An AI did Nana's grocery shopping and it bought forty kilos of lemons. Fifteen seconds, one kitchen, two people who have never existed. Every face, voice, lip movement and note of music comes from open models.

Three women who have never existed held across six scenes, from a dry field to a pottery wheel to a balcony. In three of them, two women share the frame. Each face comes from a model trained on her portraits, those portraits guide every clip, and every still is checked against them before it is kept.

A system that reads one of Kandinsky's Bauhaus paintings and plays it as music, using a colour-to-sound logic built from his own theory in the books On the Spiritual in Art (1911) and Point and Line to Plane (1926). A machine-learning model extracts the painting's elements — points, lines, planes, their colours and positions. This reading is currently semi-automated. Each element is then turned into sound by rules drawn from Kandinsky's writing, where he claimed colour and form obey a law as exact as music.

An infinite descent shaped by a personal archive of interests and favorite moments — a table, a mountain, a stretch of beach — and rendered through a model trained on painterly style. Given any new frame, a machine detects the object that echoes that archive, and dives into it. Each dive blooms into a new world.

A recording of an Athens street, split into scenes by an audio classifier. Each sound becomes an object along the path — not the thing itself, but the feel of it: speech as lips, an engine as a car, music as a guitar — scored with a song, fragments of the original street recording surfacing beneath it.

A dandelion that breathes with sound. The structure comes from code; the bristles, depth, and glow come from a diffusion model reading each frame and adding what it imagines a real plant would have.

A series of short generative films rendered from a decade of Athens meteorological records. Temperature, wind, pressure, and rain for a single day pass through a diffusion-based video model.

Generative image and video models don't keep a face: the same woman drifts from one shot to the next, and garments lose the details that make them recognisable. on-model-pipeline generates fashion stills and clips with a fixed cast. Choose the woman, the garments and the scene; it writes the prompt, draws the image and checks every detail against her master and the garment's reference. A wrong colour or a missing button is caught, explained and fixed in place, and anything the system can't see is passed back to a person to decide.

Used to make on - model, a generative fashion-ad demo for a fictional spec brand.

Artists working with unfamiliar materials, techniques, or art-historical references constantly need to look things up — but generic search returns shallow answers and generic chatbots invent details. art-assistant retrieves passages from a curated knowledge base and returns sourced, grounded answers about materials, techniques, and process.

A weekly filter for one artist's inbox. Reads incoming arts newsletters, scores each item against a personal profile, classifies by topic and urgency, and assembles a single HTML digest of what actually matters.

A no-code companion version of this pipeline is the basis of the inbox-signal workshop.

Selected product images for the new collection of ORhandmade, an Athens label of handmade clothing. Starting from the brand’s own photographs, I replaced the backgrounds, then relit and retouched each image, changing everything around the clothes and nothing in them.

City Flowers — a public art and urban innovation concept exploring how environmental data, collective feeling and neighbourhood identity can be translated into an immersive civic experience.

Organized by Onassis ONX, ACE and NYU Tandon School of Engineering.