in situ: Refik Anadol presents Living Architecture: Gehry, a groundbreaking audiovisual AI Data Sculpture that reimagines Frank Gehry’s architectural legacy through artificial intelligence (AI) and generative art. Developed by Refik Anadol Studio, the custom-built AI model Large Architecture Model (LAM) is based on advanced technology and has been trained for months on a vast archive of open-access imagery, sketches, and blueprints to transform Gehry’s architectural language into ever-changing landscapes of dynamic form, color, and movement. Augmenting this visual spectacle is an immersive soundscape composed by Kerim Karaoglu, blending AI-generated audio with material recordings captured within the Museum itself.

For this exhibition, Anadol took on the challenge of teaching AI to “learn” the architectural style of Frank Gehry, his favorite architect. This project highlighted Gehry’s unique design principles, presenting a radical and futuristic visualization of his architectural vision while paying homage to one of his masterpieces: the spectacular Guggenheim Museum Bilbao. The immersive experience engaged visitors with AI-generated projections that displayed the dream-like quality of Gehry’s iconic structures.

The dataset includes 32,147,692 photographs of architecture from Frank Gehry and others curated from publicly available image archives.

Living Architecture: Gehry unfolds across six interconnected chapters, each representing a distinct stage in the evolution of data into architectural imagination:

• Chapter I — Data Universe: A vast constellation of materials and data—archival photographs, plans, and structural documents—are compiled into a memory space.
• Chapter II — Data Plotting: As the data is processed, complex patterns emerge through the arrangement of form, texture, and spatial composition.
• Chapter III — Data Tunnel: The data transitions into immersive visual environments in the latent space, inviting viewers to engage with the evolving process of human-machine collaboration.
• Chapter IV — Large Architecture Model: By transforming Gehry’s forms, the AI generates new architectural possibilities that push the boundaries of imagination and design.
• Chapter V — Machine Hallucination: A moment of abstraction, where AI-generated structures break free from conventional spatial constraints, offering hallucinatory, dreamlike visions.
• Chapter VI — Dreams: A return to the world of architecture, now envisioned through the limitless imagination of AI—where architecture transforms in real time, continuously generating new forms that emerge and evolve in the moment.

LARGE ARCHITECTURE MODEL
Large Architecture Model Bespoke AI Model: Developed by Refik Anadol Studio, trained on vast open-access archives of architectural imagery, sketches, and blueprints. The model generates new, unrealized architectural possibilities, reimagining Gehry’s architectural language through ever-evolving form, color, and motion.

LARGE ARCHITECTURE MODEL — REALTIME GOOGLE CLOUD AI WORKFLOW
Refik Anadol Studio’s interactive AI research explores the creation of generative realities through real-time rendering.

For the Large Architecture Model (LAM), RAS has fine-tuned a custom-trained image diffusion model using Frank Gehry’s architectural language. This enables our AI to analyze, reinterpret, and adapt his aesthetic in unconventional ways, integrating it into diverse environments across the world. With billions of parameters, the model continuously generates images in an infinite loop, transforming Gehry’s distinctive forms into radically new, seemingly impossible architectural compositions, set in randomly selected locations and times of day.

These AI-generated visuals are upscaled from a compact latent space to massive 16K resolution images, projected in Gallery 208. The custom upscaling models, developed specifically for LAM, are trained to reconstruct even the finest architectural details with exceptional sharpness. Using embedding vectors derived from billions of images, the AI seamlessly expands low-dimensional representations into ultra-high-resolution outputs, preserving and extending Gehry’s aesthetic across ever-changing environments.

Running on Google Cloud’s sustainable computing infrastructure, the model operates continuously, generating new images every minute, powered by a custom-built combination of hardware and software designed for infinite exploration.