Generative AI Is Being Judged Against the Wrong Standard

The standard is concealment, framed as realism. Can the work pass as something AI was not used to produce?

That benchmark is derived from the doomsday narrative that landed before the tools did: AI as an instrument of mass job replacement, as substitution of human participation. As it applies to generative AI, this benchmark is responsible for much of the confusion and hesitancy surrounding the technology and its place in the creative process.

Of course, realism matters. But chasing the standard of concealment confines integration to those use cases where the tools are most unstable and turns every viewing into an audit, a search for the tell, conducted against work the technology cannot consistently deliver.

The guiding benchmark should be coherence, whether the imagery holds to the terms established by the story and the brand.

Every piece of creative work establishes a contract with its audience. Viewers judge the image against the story, the purpose of the ad, the identity of the brand, and what they understand all of it to represent. A generated image that is visually flawless can violate that contract while a visibly fantastic character, environment, or event can work completely if it belongs within the world of the ad. The audience does not need to mistake the frame for captured reality. It needs to understand what it is seeing and accept it on the terms established by the broader work and the brand.

The Switch

Under coherence the focus shifts from where the technology can hide, to where it can amplify. A brand asking where AI can go unnoticed is limited to substitution. A brand asking where it can extend the message is looking at everything the technology actually does well.

Where the Technology Is Unstable

The tools have real limits, and they are worth stating plainly.

The first is people. Holding the range of human expression across a sustained sequence remains the most difficult thing these platforms do. It gets harder when the audience has a reference point, because a recognizable face arrives with everything a viewer already knows about how it moves.

The second limitation is time. The most capable models can produce up to fifteen seconds of video, but stability across those seconds is not guaranteed. Nothing carries from one generation to the next. Each take begins from nothing, so a material reads one way in one shot and differently in the next, and anything of length has to be assembled rather than captured. We wrote about what that process actually involves earlier this year.

Both limitations sit exactly where the concealment strategy points a brand. That is the trap. The standard sends companies toward the two things the technology does most unreliably, then invites them to judge the whole medium by the result.

Knowing How Is the Work

We recently produced a series of films for an industrial manufacturer, each running one to two minutes, assembled from clips of a few seconds each. Environments change across those cuts. Scales change. Lighting conditions change. Every element has to hold anyway. The palette. The way a surface reads in a wide shot and again in a close. The tone, because the client's identity is what the film is carrying.

Nothing in the technology does that by itself. The tools generate. They do not decide what belongs, what to attempt, what to abandon, or how a company should sound. That is taste and judgment, applied through iteration, and it is the entire job.

This is where the idea of AI production as a commodity falls apart. Anyone can generate a clip. Producing a film that coheres across environments and cuts, inside a brand's constraints, is a skilled trade.

The Market Is Already Moving

The Interactive Advertising Bureau reported in July that nearly two-in-three video ad buyers now use generative AI for digital video creative, up from half in 2025. One-third of their ad assets will use it this year, up from one-fourth in 2025, projected to reach 43% by 2027. The economics are obvious and compelling. They are also not the explanation. No brand keeps using something that hurts it because it costs less.

Adoption is rising because fluency is rising. Companies are working through the process, learning where the technology holds and where it does not, and making the switch from concealment to coherence.

We See This Progression Firsthand

We produce work for companies across a wide range of industries. Industrial and pharmaceutical businesses. Consumer brands, fashion companies, creative studios, SaaS platforms, film production.

Most businesses take the first meeting for the cost savings. They stay for what our production model makes possible. A company comes in wanting to produce its existing work for less, and leaves with work it had ruled out as impractical to produce on any recurring basis.

Working through the process is what changes the posture. Every stage runs on human judgment, and once a company sees that, it stops asking whether the work can conceal how it was made and starts focusing on what becomes possible.

The strongest hesitation we've encountered comes from customer-facing luxury and high-end brands. Their sensitivity is understandable. The market is saturated with the cheapest output the technology produces, and in our experience that output is often used to define the whole medium. The sentiment shifts once these brands see the process and the work behind it. Arnault's public embrace of AI as an assistive tool and his efforts to integrate it across the LVMH creative process have helped move the segment. We wrote about it in July. Companies that hesitated when we first met have started experimenting on their own. Many have re-engaged with us to do work.

Where This Leaves Brands

Generative AI is not the first technology to face this test. New technologies are not adopted on arrival. They are adopted through iteration, as creators learn what the tools can carry and audiences learn what to make of the results. CGI ran that course. Audiences accept it now without thinking about it. That credibility wasn't established by using the technology to trick anyone into mistaking rendered imagery for captured reality. It was earned as creators became skilled enough to make impossible images cohere within the imaginary worlds they built. Bad CGI still gets noticed. It stopped defining the medium.

Generative AI is earlier in that progression. Its cheapest and least disciplined output remains the most visible, so the floor of the medium is still mistaken for its ceiling. That gap is closing as better work reaches the market.

Adoption will follow the same course. Brands are learning what integration actually looks like, and what opens up when they work with creators who know how to navigate the technology. That is what moves a company from hesitation to production: not a lower cost for the work it already makes, but a path to the work it had ruled out.

Realism is not the standard. Coherence is.