offer, and constraints in one page.
- Script: narration or on-screen copy broken into beats with timings.
- Storyboard: a shot list with framing, action, duration, and continuity notes.
- Generation: images, clips, voice, and music produced or sourced against the storyboard.
- Assembly: rough cut, pacing pass, sound mix, captions.
- Delivery: exports for each channel and placement, with naming conventions.
- Measurement: performance review that feeds the next brief.
Most teams skip stages two and three and then wonder why the generated footage feels generic. The storyboard is where creative control actually lives. Once you have a shot list with durations and camera notes, generation becomes a fill-in-the-blank task rather than a slot machine.
Who does what
On a small team, one person can hold all seven stages, but it helps to name the responsibilities even if they sit on one desk. The strategist owns the brief and the offer. The writer owns the script and the beats. The art lead owns style references and continuity. The editor owns pacing, sound, and captions. The analyst owns the measurement loop. When a single person wears all hats, schedule them as separate passes on different days. Editing and scripting use different parts of your attention, and switching modes mid-task is where quality drops.
A realistic time budget
For a 30-second promo with six to ten generated shots, a first version is reasonable in one working day: an hour on the brief and script, an hour on storyboard and references, three hours generating and selecting takes, two hours editing and mixing, one hour on exports and captions. Revision rounds are faster because the assets already exist. Budget two versions per campaign and expect the second to take half the time of the first.
Writing a Brief and Script That Survive Generation
Generative models are literal. If your script says "we help businesses grow," you will get abstract footage of people shaking hands in slow motion. If it says "a warehouse manager scans a barcode at 6:40 a.m. and the dashboard turns green," you get something usable. The translation layer between marketing language and visual language is the single highest-leverage skill in this workflow.
Write the script in beats, not paragraphs. Each beat should contain one idea, one location, one subject, and one action. A 30-second promo typically holds five to eight beats of two to five seconds each, plus a three-second hook and a three-second call to action. Narration runs at roughly two and a half words per second at a comfortable pace, so a 30-second spot supports about 70 to 80 spoken words. On-screen text is tighter: four to six words per card, held long enough to read twice.
Separate what the viewer sees from what the viewer hears from what the viewer reads. Three columns is enough. Visual and audio can tell different halves of the same story — the picture shows the problem, the voice explains the outcome. This redundancy is what makes short promos feel dense without feeling rushed.
Finally, write the brief with production constraints baked in. Note the required aspect ratios, the maximum duration per placement, whether talent likeness is allowed, whether the product must appear in a specific state, and which claims are legally approved. Discovering a legal constraint after generation wastes an entire pass.
Storyboarding and Shot Planning
A storyboard for AI production does not need to be beautiful. It needs to be unambiguous. A simple table with columns for shot number, duration, framing, subject, action, and continuity notes will outperform a hand-drawn board that leaves questions open.
| Shot | Duration | Framing | Subject and action | Notes |
|---|---|---|---|---|
| 1 | 3s | Close-up | Phone screen lighting up at night | Dark room, warm lamp glow |
| 2 | 4s | Medium | Manager walking through warehouse aisle | Same jacket as shot 5 |
| 3 | 3s | Insert | Barcode scanner beeping | Handheld feel, slight motion blur |
| 4 | 5s | Wide | Team gathered around a wall dashboard | Same location as shot 2 |
Two details in that table matter more than the rest. First, the continuity column: notes about wardrobe, lighting, time of day, and props that must match between shots. Second, the framing column, because a close-up, a medium shot, and a wide shot generate differently and cut together differently.
Build an animatic before you generate anything expensive. Drop placeholder stills or even colored rectangles into a timeline with the narration track and watch it end to end. If the story does not hold up with rectangles, no amount of photorealism will save it. The animatic is also where you discover that a beat you loved in the script takes six seconds to read, not three.
Generating Consistent Visuals
Style locks and reference frames
Consistency starts with references. Pick two or three images that define your look — color palette, lighting direction, lens character, texture — and reuse them across every shot in the campaign. Describe them in words as well: "soft overcast daylight, muted teal and sand palette, 35mm lens, shallow depth of field." When a model supports image-to-video or reference-guided generation, feed the same reference into every prompt in the sequence.
Character and product consistency
Recurring people are the hardest element to keep stable. Three tactics help. Keep characters in consistent wardrobe and framing across shots. Avoid extreme angle changes between consecutive shots of the same person. And generate a small library of approved character frames early, then build each new shot from the approved frame rather than from text alone. For products, generate hero stills first and treat them as the canonical asset; virtual camera moves on a canonical still are far more reliable than reimagining the product in every prompt.
Camera language and motion
The most common generation failure is motion that looks like it is drifting rather than being driven. Specify camera behavior explicitly: slow push in, static locked-off frame, handheld follow, overhead crane down. Short, simple camera moves read as intentional. Long moves that cross rooms or change subjects usually break. When in doubt, generate a static shot and add motion in the edit with a scale-and-position move.
Diagnosing bad generations
When a clip comes back wrong, change one variable at a time. If the composition is wrong, adjust the framing description. If the motion is wrong, adjust the camera note. If the texture is wrong, adjust the style reference. Rewriting the whole prompt at once teaches you nothing about which phrase caused the improvement.
Voice, Music, and Sound Design
Sound is where most AI-assisted promos give themselves away. A generic synthetic voice over a generic music bed sounds exactly as cheap as it costs. Three improvements fix most of it.
First, pace the read. Synthetic voices default to a flat, even cadence. Insert commas, break long sentences, and generate the narration in short segments you can nudge independently. Thirty seconds of narration is easier to control as six five-second clips than as one file.
Second, treat the music as a pacing tool, not wallpaper. Choose a track with a clear downbeat and cut your shots to it. A four-second shot landing on a beat feels intentional; the same shot landing mid-phrase feels loose.
Third, build a small sound-effects layer. A whoosh on a transition, a subtle click on a UI moment, a low hum under a data shot. Six to ten effects across a 30-second spot will do more for perceived production value than doubling the resolution of your footage.
For mixing, keep dialogue or narration clearly on top, music ducked beneath it, and check the final loudness against your platform's target — many social platforms normalize to around -14 LUFS, so mixing much louder than that simply gets turned down. Always preview on a phone speaker. That is where most of your audience will hear it.
Editing and Multi-Format Delivery
The first three seconds carry the whole promo. Open on motion, a striking frame, or a question the viewer already has. Do not open on a logo, a title card, or a slow establishing shot of a building. Save the branding for the last two seconds and the end card.
Structure the timeline in layers: picture, narration, music, effects, captions, graphics. Keep captions as a separate layer rather than burned into the master so you can toggle them per platform. Most social placements benefit from burned-in captions because a large share of views happen with sound off, while website embeds often look cleaner with an optional subtitle track.
Export a master at your highest working resolution and native aspect ratio, then derive the rest. Vertical 9:16 for short-form social, 1:1 for feed placements, 16:9 for web and pre-roll. Do not simply crop. Reframe: recompose each shot so the subject stays in a natural position within the new frame. Safe areas matter too — keep text and key action away from the very top and bottom of vertical exports, where platform interfaces overlap.
Use a naming convention that survives a shared drive: campaign, version, ratio, duration, date-free revision letter. Version control prevents the classic mistake of publishing an old cut because nobody could tell which file was newest.
Analytics: Closing the Loop
Metrics that actually inform creative decisions
Reach and views are vanity numbers for creative purposes. The useful ones are hook rate (the share of viewers still watching at three seconds), hold rate (average watch time as a share of duration), completion rate, click-through rate, and cost per acquisition where you are running paid distribution. Each maps to a specific creative decision: hook rate tests your opening shot, hold rate tests your pacing, completion rate tests your length and payoff, click-through tests your call to action.
A simple testing framework
Change one thing per test. Common high-value tests include opening shot (product close-up versus human face), narration versus text-only, duration (15 versus 30 seconds), and call-to-action wording. Run each variant long enough to gather a meaningful sample, and resist declaring a winner after a few hundred impressions.
Feeding results back into the brief
The measurement loop is only worth building if it changes the next brief. Keep a living document of what worked: winning hook types, ideal duration for each placement, phrases that drove clicks, and shots that consistently underperformed. Ten campaigns in, this document becomes your real competitive advantage — far more valuable than any single model upgrade.
Tool Categories, Selection Criteria, and Common Mistakes
The categories you will need
Think in categories rather than brands, because the leaders change quarterly. You need a script and ideation assistant, a still-image generator for style frames and hero assets, a video generation model for motion shots, a voice tool for narration, a music and sound library, a nonlinear editor, a captioning tool, and an analytics dashboard. Many suites bundle several of these; bundling saves time but rarely wins on every axis.
Selection criteria
Evaluate each tool against six questions. Does it support the aspect ratios you need natively? Can you seed it with reference images for consistency? What are its commercial usage terms for generated output? How fast is a typical render at your working resolution? Does it export clean files that drop into your editor without transcoding? And does it handle your data responsibly — meaning your unreleased product footage and scripts are not used to train models or exposed to other users.
Mistakes that cost the most time
Chasing photorealism before nailing the story. Generating 40 clips before locking a storyboard. Using a different style reference for every shot and hoping the edit hides it. Ignoring aspect ratio until after the master is finished. Skipping the sound pass because the picture looks good. And publishing without a legal or brand review on claims, likeness, and music rights. Each of these costs a rebuild rather than a tweak.
FAQ
How long should a business promo be?
Fifteen to thirty seconds for paid social, thirty to sixty for website hero and pre-roll, and up to ninety seconds only when you have a genuine narrative or a complex product to explain. Shorter cuts almost always outperform longer ones on cold audiences.
Can I use AI-generated footage in paid advertising?
In most cases yes, provided you follow the platform's disclosure rules for synthetic media and you have cleared music, voice, and any recognizable likeness. Read the usage terms of each generation tool you rely on, since some restrict certain commercial categories.
How do I keep characters consistent across shots?
Generate a small set of approved character frames first, keep wardrobe and lighting fixed in your descriptions, avoid dramatic angle changes between consecutive shots, and build each new shot from an approved reference rather than from text alone.
Do I still need a human editor?
For anything customer-facing, yes — or at minimum a human editing pass. Models generate material; pacing, rhythm, and the decision about which two seconds of a six-second clip to use are still editorial judgments.
What is the fastest way to improve a promo that underperforms?
Test the first three seconds first. Hook rate is usually the biggest lever, and reshooting or re-cutting only the opening is far cheaper than rebuilding the whole spot.
How many variants should I produce per campaign?
Three to five is a practical range: one narrative-led cut, one product-led cut, one text-heavy cut for sound-off viewing, and one short cut for placements with tight duration limits. More variants only help if you can measure them.
