The interactive installation Deep Mirror is structured as a hybrid apparatus that integrates physical manipulation, algorithmic processing, and real-time AI-generated images.
Two tables positioned in front of a wall projection organize the field of interaction. On the first table rests a physical object-model with movable and articulated parts, mounted on a rotating base. Users can rearrange its components, alter its spatial configuration, and rotate it freely. A camera captures internal images of this object, continuously transmitting the video stream to an AI-based image generation system, which reinterprets the captured material according to variable atmospheric parameters.
On the adjacent table, a 32-inch screen displays the system’s interface. As participants approach, they gain access to the Deep Mirror controls and can intervene in the application of so-called Operative Atmospheres: parametric sets that modulate lighting, materiality, volumetric density, tonality, and the intensity of algorithmic intervention. The transformations become immediately visible in the projection, establishing a continuous loop between physical gesture, computational processing, and visual updating.
In the system’s interface the interaction unfolds across two complementary layers.
In Basic Mode, the system offers a synthetic and intuitive control surface organized around a small number of high-impact parameters, enabling fluid and accessible atmospheric exploration. In this layer, the user can load and edit more than 500 operative atmospheres recorded in the earlier sessions.
In Advanced Mode, users access the full set of variables (multiple prompts, weights, steps, depth map parameters, and noise controls), allowing for refined parametric manipulation of the generative process.
This dual architecture does not establish a hierarchy between simplicity and complexity, but rather two synchronized perspectives on the same machinic state. Any modification carried out in one layer immediately reverberates in the other, ensuring continuity within the creative process. In this sense, Deep Mirror operates not merely as a visualization tool, but as a feedback system between human imagination and machinic imagination, in which the physical object functions as a material trigger and the AI as a field of probabilistic variation.
The installation thus constitutes an experimental environment in which architectural atmosphere ceases to be a static representation and instead emerges from a recursive dynamic between tactile manipulation, technical inscription, and real-time algorithmic updating.
