This is a casting question, not an ideological one.
The conversation about AI models tends to arrive pre-loaded — either it is the end of creative work or it is the obvious future and everyone else is behind. Neither framing helps a brand decide what to do this quarter.
The useful question is narrower: for this specific job, which produces the better result? Sometimes the answer is a generated model. Frequently it is a person. Often it is both, doing different work.
Here is how we make that call.
Where AI models win
Volume and consistency
When you need forty variations of the same visual across sizes, languages, and placements, an AI model produces them with a consistency no repeat booking can match. The same face, the same styling, the same lighting, across a campaign that runs for months.
Speed against a deadline
A shoot has a floor on how fast it can happen: booking, scheduling, location, crew, post. When a launch date moves up or a campaign needs a new market's assets in four days, generated visuals compress a three-week production into a short one.
Scenarios that are expensive or impossible to shoot
A location you cannot access, a season that is not happening, a product that does not physically exist yet, a set that would cost more than the campaign. This is where the economics are not close.
Catalog and utility imagery
Product-in-context shots, size and variant imagery, background lifestyle work. High volume, low emotional weight, and the audience is not looking for a person to relate to.
Representation across a wide audience
Showing a product across many skin tones, ages, and body types is expensive to cast properly and frequently gets done badly or not at all. Generated imagery makes breadth achievable — provided the intent behind it is real and not a substitute for hiring diversely everywhere else.
Where human talent wins, decisively
Trust and recommendation
The entire mechanism of creator marketing is that a real person with a real reputation is putting it on the line. A generated figure cannot vouch for anything, because it has nothing to lose. For a recommendation-driven purchase, this is not a small gap — it is the whole product.
Anything that happened
Events, service, food, atmosphere, the room on a Friday night. If the content is evidence that something is real, generating it defeats the purpose and, if discovered, converts your best asset into a liability.
Community
Creators bring audiences, comment sections, replies, and their own judgment about what will land. You are buying a relationship, not an image.
Culture and local credibility
In a city like Miami, being demonstrably of the place is a real differentiator. A generated model can be styled to look local. It cannot be local, and the local audience knows the difference faster than you would expect.
Anything with an implied endorsement
Testimonials, reviews, before-and-afters, results claims. A generated person delivering a testimonial is not a creative choice. It is a fabricated endorsement, and it is the fastest way to lose a category's trust permanently.
The disclosure rule we work to
Our position is simple and we apply it consistently: if a reasonable viewer could mistake generated imagery for a photograph of a real person, say so.
Practically:
- Disclose in the caption or on the asset — not buried in a footer nobody reads
- Never generate a person delivering a testimonial, review, or results claim
- Never generate imagery of a real, identifiable person without their written agreement
- Never present generated content as documentation of a real event, room, or service
- Keep human creators disclosed as partners, exactly as advertising rules already require
Disclosure costs almost nothing in performance and removes essentially all of the downside risk. Brands that skip it are trading a rounding error in engagement against a category-level trust problem.
How they work together
The most effective structure we run is not a choice between them — it is a division of labor:
- AI models carry the campaign surface. Hero imagery, paid variations, seasonal refreshes, the assets that need to be identical across thirty placements.
- Human creators carry the proof. The unboxing, the review, the walkthrough, the in-room content, the recommendation.
- Owned production carries the truth. Your actual product, space, team, and service, shot properly, once a quarter.
Budget generally follows the same split: the largest share to the layer that carries trust, the efficiency gains taken where volume lives.
Four questions before you generate anything
- Is this asset a claim? If it asserts that something happened or that someone endorses you, do not generate it.
- Would a customer feel deceived if they learned how it was made? If the honest answer is yes, disclosure will not save it — the concept is wrong.
- Is a real person available and better? Often the honest answer is yes, and the only advantage of generating is that it is easier internally.
- Are we replacing work we should be paying for? Using generated imagery to avoid paying creators for work they should be doing is a short-term saving with a long-term reputation cost.
The bottom line
AI models are a production tool with a specific and genuinely useful set of applications, and a clear line past which they damage the thing they were meant to help. Use them for volume, speed, and scenarios you cannot shoot. Use people for trust, proof, and community. Disclose either way.
For how we build and deploy AI models for brands, read this. For the human side, our talent roster and creator roster strategy cover it.
Trying to work out which side of the line your campaign sits on? Let's talk.