Where AI Images Actually Fit in an Agency: Concepts and Decks, Not the Hero Shot
By The Fellowi Team · · 7 min read
Most writing about AI in agencies is either a threat or a miracle. Here is the version from people who build the tool: it is very good at the part of the job where you need twenty ideas by Thursday, and it is not good at the part where the client's actual jar of face cream has to look exactly like the jar of face cream.
Knowing which side of that line a task falls on is the entire skill.
Where it earns its place
Concept rounds. The bottleneck in early creative is rarely talent, it is how many directions you can afford to show before someone has to commit. A generated round lets a team put six visual worlds in front of a client instead of two, and lets the two survivors be chosen for the right reasons rather than because they were the only ones with art.
Moodboards that are not other people's work. The traditional moodboard is stitched from photographers' portfolios and stock - which means the board looks better than anything you can actually shoot, and half of it is uncleared. A generated board is closer to what the campaign will really look like, and nobody has to be asked.
Social variation at volume. One approved concept, fifteen crops, four seasonal palettes, three formats. This is the least glamorous use and it is where the hours actually go.
Pitch work that may never be paid for. Speculative decks are where agencies burn money hardest. Making them cheaper to produce changes what you can afford to pitch for.
Where it still loses to a photographer
The product itself. Reference images hold a silhouette, a colourway and a mood convincingly - see generating from your own photo for how that works - but exact logo geometry, stitching, and where a label sits are precisely where generation still drifts. If a lawyer or a buyer would notice the difference, shoot it.
Anything claiming to be documentary. A real venue, a real event, a real staff photo. Passing generated imagery off as a photograph of something that happened is a credibility problem, not a technique problem.
Faces of real people. You need permission and the rights to any photograph you use as a reference. This is the constraint that catches agency work most often, because it feels like it has been solved and it has not.
The commercial terms, briefly
Anything generated can be used in client work, sold, or published. We take no share and require no credit, and there is no separate commercial tier to buy - the details and the three honest caveats are in selling what you generate. The caveat that matters most in an agency context: nobody can warrant that machine-generated output is protectable, so it should not be the thing you promise a client owns outright.
What it costs when the volume is real
Generation is priced in coins, and a failed generation refunds itself - the reasoning is in why the image API is priced in coins. The honest unit for an agency is not the cost per generation but the cost per approved asset, because the real number includes the rerolls. Budget three to five generations per keeper on concept work and you will not be surprised.
The larger coin packages exist for exactly this shape of buyer; the per-coin comparison across every tier is in which package is the best value, and current prices live on the product page.
Wiring it into a pipeline
The studio is fine for a person. A team producing variations at volume usually wants the public API: a key, a POST, a poll, a download, with fixed failure codes so a batch script knows what happened. Two practical notes. Generated files are retained for a limited window, so download and archive on your side - treat us as a renderer, not an asset library. And if you run work for several clients, a key per client keeps the billing conversation simple.
The pitch we would not make
We are not going to tell you this replaces your photographer, your retoucher or your art director. It replaces the twenty hours a week they spend making things nobody will ever approve. That is a smaller claim, and it is the one that survives contact with a real campaign.
Concept rounds run in Fellowi Images, and the ones that need motion for a feed go through Fellowi Video.
Questions this post answers
Can an agency use these images in paid client work?
Yes. Our terms place no restriction on commercial use and we take no share, so generated images can go into a deck, a campaign or a deliverable without a separate licence.
Can it reproduce a client's actual product faithfully?
Not reliably. Reference images hold a shape, a colourway and a mood well, but exact logo geometry, stitching and label placement are where generation still drifts - which is why we call it a concept tool, not a product-photography replacement.
What about using a real model's face?
You need that person's permission and the rights to use their photo, exactly as you would for any other use. Generating from someone's picture does not create a model release.
Keep reading
You Can Sell What You Generate Here. Most Generators Do Not Let You.
No share, no credit, no separate commercial licence. And three honest limits that no generator can write away for you.
Wiring an AI Image API Into an Agency Pipeline: Keys, Ledgers and Archives
The API docs tell you how to make one picture. An agency needs to make thousands for a dozen clients without losing track of who paid for what. Six rules from the way the API actually behaves.
Podcast Cover Art With AI: The Square That Has to Work at Thumbnail Size
Your show is judged at the size of a fingernail, in a grid of a hundred other squares. Five rejected drafts, what was wrong with each, and the rules they add up to.