Using machine learning to generate visual patterns

Machine learning can turn data, images, sound and movement into a living visual language. Instead of asking a system to reproduce one finished picture, an artist can train it to recognise relationships between colour, texture, scale and repetition, then use those relationships to create fresh compositions.

This approach sits comfortably within the experimental spirit of ITP 30 Show, where interactive art, design, media and technology meet. A pattern may respond to weather, audience movement, archived material or an unpredictable stream of information, giving viewers a work that changes each time they encounter it.

For Australian creators, the source material can be intensely local. The Bureau of Meteorology offers rich weather signals, while Sydney light conditions, Melbourne tram movement, Brisbane humidity or seasonal bushfire data can all become inputs for a generative artwork.

The most compelling projects treat machine learning as a creative collaborator rather than a shortcut. Human choices remain central: selecting the dataset, defining the visual rules, deciding what should remain recognisable and shaping the experience around the final display.

From data to visual language

A machine-learning artwork begins with a dataset. This could include photographs of eucalyptus bark, satellite images, handwritten marks, fabric samples or thousands of small illustrations. The system searches for recurring features and represents them as numerical relationships, allowing it to compare and recombine visual elements.

Generative adversarial networks, diffusion models and neural style-transfer systems each produce different results. A diffusion model may gradually build an image from noise, while a style-transfer workflow can apply the rhythm of one collection to the structure of another. The choice affects whether the outcome feels orderly, organic, fragmented or dreamlike.

Designing patterns with rules

A useful pattern generator needs constraints. An artist might limit the palette to colours found around Sydney Harbour, require shapes to follow a grid, or let the density of marks change with wind speed. Rules give the model a framework while leaving room for variation.

Small changes can create major differences. Adjusting the level of randomness may turn a calm tiled surface into a turbulent field. Changing the frequency of retraining can make a work feel stable or constantly in motion. This makes iteration essential: screenshots, sketches and test prints help reveal which settings produce meaningful visual behaviour.

Making weather visible

Weather data is especially effective because it connects an abstract system to daily experience. Temperature, rainfall, cloud cover and wind direction can control gradients, movement or the arrival of new shapes. A viewer might see a storm as a dense cluster of dark marks, while a clear afternoon produces open space and bright tones.

The weather-powered painting featured in the ITP 30 Show journal demonstrates how environmental information can become part of an artwork’s structure. For an Australian audience, linking a piece to local conditions could make a gallery installation respond to a heatwave in Perth, a cool change in Melbourne or a wet season afternoon in Darwin.

Working with place and culture

Pattern generation can build a strong sense of place without simply decorating a generic model with local symbols. Artists might draw from urban textures, public transport maps, coastal erosion records or community-submitted photographs. A project shown during Vivid Sydney could transform real-time city light levels into a shifting projection, while a regional exhibition might use agricultural or river data.

Care is needed when working with cultural material. Indigenous designs, stories and visual systems should not be treated as freely available training data. Permission, attribution and collaboration matter, particularly when a project draws on Aboriginal or Torres Strait Islander knowledge. Ethical sourcing gives the artwork a stronger foundation and avoids turning cultural identity into an aesthetic effect.

Building an interactive experience

Visual patterns become more engaging when visitors can influence them. A camera can track movement, a microphone can translate sound into colour, or a touchscreen can let people alter scale and density. In a Melbourne gallery, a work might respond to the pace of visitors moving through the room; in Brisbane, humidity readings could gradually soften its edges.

Interaction should support the concept rather than act as a technical demonstration. Clear feedback helps people understand that their actions matter. A delayed response can suggest memory, while immediate transformation creates a playful sense of control. Designers should also provide alternatives for visitors who cannot use touch, sound or movement-based controls.

From prototype to public showcase

A generative system may work beautifully on a laptop but fail on a large screen, projection surface or gallery network. Artists need to test resolution, colour calibration, processing time and data reliability. Local conditions matter too: strong daylight in an Australian exhibition space can flatten subtle tones, while unstable internet access can interrupt a live data feed.

The presentation surrounding the model is part of the work. Brief project notes, author credits and process images can show audiences how the visual pattern emerged. A digital project about probability and audience behaviour could sit beside the instant-play casino example, prompting discussion about interfaces, chance and the persuasive design of interactive systems.

The Australian creative market also rewards adaptable outcomes. A prototype might become a projection for a gallery, a browser-based artwork for remote visitors, a printed edition for a design shop or an educational activity for a community centre. Considering these formats early helps artists balance ambitious machine-learning experiments with practical production needs.

Submit a project to an exhibition, document your experiments and invite audiences to see how data becomes form. Whether your source is weather, movement, sound or an everyday archive, machine learning can provide a responsive new surface for artistic thinking.