The work begins with an eye trained over twenty years. Carnegie Mellon, fifty international architectural projects, the Hugh Ferris Memorial Prize. That training is the part of the process no tool can replicate, and it shows up in everything: the composition, the proportions, the way light is read, the decision about what stays and what goes.
The execution layer uses AI image generation. The specific models and platforms shift over time — the field is moving quickly, and the tools that produce the best results in a given month are not always the tools that produced the best results six months earlier. What stays constant is the way the work is directed: detailed prompts, refined across hundreds of iterations, filtered through a curatorial standard that rejects roughly six out of every ten outputs before anything reaches the library. The Lens — the way of seeing that holds the library together — is the thing being applied at every step. The tool is the brush. The eye is the work.
We use the language watercolor-inspired, never watercolor, in everything we publish about the illustrations. The work draws on the visual vocabulary of traditional watercolor painting, but it is not made with paint and water, and the people who do make work with paint and water deserve language that doesn’t blur the difference.
When a tool’s underlying training data is contested — and AI image models’ training data is contested — the most useful thing a working artist can do is run their own process with care. We have four standing commitments.
Clean prompts. Every prompt describes the visual qualities of the image: light, palette, subject, mood, composition, the texture of the medium. No prompt names a specific living artist, a branded property, a recognizable person, or an iconic copyrighted work as a reference. The work is built from visual language, not from pointing at someone else’s body of work and asking the tool to mimic it.
Multiple generations and active rejection. Every published image comes from a batch where most of the outputs were rejected. The selection step is the curatorial work, and it’s where the brand’s standard lives. Anything that looks suspiciously polished or unusually familiar is set aside rather than published.
Reverse image search before publication. Hero images and a sample of every collection are run through reverse image search before going live. This is a quality gate that catches both accidental similarity to existing work and any rare case where a tool has converged on something close to a memorized image from its training data.
Documented process. Prompts and raw generation files are kept as backup for every published asset. If a piece were ever questioned, the documentation is there.
These four commitments do not solve every concern about AI image generation. They solve the concerns we can solve through our own process, which is the part of this we can take responsibility for.
The training of large image models on scraped artist work, without consent and without compensation, is a real and unresolved problem. We don’t think this is settled. We don’t think it’s a non-issue. We think the artists who have raised it are right that something happened to their work that they didn’t agree to, and that the fact of public visibility on the open web is not the same as a license to ingest their work into a commercial system.
We would support the building of better infrastructure: opt-in training datasets, durable opt-out registries, attribution standards, transparency about what was used to train any given model. Some of this is being built now, late and imperfectly. We hope it gets better, and we will use the tools that are built on better foundations as they become available and capable.
We are not in a position to wait for that infrastructure to be perfect before working. Like every working artist using these tools today, we’re navigating a situation that the technology, the law, and the industry have not yet sorted out. We’re trying to navigate it with our eyes open.
The honest part. There is one question we cannot fully resolve, and we’d rather name it than pretend it isn’t there.
When a small business licenses a Monnaco illustration for a hotel’s brand identity, that’s a project a traditional illustrator might otherwise have been hired for. We can tell ourselves that some of our buyers would never have hired a traditional illustrator at any price — they would have used free stock or nothing at all — and for some real share of our buyers that’s true. But not for all of them. Some portion of what we do exists in a market where traditional illustrators also work, and our presence in that market changes it.
We don’t think we’re displacing illustrators on the strength of being cheaper. We’re making something with a specific sensibility and a curatorial standard that doesn’t compete with what a commissioned illustrator would do. But we’re honest enough to know the markets overlap, and the overlap matters.
The best thing we can offer is not a tidy resolution. It’s a commitment to operate the way a thoughtful participant in a forming market should: clear about what we are, careful about how we make the work, transparent about the questions we haven’t answered. We’d rather be a brand that names the difficulty than one that pretends it away.
This page exists because the question is reasonable, and because anyone evaluating The Collection for serious work — a hotel brand, a publisher, a creative director licensing for a campaign — deserves more than a marketing answer. If you have a question this page doesn’t answer, or you think we’ve gotten something wrong, write to ana@themonnacocollection.com. The conversation is part of how a position like this gets sharper over time.
— Ana Carolina