What it had to solve
No client brief: Dpt.'s chief innovation officer Hugues Bruyère wanted to see how generative AI behaves as a tactile, responsive medium.
The idea
A camera watches paper, clay and your hands; an image-to-image diffusion pipeline repaints them live — touch is the prompt.
How it works
Paper, clay and everyday objects arranged before a camera feed an SDXL Turbo image-to-image pipeline; each gesture returns an instant visual response. The research grew into Art Morph at the Cleveland Museum of Art and The Latent Space at SFMOMA.
What it did
Art Morph lets Cleveland visitors turn fabric and objects into artworks drawn from the museum's Open Access collection; The Latent Space (November 2025) was a live SFMOMA performance with photographer Alejandro Cartagena on homeownership and training-data bias.
Background
Dpt., an experiential design studio, runs Hands-on Hallucinations, an open-ended R&D project led by chief innovation officer Hugues Bruyère. Instead of prompting generative AI with words, he arranged paper, clay and objects before a camera connected to an SDXL Turbo image-to-image pipeline: every gesture produced an instant visual response, making the model a hands-on, responsive medium rather than a text-prompt tool.
The exploration became Art Morph at the Cleveland Museum of Art: in the reimagined ArtLens gallery — 21 new experiences the museum built with Design I/O, with Dpt. collaborating on two installations — visitors place fabric, paper or objects on a platform and watch AI transform them in real time into artworks inspired by the museum's Open Access collection. Materials are refreshed regularly so the installation keeps evolving.
The second offshoot, The Latent Space (November 2025), was projected live at SFMOMA: Bruyère collaborated with Mexican photographer Alejandro Cartagena on the question 'what does it look like to dream of a home?', training custom LoRAs on Cartagena's photographs of homes and suburbs while visitors handled objects that steered the images — a work about representational bias in training data and contemporary Mexico's ideal of homeownership.
Why it lands
It replaced the text prompt with gesture and material, making the model respond like a physical medium
The museum applications came after the fact, products of deliberate open-ended research
Art Morph ties its output to the museum's own Open Access images, so play stays collection-based
The Latent Space used the same foundation to interrogate training-data bias rather than just entertain
What you can take
Open-ended R&D with no application in mind can be the strategy — the same exploration became two very different museum commissions.
Since then
Dpt. says Hands-on Hallucinations continues, exploring new models, interactions and output formats; the studio frames the two museum projects as proof that untargeted R&D leaves you with knowledge and tools ready when the right institution arrives.
FOLLOW THE EVIDENCE
The sources
- Dpt. | Hands-on Hallucinations research blooloop.com