Eugène: a dreamlike Paris, rare voice in AI art

Working under the name Eugène, this Paris-based AI illustrator and creative director has built one of the most instantly recognizable visual worlds in generative art today, an idealized, geometrized version of Paris, all Haussmannian architecture, bold pattern and near-total stillness, rendered with a warmth and consistency that feels almost impossible to sustain across hundreds of images.

But the images are only half the reason this interview is worth reading. Behind them is two decades of experience directing campaigns long before generative AI existed, and a way of thinking about images that owes more to print culture, cinema and art direction than to any AI tool. That background shows in how he talks: original, occasionally surprising, never repeating the usual lines about AI and creativity, and consistently more instructive than most conversations in this space.

Eugène: a dreamlike Paris, rare voice in AI art

We talked with Eugène about where that visual consistency actually comes from, what shaped his eye long before AI entered the picture, and what he’d tell an AI illustrator or artist just starting to build a visual language of their own…

Can you introduce yourself?

Before prompts, before pixels became my main material, I grew up inside printed images.

My father was a photoengraver, so magazines, books, proofs, paper, ink, grain, contrast and color separations were part of my early visual landscape. I do not come from prompts. I come from printed matter, visual culture and art direction.

That still matters.

My illustration work is published under the name Eugène, which is my middle name. I am a French AI illustrator, creative director and designer, based near Paris.

Before working with generative AI, I spent more than twenty years in creative agencies, leading campaigns and visual projects for brands such as Dior, L’Oréal, Samsung, Paco Rabanne, AXA, Disneyland Paris. That background still shapes everything I do. I come from advertising, luxury, brand storytelling and visual direction, not from pure technology.

In a way, I was building creative systems before I had AI tools. In 2000, I co-founded Computerlove, a curated platform for digital design. In 2009, I worked on the Disney Toonificator for Disneyland Paris, an online experience that used facial recognition and machine learning to transform visitors into Disney-style characters. Fifteen years later, I am still exploring the same idea: using machines to extend visual storytelling.

I started using generative AI intensively in 2023. Today, under the name Eugène, I work both as an AI illustrator and as the founder of Eugène Studio, a generative studio based near Paris. The two practices are connected: the illustration work proves the visual signature, while the studio work turns that signature into systems, workflows and tools that can scale.

My illustration work is represented by Senso Art, with commissions in editorial, luxury, institutional and campaign contexts. I have worked for clients such as Connaissance des Arts, Pentagram NYC, Éditions du Seuil, La Banque Postale, Saint-Gobain, Coloradan Magazine.

I don’t prompt anymore. I specify.

At first, like everyone, I prompted. Now I build contexts, visual ethos, constraints, references, negative rules and production systems. I use tools like OpenAI Codex, Hermes and Claude to structure visual worlds and produce coherent image series, not just isolated lucky images.

My workflow has four layers: taste, constraints, systems and curation. Taste defines what belongs to the world. Constraints protect the style. Systems make the work repeatable. Curation keeps the human eye in charge.

Through Eugène Studio, I also build tools and agents. For example, I work with supervised editorial agents: small creative systems that can research a subject, propose narrative angles, generate visual directions, and then leave the final decision to human curation.

The system proposes. The human chooses.

I also create practical tools like Light Lab and Expression Lab. Instead of asking an AI for “cinematic light” or “a subtle emotion”, I try to make the intention visible and controllable: sun direction, camera settings, facial action units, constraints, negative rules. I don’t want the model to guess my taste. I want to give it a structure it can work inside.

And through CreativeAI.fr, I train creative and marketing teams to integrate generative AI into real production workflows. Since 2023, CreativeAI has trained more than 600 professionals, with over 3000 hours of training delivered.

A beautiful AI image is easy. A recognizable visual language is rare.

Where do you find your inspiration?

My visual world is fed by painting, cinema, editorial illustration, architecture, Paris, color, awkward human moments and slightly impossible atmospheres.

I am drawn to images that feel constructed but alive. Scenes with tension. Characters who seem to step out of a film that does not exist yet. Colors that feel slightly too intense to be realistic, but too precise to be random.

I like images that borrow from both painting and cinema. I am influenced by non-AI artists, of course. That is essential. AI should not become its own closed aesthetic loop.

I am interested in image-makers who understood that an image is not just something beautiful, but a small system of tension: Guy Bourdin for color and narrative ambiguity, Saul Steinberg for visual intelligence, Tomi Ungerer for irony, Lorenzo Mattotti for pictorial intensity, Alexey Brodovitch for editorial rhythm, and Jacques Tati for the choreography of people inside architecture.

I also look at magazine culture: old issues of Graphis, Twen, Harper’s Bazaar, Vogue Paris, book covers, editorial layouts, old advertising, cinema frames, interiors and street photography.

The references are often less about style than attitude. I organize research as a living visual archive: moods, colors, compositions, gestures, textures, printing accidents, characters, interiors, architectural details and small obsessions. I don’t use references to copy them. I use them to understand what kind of tension an image can hold.

Sometimes an idea comes from something very indirect: a badly lit café, a strange coat in the street, an old printed page, a hotel lobby, a corporate meeting that suddenly feels like surrealist theatre.

One unexpected source of inspiration is organizational absurdity. A broken workflow, a vague brief, a “temporary” process that has lasted two years. These things can become visual material. They are not beautiful in the classical sense, but they contain tension, irony and human truth.

The unexpected sources are often the best ones. AI is very good at amplifying small sparks, but only if you give it direction.

Maybe AI illustration is not the birth of a new kind of image. Maybe it is the return of art direction.

Any advice for someone starting AI art?

My first advice would be: don’t start with tools. Start with taste. The tools change every month. Taste takes longer to build, and it is much harder to fake. Look at images outside AI. Study composition, color, rhythm, framing, narration, printing, photography, cinema, illustration and design. Ask yourself what kind of image you would still like if nobody told you how it was made.

I work with AI every day, but I don’t really like most “AI art”. Too many AI images are made by people who discovered image-making before they discovered images. They are technically impressive, but culturally empty. I don’t think AI is magically revolutionizing creativity. I think it is making visual culture more important than ever. The machine can generate an image. It cannot decide why that image should exist.

If I had to start again, I would build a stronger visual system earlier. At the beginning, it is tempting to explore everything because AI can do almost everything. That is also the trap. The real work begins when you decide what belongs to your world and what does not.

One mistake I made was spending too much time chasing beautiful isolated images. A beautiful image is nice. A coherent series is more powerful. A recognizable visual language is stronger still.

Another lesson: negative rules are often more useful than positive descriptions. Saying what must never happen can protect a style better than writing ten paragraphs about what it should feel like.

My short manifesto would be:

Taste before tools.
Systems before prompts.
Series before images.
Curation before automation.

For social media, I would give the same advice: don’t only post results. Share a point of view. Show what you are exploring, what you are questioning, what you are learning. People are not only interested in images. They are interested in the intelligence behind them. I don’t trust likes very much. A like is attention, not value. Especially when it comes from ‘friends’. For me, the real test is whether someone is willing to publish, commission, exhibit, buy or use the work.

The advice I wish I had received earlier is simple: AI does not replace direction. It makes direction more important. The better your eye, your references, your constraints and your taste, the better the machine becomes.