THE GALLERY
AI EPISODIC SERIES
2026
The Gallery is a radically experimental project unfolding as a dystopian episodic series set inside an art gallery that exists nowhere, suspended on the threshold between the real and the artificial. This threshold is not only metaphorical, but literal. At times, the gallery appears embedded in familiar landscapes; at others, it floats in deep space, disintegrating and exposing its true nature as a structure made of pixels, simulation, and hunger. In this sense, the gallery becomes both a stage and a creature: a liminal architecture that embodies the unstable condition of contemporary art itself, forever poised between presence and abstraction, materiality and image, ritual and spectacle.
Within The Gallery, generative video models are forced toward what suits them least and what they have likely been trained for the least: not action cinema, not realism, not narrative efficiency, but performance, symbolic gesture, suspension, minimalism, stillness, and ritual. In other words, they are bent away from mimesis and toward something more fragile, opaque, and disturbing: the language of presence without presence, action without explanation, cinema emptied of its conventional certainties.
Each episode revolves around the preparation of a new surreal event inside the gallery. The protagonists organize exhibitions, performances, apparitions, and impossible situations that always culminate in an uncanny rupture, an event that reveals the true nature of the place: monstrous, famished, and never neutral. The gallery is not a container for art, but a devouring organism at the edge of the world, a stomach made of architecture, flesh, and pixels, feeding on bodies, images, and symbolic residue.
The director of the gallery is myself, embodied through my digital avatar, accompanied by my two unborn performer-assistants, Marco and John, synthetic figures generated through AI. In each new episode, they are joined by the avatars of real collaborators — dancers, performers, and actors who choose to enter the project and lend themselves to its unstable ontology. Together, we inhabit a world in which the distinction between performer and simulation, artwork and host, author and generated entity becomes increasingly uncertain.
The Gallery is a project about the seduction and failure of images, about the violence hidden inside immaculate surfaces, and about the possibility of using AI against its own dominant tendencies, pushing it away from smooth illusion and toward something stranger, poorer, more symbolic, and more alive.
Shows
Escape Media homepage cover feature
Credits
Concept and art direction - Francesca Fini
THE GALLERY - PILOT
AI SIMUL / ACTION # 1
At the center of this episode, two statue-like golfers move inside the aseptic space of The Gallery, transforming athletic precision into a ritual of dripping, collapse, and chromatic release. What matters is not spectacle, but tension: the tremor of a muscle, the hesitation of a glance, the density of a gesture that the machine cannot truly predict, only distort and transfigure.
From a canvas floating in the void to the trace of a cave painting in a primordial grotto, the work moves through different layers of image-making, tracing a continuity between the most ancient mark and the contemporary pixel. Art does not change in essence, only in the density of its simulation.
When the light flickers and the image reveals its wireframe nature, what remains is not the illusion of life, but the persistence of an inner force. In that sense, the soul of the simulation becomes the last possible act of resistance.
The 3D reconstruction of The Gallery, with Claude, Phyton and Blender
Spatial coherence is one of the central problems in my AI-generated performances. What matters to me is not simply producing a striking image, but constructing an environment that can sustain multiple shots and symbolic actions without collapsing in perspective, proportion, or internal logic.
Although many commercial tools are now addressing this issue, especially through Gaussian splat workflows, I still find the most reliable solution to be more grounded: rebuilding the space in 3D.
Recently, I developed a faster workflow by combining Claude, Python, and Blender. I begin with a single view of an AI-generated space, then generate the remaining views to establish a more complete spatial structure. From these images, my custom Claude-based script produces Python code that reconstructs the architecture directly inside Blender.
What emerges is not a mere visual approximation but a parametric architectural model that can be refined through language: dimensions, proportions, spatial relationships, and structural details can all be adjusted before manual intervention begins. The result is not a perfect replica of the source images but a solid and highly flexible foundation.
From there, I refine the environment manually, texture it, light it, and render it as a coherent space able to support the entire visual sequence. I also position provisional figures within it, later replacing them with final characters developed through image-generation and editing tools while maintaining continuity through reference datasets.
For this particular video test, I worked with the raw geometry of the environment in a deliberately reduced state, with basic lighting and no textures, using Dreamina AI and Seedance 2 for character animation.