For as long as advertising has existed, making a film has meant booking a place, a crew and a day. Everything else in a campaign, the idea, the casting, the media plan, has been arranged around that fixed, expensive event.
That assumption is now breaking. Not gradually, and not only at the cheap end of the market. The cost of producing a broadcast quality film has fallen by something close to two orders of magnitude in under three years, and the brands proving it are not experimental startups. They are the largest advertisers in the world.
The useful question is no longer whether AI will change brand production. It is what specifically changes, and in what order. Here are the five shifts that matter.
1. The set disappears before the actor does
The first misreading of AI advertising is that it replaces people. In practice, what brands have actually replaced first is the place.
Coca Cola rebuilt its "Holidays Are Coming" campaign using generative tools to create wintry, atmospheric cityscapes that were never built or filmed. Volvo produced fully AI generated spots specifically to localise campaigns for regional markets such as Saudi Arabia. Toys "R" Us made an origin story film with no location at all.
Look at what those have in common. The environments are synthetic. The creative idea, the brand strategy and the human direction remain entirely human. Adobe reports that 81% of creative professionals already use generative tools, with 62% cutting task time by around 20%. That is not replacement, it is compression.
For a campaign involving a celebrity, this is the most immediately relevant shift, because the set is usually the reason a shoot has to happen at a particular place on a particular day.
2. The shoot becomes a capture
Once environments are generated, the nature of the star's involvement changes fundamentally.
A traditional shoot requires the talent present for the duration of production. A capture session requires them present once, briefly, to record the raw material from which everything else is built. H&M demonstrated this at scale in July 2025, launching campaigns featuring AI generated digital twins of 30 human models, allowing consistent on brand visuals without scheduling a physical shoot for every collection update.
The economic consequence is significant. A traditional celebrity campaign in India runs anywhere from ₹20 lakh to ₹5 crore once production and talent fees are combined. Remove the set, the crew, the location, the travel and the reshoots, and what remains is the talent fee and the assembly cost.
But the strategic consequence matters more than the economic one. When a star's contribution is a capture rather than a shoot, their availability stops being the constraint that shapes the entire campaign calendar.
3. One film becomes many
This is the shift most brands underestimate, and it is where the real value sits.
When production requires a shoot, a brand gets one film and perhaps a handful of cuts. When it does not, the same capture produces dozens of variants: different lengths, offers, regions and languages, tested continuously against each other. Indian practitioners now talk in terms of generating thirty unique video variants a week simply to keep feeding the Meta and YouTube algorithms.
That cadence is not a cheaper version of the old model. It is a different model. Creative fatigue in performance marketing now sets in within days rather than months, and only a production pipeline without a camera can refresh at that speed.
For India specifically, this is the decisive advantage. A skincare brand talking to a customer in Ludhiana needs different linguistic and cultural nuance than one talking to a customer in Coimbatore. Producing twenty versions of a product demo with human actors is a logistical impossibility for most brands. Producing them from a single capture is routine. Deloitte's 2026 TMT predictions anticipate India's virtual influencer market growing an additional 15 to 20% annually, and vernacular reach is the reason why.
4. The audience starts judging why, not how
None of this works if people reject it, and there is real evidence they might. Consumer sentiment toward AI advertising has moved in the wrong direction, with roughly 12% more consumers reporting negative feelings than in 2024 and Gen Z showing around 39% negative sentiment.
But the backlash is not indiscriminate, and the pattern in it is the single most useful thing a brand can understand right now.
When Guess ran an AI generated model in Vogue's August 2025 print issue, readers announced subscription cancellations within hours and working models objected publicly. Coca Cola's AI holiday work drew heavy criticism. Yet an unofficial Liquid Death spot made with Veo 3 for roughly $800 in credits spread widely and landed well. Airbnb's AI assisted spots provoked curiosity rather than anger, and the brand leaned into the debate directly.
The distinction is not whether AI was used. It is whether AI appears to have replaced a human who should have been there. A synthetic model standing in for a working model reads as a job taken. A synthetic environment, or a star's own licensed avatar, does not.
This is why the next phase belongs to licensed likenesses rather than invented ones. An AI avatar of a star who consented, approved the campaign and was paid is not a person being replaced. It is a person working, without the Tuesday and without the flight.
5. The face becomes a licence, not a booking
The final shift is commercial and legal rather than creative, and it is the one that decides who actually wins.
Brands have always bought a celebrity's time. What they will increasingly buy is a defined right: a specific campaign, in a specific category, for a specific duration, across specific territories. The Indian market has already begun pricing this. Celebrity fees for limited usage digital rights have moved into the ₹15 to 50 lakh range, while platform based licensing of pre cleared avatars starts far lower.
The compliance layer arrived faster than most marketing teams expected. In the United States, the FTC applies existing deceptive advertising and endorsement rules to AI content and in 2026 clarified a double disclosure requirement, meaning brands must disclose both a paid partnership and AI involvement where both apply. New York requires disclosure of AI generated synthetic performers. The EU AI Act adds its own requirements from August 2026.
In India, the amended IT Rules introduced the Synthetically Generated Information framework, requiring AI generated content to carry labels and embedded identifiers. ASCI's influencer guidelines require clear labelling, with a stricter obligation where an avatar is modelled on a real person. And across 2025 and 2026, Indian courts have repeatedly confirmed that a person's name, image, voice and likeness cannot be commercially exploited without consent, and that this protection extends to digital twins.
The practical consequence is blunt. An unlicensed AI likeness is not a creative shortcut, it is a legal liability. What a brand now needs is direct proof of consent from the celebrity or their estate, audit logs recording how and where the avatar was used, and proper governance around the underlying data.
What the next brand shoot actually looks like
Assemble the five shifts and the picture is clear.
A brand briefs a campaign. Instead of negotiating a shoot window against a star's film schedule, it licenses that star's avatar for a defined scope. The star reviews the brand, the category and the usage, and approves it. The films are assembled rather than shot: one capture, many environments, many languages. Every output carries its AI disclosure and provenance metadata by default. The star approves the finished work before it goes live. The licence expires on schedule.
Nothing in that sequence requires a set. Every part of it requires consent, scope and proof.
That is the real bottleneck, and it is worth being precise about it. The generation quality problem is largely solved. The cost case is proven. The localisation case is proven. What has not been solved at scale is the unglamorous part: a rights layer that lets a brand know, with certainty, that the face in its campaign is licensed, approved, disclosed and defensible if challenged.
The brands that win the next five years will not be the ones that adopted AI production earliest. They will be the ones that adopted it with clean rights, because the cost of a beautiful campaign built on an unlicensed face is not the production budget. It is the injunction.
Player Two operates a licensed AI avatar marketplace: consent backed talent, scoped campaign licensing, and production workflows built for brands that need their rights to hold up.