Hugues Bruyère, known online as smallfly, is a French-born artist, originally from Avignon and based in Montreal, who works at the intersection of physical interaction and AI image generation.He is also co-founder and Chief of Innovation at the design and technology studio Dpt., where the same real-time AI techniques inform his personal art practice.,
Bruyère's early experiments combined Neural Radiance Fields, which let him generate new camera viewpoints of natural and urban scenes, with Stable Diffusion, ControlNet, TemporalKit and Ebsynth to restyle the resulting video.He later moved to real-time image-to-image diffusion pipelines built on Stream Diffusion and SDXL Turbo, steering the generated imagery live using physical materials such as clay, a tracked magnifying glass, and handheld light alongside text prompts.,He also explored dropping ink into milk as a live input for the diffusion pipeline, displaying the resulting images in circular plates or steel-tray triptychs.
Writing in Fast Company, journalist Jesus Diaz described this work as a sign that “there's hope for artificial intelligence and the human brain to create together, rather than becoming adversarial entities in the AI hamster race”.In January 2024 Bruyère staged his first outdoor mixed-reality experiment, using a Meta Quest 3 headset in a back alleyway near his home to revisit a 2019 augmented-reality piece in which a viewer's circling motion around a virtual sculpture could speed, slow or pause a recorded physics simulation. Viewers who saw the piece online compared it to William Gibson's 2007 novel Spook Country and the 2016 video game SUPERHOT.
In 2025 Bruyère designed and developed The Latent Space, an interactive live-performance artwork made in collaboration with the artist Alejandro Cartagena and presented through Fellowship from November 2025. A camera captures physical gestures and materials placed in front of the lens, which a real-time diffusion pipeline built on SDXL Turbo and the Hugging Face Diffusers library reinterprets into evolving imagery. The system uses custom LoRAs trained on datasets derived from Alejandro Cartagena's own work as behavioral biases that shape the generated images, part of a project the two artists describe as addressing “the ideals of homeownership in contemporary Mexico”.The piece continues Bruyère's ongoing series Hands-on Hallucination, and follows an earlier NFT project, Octreefied VI, documented on CreativeApplications.Net in 2021.,
