Two years ago, AI in marketing meant a chatbot writing captions nobody read. In 2026 it means we can produce a one-minute cinematic commercial for a snack brand, with the look of a feature-film trailer, from a room in Adma, in days. We did exactly that for Snips. It also means half the brands in the region are about to flood their feeds with the same shiny, weightless footage, and the other half will be terrified of the technology and wait. Both halves are wrong. This is what we have learned actually works, from inside the work.
What changed, in plain language
The models that generate video crossed a line in the last eighteen months. The shots hold together. Characters keep their faces from one scene to the next. Camera moves can be directed rather than hoped for. Sound can be generated with the picture. Image models can now take a real product shot and place it, correctly lit, into a scene that never existed, and keep the product looking like the product.
For our own work we use a stack that changes every quarter. Today it means keyframes built with image models such as Google’s Nano Banana to lock the product and the look, video generated with Seedance from those keyframes, voice and sound from dedicated audio models, and a human edit at the end that no model has replaced. The point is not the tool names. By the time you read this, some of them will have changed. The point is the pipeline: strategy, script, controlled stills, controlled motion, human edit. Skip the first two steps and the last three produce beautiful noise.
Where the savings are real
Testing ideas. A traditional shoot forces you to bet the whole budget on one idea. With generative production, a brand can see three directions moving before it commits to one. The cost of being wrong drops, which means the appetite for being bold goes up.
Looking like a national brand at a regional budget. For Marquise Jewelry, we produced photography at the standard of a studio campaign, with models, sets and light, without the studio bill. That is not a trick. It is the same craft, art direction and retouching judgment, applied to a new kind of camera.
Speed across markets. A brand present in Lebanon, the UAE and Saudi Arabia can now adapt a film for three markets, in two languages, with local details, in the time it used to take to schedule one shoot. For companies with the Gulf in their plans, that is the single most useful thing AI production offers.
Volume where volume matters. Product content for e-commerce, variations for paid media testing, seasonal refreshes. The work that used to be done badly because there was never enough budget can now be done properly.
Where it is a trap
We will be as candid about the failures as about the wins, because the failures are where a client’s money goes.
- Text. Models still drift on written words. In one of our own Seedance renders, the pack copy shifted slightly despite anchored keyframes. We caught it because someone was looking. The rule now: on-pack text is locked with reference stills and checked frame by frame, or it is shot for real.
- Brand consistency. Left alone, generative tools produce generic beauty: the same golden light, the same slow-motion glamour, the same faces. A brand’s distinctive assets, the colour, the mark, the typography, the specific way it speaks, have to be imposed on the model by people who know the brand. Otherwise you are paying to look like everyone else, faster.
- Rights and disclosure. Platforms increasingly label AI-generated media and audiences increasingly notice it. We disclose, we keep records of what was generated and from what, and we do not use anyone’s likeness or a competitor’s work as a reference. A brand that is caught faking is a brand that has to rebuild trust from zero, and trust is the slowest asset to grow.
- Strategy. No model knows your customer, your margin, your branch that is under-trading, or the price your market will bear. AI produces. It does not decide. The companies getting hurt are the ones who let the ease of production substitute for the discipline of strategy. Rory Sutherland’s line is the best warning we know against letting the tool set the agenda:
The opposite of a good idea can also be a good idea.
Rory Sutherland, Alchemy (2019)
Sutherland means that logic alone does not find the best answer in marketing; judgment does. A model gives you the average of everything that was ever made. The average never built a brand.
The five rules we work by
- Strategy first, always. Every AI production starts with the same document as a traditional one: who it is for, what it must make them think, what it must make them do, and how we will know.
- Real assets in, real assets out. Real product photography, the real logo, real colours as references. The model fills the world around the truth; it does not invent the truth.
- A director in the room. Someone who has shot the traditional way decides the frame, the pace and the cut. Prompting is not directing.
- Check it like a lawyer. Text, hands, logos, reflections, cultural details, Arabic typography. Every frame. Then a second person.
- Disclose and document. What was generated, with what, from what. The client owns the record.
What this means for a CEO in Beirut, Dubai or Riyadh
You will be pitched AI by everyone this year, mostly as a discount. Treat it as a capability instead, and ask the same questions you would ask of any production partner: show me the strategy this served, show me the numbers it moved, show me a failure and what you learned. Philip Kotler’s old warning has never been more precise:
Marketing takes a day to learn. Unfortunately, it takes a lifetime to master.
Philip Kotler
AI compressed the day. It did nothing to the lifetime. The agencies worth hiring in 2026 are the ones who were already good at the lifetime part, and who picked up the new tools with the same standards they held for the old ones. That is how we run our studio and its AI film arm, Attentionplease.ai: business first, brand second, technology third, and all three in the same room.
Frequently asked questions
Is AI-generated content allowed in advertising on Meta and other platforms?
Yes, with conditions that keep tightening. Platforms label realistic AI media, some ad categories require explicit disclosure, and every platform bans deceptive use of real people’s likeness. We disclose by default and keep a production record for each asset so a client can answer any question about how it was made.
Will AI production replace our agency or our in-house team?
It replaces a share of production hours, not judgment. Strategy, positioning, offers, media planning, sales alignment and the taste that makes work distinctive are unchanged. What changes is how much of the budget goes to making versus thinking, and in most companies that ratio badly needs to move toward thinking.
How fast can an AI commercial be produced?
From an approved script, days rather than weeks, with most of the time spent on direction, review and correction rather than rendering. The script and the strategy behind it take as long as they always did, and skipping them is how AI commercials end up looking like everyone else’s.
Does AI production work for Arabic-language campaigns in the Gulf?
It does, with extra care. Arabic typography and dialect are exactly where models make the mistakes an audience notices first, so text is set by people, voice is checked by native speakers of the target market, and cultural details are reviewed frame by frame before anything ships.
