There is no button
tl;dr
AI creative isn't a button. Hybrid production is a dial the brand holds, the bottle is always real, and the questions that matter are what's real, what's generated and who's steering.
Somebody tested the button
Most arguments about AI creative assume a button. You type a sentence, press it, and a campaign falls out. People who love AI picture the button saving them a fortune. People who hate it picture the button putting them out of work. Both sides are arguing about the same imaginary machine.
This year somebody built the button and tested it. Ipsos and Syracuse University's Newhouse School took ten real ads from major brands and had AI remake each one with no human creative intervention. Gemini reverse-engineered the brief and wrote a shot list. Sora 2 generated the spot. Nobody touched the output. Then 3,000 consumers watched both versions.
The button held up better than the skeptics expected. Only a quarter of viewers were confident the AI ads were AI. But the human-made ads beat them by 14% on short-term sales impact and 17% on long-term brand health. Ipsos titled its report AI Ads Are Good Enough and spent the rest of it explaining why good enough loses. Its recommendation was to keep humans in the loop and build hybrid workflows.
That's the part of the conversation almost nobody has. Last week we wrote about what the budget cuts are taking out of the category in The outside line goes first. This is the other half: what the work actually is. Between the button and the camera there's a whole method, and it's where most serious creative work now happens.
Nothing is pure anymore
The second myth is that creative work comes in two flavors, AI and not. A brand asks for no AI and assumes it knows what it's asking for.
Start with the phone in your pocket. Every flagship smartphone shoots several frames for each picture and runs them through machine-learning models before you ever see the image. Brands that ask for authentic, unpolished content shot on a phone are asking for one of the most heavily processed images in modern photography. More machine learning touches a phone snapshot before anyone sees it than touches a raw file off our Hasselblad.
Then the edit. Photoshop has filled gaps with content-aware tools since 2010, and Generative Fill has been in the release version since 2023. Adobe reported that Firefly generations across its tools topped 16 billion by the end of 2024. Lightroom's AI Denoise cleans up high-ISO files that photographers used to throw away. A great deal of the retouched product photography you've seen in the last three years passed through one of these tools, whether or not anyone mentioned it.
Then the people approving the work. McKinsey's global survey found 71% of companies use generative AI in at least one business function, and marketing and sales is where they use it most. The brief, the copy, the deck: odds are decent some of it was drafted with help.
Pure is gone. Nobody had to announce it. The questions that matter now are where AI touched the work, how much, and who was steering.
What hybrid actually means
Hybrid production is the method between the button and the pure capture. Here's what it looks like in our studio, start to finish, for a single hero frame.
The reason. Before anything gets lit or generated, two directors decide what the image is for, who it has to move, and what it should make them feel.
The capture. The real bottle goes on set and gets lit by hand on a Hasselblad: backlight through diffusion, polarized key, strip lights for the edges, a spot for the label. Every final is a composite of several exposures, each one optimized for a different part of the glass.
The extension. Generative tools build the world around that capture: the bar, the light through the window, the season. The render is position-matched to the camera angle so the real bottle sits in it naturally.
The return. The real label, the real glass and the real liquid go back in by hand in Photoshop, because labels carry legal weight and models still invent typography.
The finish. A film grade, a second pair of eyes from someone who also came up behind the bar, and a disclosure label on delivery.
The generative step is one stage of five, and it only works because of the stages around it. We learned that the hard way while writing about digital twins. An image model can render the way light refracts through a filled bottle, flipping the background left to right inside the glass, but only when it's told exactly what the glass should do. Getting that right depends on someone who already knows what correct looks like. That knowledge came from ten years of lighting glass, and no model shipped with it.
Hybrid is faster traditional production. There's no magic in it. It takes fewer shoot days and more decisions per frame, and every one of those decisions belongs to a person.
Real is a setting
The biggest misunderstanding about hybrid work is that it's all or nothing. It works as a dial, and the brand holds it. The setting can change project by project and element by element.
The bottle and label: always real. Never generated as a final.
The drink: built and shot in the studio for hero frames, or extended and varied around a real capture.
The people: cast and shot, generated, or licensed from a real model. The brand's call, every time.
The setting: a location, a studio set, or built around a real capture. Chosen per brief.
Light, season, time of day: set on the day, or adjusted in the render. Chosen per brief.
A brand uneasy about synthetic people can say so, and the frame uses real ones. A brand that wants a full traditional shoot for the launch and generative extension for the always-on feed has a production plan. A studio that works this way can do the old method, the new method and every point between them. Ask yours where the dial is set on each project. If they can't tell you, that's the answer.
There's also a setting most brands haven't heard of yet. Cast a real model, shoot them in the studio, and license their likeness for generative work, paying them each time it runs. New York's synthetic performer law requires disclosure for digital humans who aren't recognizable as any identifiable real performer, so a recognizable, licensed person falls outside that definition. A separate New York statute governs the license: a digital replica clause needs a reasonably specific description of the intended uses, or a performer represented by counsel or a union. We covered the law in more depth here.
Done properly, the model gets paid every time their face works, the brand gets real people at generative speed, and the audience's sharpest objection, synthetic people, no longer applies. We'd still label it, because our disclosure standard was never waiting on a statute.
Where audiences draw the line
The dial matters because audiences do have a line. It just isn't where most memos put it.
The 2026 Bentley University-Gallup survey found 49% of Americans view businesses' use of AI in ads negatively. Read one level down and the picture changes. Seventy-five percent are fine with AI helping brainstorm or draft ads, and 53% accept AI making the final images. The objection concentrates on one thing: 62% say AI-generated people or voices are unacceptable, even when disclosed. Among adults 18 to 29, that rises to 73%.
The other line is honesty. In the Ipsos study, 79% of viewers said companies using AI should have to disclose it. Researchers at NYU Stern found fully AI-generated ads lifted click-through by as much as 19% over human-made ones, until they were labeled. The label alone cut click-through 31.5%. Audiences liked the work. They reacted to the word. Clicks and brand building aren't the same scoreboard, either: the Ipsos ads that lost were measured on sales impact and long-term brand health, where craft compounds.
Put those together and the brief writes itself for most brands today. Keep the product real. Use generative tools for the world around it. When the frame needs a face the audience has to trust, cast a real person or license one. Label the result honestly and let the work carry it. That’s a setting on the dial, and it’s available to any brand that asks for it. We publish our own disclosure standard, Rendered, Not Recorded, so nobody has to guess what was made how.
None of that puts synthetic people off limits. A generated cast is a legitimate setting when the idea calls for one: a character no one could cast, a world no crew could reach, a cast that has to stay consistent across every season of a campaign. Today it carries a cost, and the numbers above show what it is. A brand that chooses it should choose it on purpose, label it, and never let a synthetic face stand in for a real customer, a real bartender or a real testimonial. That line holds however sentiment moves. We think the broader objection will move, because every image technology has stood this trial and audiences keep settling on what they once refused. Nobody knows where the line sits in five years. Work built to hold up with the label on will be fine wherever it lands.
Photography stood this trial first
In his review of the Salon of 1859, Charles Baudelaire put photography on trial. The new medium fed the crowd's appetite for exact copies, he argued, and had no business near art. "If photography is permitted to stand in for art in some of its functions," he wrote, "it will soon have supplanted or corrupted it entirely."
The man who wrote that sat for Nadar's camera more than once. Anyone who knows Baudelaire's face knows it from photographs, Nadar's and Étienne Carjat's among them. Photography went on to become an art in its own right, and it carried his likeness further than any painting did.
Brian Eno described the same cycle in his 1995 diary, later published as A Year with Swollen Appendices. Whatever feels weird, ugly or nasty about a new medium, he argued, "will surely become its signature." Film grain was once a defect. Today we run a Kodak Vision3 emulation over frames that never touched film, because the grain is part of how the work feels.
Dorothea Lange said the camera teaches people how to see without a camera. The tool was never the one doing the seeing. That was true of her Graflex, it's true of our Hasselblad, and it's true of every model we run.
Who turns the dial
The debate about whether AI belongs in creative work has mostly been settled by the work itself. It's in the phone, the retouch, the brief and the approval chain. What hasn't been settled is who decides how it's used, and that's the conversation worth having with any studio, agency or in-house team.
Ask where the dial sits on each project. Ask what's real, what's generated and why. Ask who on the team knows what the glass is supposed to do before the model gets it wrong. Those answers will tell you more about the work than any label.
There is no button. There's a camera, a reason for every frame, and people with enough years behind them to know where the dial belongs.
FAQ
What is hybrid production?
Hybrid production starts with a real studio capture of the product and uses generative tools to build the scene around it. Human directors set the reason for each frame, real labels and glass are composited back by hand, and the brand decides which elements stay real.
Is AI-generated advertising just pushing a button?
No. In a 2026 Ipsos and Syracuse University test, fully automated AI ads with no human creative input underperformed human-made ads by 14% on short-term sales impact and 17% on long-term brand health. Ipsos recommended hybrid workflows that keep humans in the loop.
Can a brand choose which parts of an image are real?
Yes. In hybrid work the product, the drink, the people and the setting can each be captured for real or generated, project by project. Ford Media Lab keeps the bottle and label real in every final.
Can AI-generated advertising use real models?
Yes. A brand can cast and shoot a real model, then license that person's likeness for generative work with a contract that specifies the uses. Under New York General Business Law 396-b, a recognizable real performer is outside the definition of a synthetic performer, and New York General Obligations Law 5-302 sets contract requirements for digital replicas.
Do audiences reject AI in advertising?
Mostly they reject synthetic people and undisclosed use. In the 2026 Bentley-Gallup survey, 75% accepted AI for brainstorming and drafts and 53% for final images, while 62% rejected AI-generated people or voices.
by // Kyle Ford
