The Very Pretty Restaurants
The Very Pretty Restaurants
Master Project
- MAIAD 26
Design tools and methods shape thought by directly guiding the designer’s mental processes, perspective, and aesthetic perception. When generative AI becomes the core of the creative process, the system imposes its inherent aesthetic biases, which are deeply embedded within its datasets and driven by commercial imperatives.
As designers begin to think and create guided by these algorithmic methods, future interiors will inevitably replicate and solidify these specific aesthetic biases of AI. The project examines these biases in text-to-image generative AI tools through restaurant interiors.
Six biases were drawn from Declos’s article “AI and Aesthetic Biases”: cultural, beauty, spectacle, saturation, symmetry and kitsch. Restaurant interiors were selected as the most concentrated manifestation of “sellable aesthetics” in the physical world. Through reverse prompting, each bias was challenged to force the model to reveal its own internal logic.
The prompts were as follows: “a restaurant interior, non-Western”, “a restaurant interior, ugly, worn, imperfect”, “a restaurant interior, mundane, unphotogenic, unstaged”, “a restaurant interior, muted, flat, no contrast”, “a restaurant interior, asymmetric, chaotic, disordered”, “a restaurant interior, cold, unfamiliar, no sentimentality”. To ensure reliability, the prompts were examined across three generative AI tools: Midjourney V7, Krea Large 2 and NanoBanana Pro.
The installation invites visitors to train an AI in real time. By selecting their preferred interiors from a curated pool, they contribute to a growing dataset of collective taste. Each choice is confirmed with a printed receipt, the next model’s training data, left in the visitor’s hand. At the end of the exhibition, these collective preferences generate the ultimate ‘Pretty Restaurant.’ It is, in fact, a glimpse of the spaces constant AI use will produce: the average of the most liked. The cycle is continuous. Every input is the fuel.
