AI & Adaptive Media
Studio A N F develops AI and adaptive media systems in which neural networks become part of a larger computational environment. Models can generate imagery, interpret sensor input, transform live data or influence the behaviour of a realtime visual system.
The objective is not simply to generate images with AI. Studio A N F integrates neural models with realtime graphics, simulation, sensing and generative software to create systems that can perceive, generate and adapt continuously.
The studio has worked with generative neural networks since 2018, from early GAN-based experiments to contemporary diffusion and realtime inference pipelines. AI is treated as a computational material — one layer within a broader system rather than a visual genre in itself.
AI as part of a realtime system
- Neural generation. GANs, diffusion models and other generative networks can produce or transform visual material continuously as part of a running installation.
- Realtime integration. Neural pipelines can operate alongside TouchDesigner, Unreal Engine and custom software, connecting AI output directly to realtime graphics and simulation.
- Adaptive behaviour. Cameras, sensors, environmental conditions, live data or audience activity can condition how the system generates and evolves.
- Computer vision. Neural models can interpret movement, objects, spatial activity and other visual information before translating it into abstract inputs for the artwork.
- Project-specific models. Fine-tuning, custom datasets and specialised workflows can be used where existing models do not provide the required visual language or behaviour.
- Local inference. Where required, models can run entirely on installation hardware without depending on external AI services or cloud APIs.
From generation to adaptation
A conventional generative-AI workflow produces an output from an input. An adaptive media system introduces a continuous feedback loop: the environment provides data, the system interprets it, computation changes its internal state, and the resulting output becomes part of an evolving realtime experience.
Neural networks can operate at different points inside this loop. They may generate visual material directly, analyse camera input, classify environmental information, transform data into latent representations or influence the parameters of an existing generative system.
This makes AI particularly useful when combined with media architecture and interactive environments, where an installation may need to remain visually coherent while responding to changing conditions over long periods of time.
Selected AI work
AI Hypercore: Digital Pop Art Maximalism — 2024
Generative AI used to construct highly dense synthetic visual worlds exploring contemporary image culture and machine-generated aesthetics.
Hypercore Mandelbrot Set — 2024
AI-augmented generative work combining the mathematical structure of the Mandelbrot set with neural image synthesis.
Generative Adversarial Portraits — 2020
Early GAN-based work exploring the latent visual space of machine-generated portraiture.
GAN Sculptures — 2020
Physical sculptures derived from neural generative processes and translated from computational forms into 3D-printed objects.
How Studio A N F works
01 — System
Studio A N F defines what role AI should play within the project: generation, perception, transformation, classification or adaptive control.
02 — Pipeline
The appropriate models, realtime engines, data sources and hardware are combined into a working computational pipeline.
03 — Prototype
Models and system behaviour are tested against the actual visual, performance and latency requirements of the installation.
04 — Integration
The neural pipeline is integrated with the project's realtime graphics, sensors, media servers and display infrastructure.
05 — Deployment
The complete system is delivered for ongoing operation, including the required models, software configuration and technical documentation.
Does the AI system require a cloud service?
No. Where privacy, reliability or long-term operation requires it, Studio A N F can run neural inference locally on the installation hardware. Models and weights remain part of the deployed system, eliminating dependency on external APIs and allowing installations to continue operating independently of third-party AI services.
How is this different from prompting an AI image model?
Prompting produces an output. Studio A N F designs the computational system around the model: how it receives information, how it generates or interprets data, how its output interacts with realtime graphics, and how the entire process behaves continuously in physical space.
The model is therefore not the finished work. It is one component inside a larger system that can run, perceive and evolve.