The “one hour” promise sells a draft, not a commercial
Generative models have learned a lot: from a text prompt or an image they create a short video clip with camera movement, light and weather. For advertising this opened up things that used to require a studio, 3D specialists or a trip to the mountains.
But a clip is not yet a commercial. An ad needs the hero to stay the same from the first shot to the last, the logo to match the brand book, and the message to land in fifteen to thirty seconds. These are exactly the requirements where quick generations stumble.
The road from first experiments to scenes with dialogue took three years — long enough for whole products not only to appear but to shut down.
Generative video: from silent seconds to scenes with sound
- Meta unveils Make-A-Video — short clips from a text prompt
- OpenAI shows the first Sora demos
- Sora opens to users: videos up to 20 seconds in 1080p
- Google Veo 3 — the first Veo model to generate video with sound and dialogue
- Sora 2: synchronised speech, sound effects and more believable physics
- OpenAI shuts down the Sora app
Meta AI, OpenAI, Google, TechInformed
The last line is the main practical takeaway. The app used in 2024 to make the first high-profile brand film lasted less than eighteen months. A process built around one model becomes obsolete with it. A process built around direction survives any change of tools.
Where AI already beats a shoot
- Characters that don’t exist. A talking object, a fairy-tale creature, a brand mascot — without puppets, costumes or months of 3D animation.
- Worlds and locations. Another season, mountains, a city of the future, an interior you don’t have to rent.
- Impossible camera work. Flying through an object, moving from macro to a wide shot, a move no crane could repeat.
- Testing the idea. An animatic built from generated frames shows the client the video before production — it is easier to agree on a picture than on words in a script.
- Versions. Different seasons, audiences and formats — 16:9 for TV, 9:16 for Reels, square for the feed.
Four places where AI video breaks
Character consistency and scene logic. The model does not remember the hero between shots. In one shot the mascot has two fingers, in the next three; clothing colour, proportions and facial features drift. For a brand this is critical: a character is the logo, only alive. Developers admit the weakness themselves — OpenAI listed the model’s limitations openly in its description of Sora.
It may struggle to simulate the physics of a complex scene, and may not comprehend specific instances of cause and effect (for example: a cookie might not show a mark after a character bites it).
Even showcase videos get it wrong. In one of the early Sora clips a train crosses the Glenfinnan Viaduct in Scotland. The real viaduct carries a single-track line; the model laid a second track on it. Anyone who has been there notices at once. A brand whose product or city gets “filled in” by the model like this will hear about it in the comments.

Text and logos. Lettering melts in generation, brand marks distort, packaging sprouts words that don’t exist. That is why captions, the logo and the packshot are added as graphics on top — just as in classic production.
Rights. A telling case is the comic “Zarya of the Dawn”. In September 2022 the U.S. Copyright Office registered it in full; in February 2023 it revised the decision: the text and the selection and arrangement of panels stayed protected, but the Midjourney images did not, because they “are not the product of human authorship”. In January 2025, after reviewing more than 10,000 comments, the Office set the approach out in a report: a prompt alone does not make a person an author; what is protected is the creative contribution — selection, arrangement, modification. For a business, this argues for a process where that contribution exists and is documented.

Labelling. In Kazakhstan the Law on Artificial Intelligence was signed on 17 November 2025 and came into force on 18 January 2026. Synthetic content is labelled twice: in machine-readable form — a tag inside the file — and with a clear notice for the viewer, such as “Created by AI”. Missing labels are fined; for large businesses a repeat violation costs up to 200 MCI (monthly calculation indices). Real people’s faces are used only with their consent.
Brand reputation. Viewers spot defects faster than you might think. In December 2025 McDonald’s pulled an AI-generated Christmas ad in the Netherlands after a wave of criticism. We look at why such campaigns put audiences off in a separate piece on why viewers criticise brands’ AI ads.
The fix is a process where AI is just one of the tools
A good AI commercial is built the same way as any advertising film. Only the way of getting the image changes — and the scale of selection. For the one-minute Toys"R"Us film, hundreds of iterations were cut down to a couple of dozen shots that still needed corrective visual effects. For Coca-Cola’s 2025 holiday ad the count ran into the tens of thousands.
The Decoder, TODAY, OpenAI
How a director-led AI commercial is built
- 1Idea and scriptwhat the viewer should understand and feel, which shot is the key one, how the video ends
- 2Style and referencescharacter sheet, locations, light, brand palette, and the hero’s objects created in advance as separate references
- 3Generationeach shot in several versions, based on frames from earlier successful generations
- 4Selectionthe director checks characters’ scale, object positions, face and clothing; defects are not cropped, the shot is generated again
- 5Compositing and graphicslogo, titles, packshot, colour and the AI label done by hand, following the brand book
- 6Edit and soundrhythm, music, voice, sound effects and adaptations for every platform
The most underrated stage is the first one. How to brief a contractor is covered in our guide on writing a brief for a commercial.
What it looks like in practice
“The Caring Router”, an autumn vertical video for almaPLUS: a router with eyes on its screen saves autumn from slow internet and cuts its own price. The character, the interiors and the mountain view were created with AI, while the edit, graphics, titles and sound are classic post-production. AI delivered what a camera cannot film — a living router — and everything to do with the brand and the message was made by hand.
almaPLUSThe Caring RouterA router with eyes on its screen saves autumn from slow internet and cuts the price itselfSee it in the portfolio
Long form tests the tool harder than advertising does. We made two short films with AI: “The Secret of the Forest Guardian”, about a wooden forest guardian, a sprout in his palms, a fire and a blue flower, and “While the Clock Ticks”, about an elderly man, a cryopreservation programme and life in a city of the future. In film you see whether AI can hold light, texture and the characters’ emotions through a whole story, not just one shot.
CaseFestival FilmsA forest guardian’s fairy tale and a story of timeSee the case
What to ask a contractor before you start
- Who directs the project, and who is responsible for selecting shots?
- How will the hero’s appearance and the scale of objects be kept consistent?
- How will the logo, captions and packshot get into the video?
- Where and how will the AI label appear — both the visible and the machine-readable one?
- Which source files and rights will you receive at the end?
If you are planning an AI commercial for a brand in Almaty or across Kazakhstan, start with the idea rather than the choice of model: models change every year, a good idea does not. You can see how this works across genres in the portfolio, and the calculator on the site will show the exact cost for your task — five to seven questions, one minute.
Frequently asked questions
Can a commercial be made entirely by AI?
The imagery, largely yes: characters, locations and camera moves are already generated, and newer models add sound and dialogue. But the script, shot selection, logo, titles, edit and sound mix are still done by people — otherwise the video comes out with defects.
Why does a character’s face change in AI video?
The model keeps no memory of the hero between shots and creates it anew every time. Consistency comes from references, a character sheet, repeated generations and strict selection.
Does AI advertising have to be labelled in Kazakhstan?
Yes. The Law on Artificial Intelligence has applied since 18 January 2026: synthetic content must be labelled in machine-readable form and carry a clear notice such as “Created by AI”. Violations are fined.
Who owns the rights to a video made with AI?
The U.S. Copyright Office’s position is that the human creative contribution — script, selection, arrangement, refinement — is protected, not the prompt itself. So rights and source files should be set out in the contract with your contractor.
When is AI video a better choice than a shoot?
When you need a character or world that does not exist, a difficult location, another season or many versions for different formats. For real people’s live emotions and product shots, a classic shoot is often more reliable.
Sources
- Meta AI — Introducing Make-A-Video, 29.09.2022
- OpenAI — Sora: Creating video from text (описание модели и её ограничений)
- OpenAI — Sora is here, 09.12.2024
- Google — Veo 3 and generative media models at I/O 2025, 20.05.2025
- OpenAI — Sora 2 is here, 30.09.2025
- TechInformed — OpenAI’s Sora shutdown adds to signs of a harder AI phase, 01.04.2026
- Wikipedia — Glenfinnan Viaduct
- U.S. Copyright Office — Zarya of the Dawn (Registration # VAu001480196), 21.02.2023
- U.S. Copyright Office — Copyright Office Releases Part 2 of Artificial Intelligence Report, 29.01.2025
- Zakon.kz — Токаев подписал закон «Об искусственном интеллекте», 17.11.2025
- NUR.KZ — Новый закон вступил в силу в Казахстане, 18.01.2026
- Tengrinews.kz — Крупные штрафы появятся в Казахстане за отсутствие ИИ-маркировки, 13.01.2026
- Informburo.kz — В каких случаях можно не подписывать, что к видео или фото применили ИИ, 23.01.2026
- Decrypt — Toys‘R’Us Ad Made With AI Ignites Controversy, 25.06.2024
- The Decoder — Coca-Cola once again turns to AI for its global “Holidays Are Coming” Christmas campaign, 04.11.2025
- TODAY — Coca-Cola Releases Another AI-Generated Holiday Ad, Despite Last Year’s Backlash, 2025
- BNN Bloomberg / AFP — Not lovin’ it: McDonald’s pulls Dutch AI Christmas ad, 10.12.2025


