25 September 2020

On Creating with Hybrid Generative Systems

How can generative systems enhance creative exploration?
Dr Camilo Cruz reflects on designing a generative system to grow 3D prints.

Designing is an inherently generative process. Instinctively—or perhaps by training, as this is what I was taught to do in architecture school—designers are almost immediately drawn to put together something that they think addresses the problem at hand. These prototypes are excellent at revealing what works and what doesn’t in order to attain the desired outcomes, and this is a method—we were told—that even proto-designers, such as tool makers and artisans used, before scaled drawings and modelling technology were ubiquitous. They made things, put them to work, and if they weren’t good enough, they would adjust them in the next version. This is how they built up know-how, and ultimately advanced their craft. The ‘trick’ is to do this enough. To iterate between generating and testing so you assess all the possibilities in the spectrum and settle for the best one. But what if a better alternative is still out there? What if, because of contextual constraints, time limitations or a creative rough patch, that better alternative has escaped the designer’s creative capabilities or imagination and never comes to be?

Artistic production could also be considered generative in nature, and for similar reasons: when the goal is to express an artistic idea and convey a sensorial experience, exhaustive search and careful refinement are the methods of the trade. However, one notable difference between designed artefacts and works of art lies in the fact that the former always have a practical purpose that supersedes the artistic character of the artefact. The task of designers is twofold, as it requires focus not only on addressing the practical problem at hand, but also on how the solution provided will impact the environment where it will be deployed. This is especially true for the design of physical objects meant to be looked at, inhabited, or interacted with (for more on this see Lawson 2006), as they will have aesthetic and environmental implications deeply related to cultural and contextual issues, which are unpredictable, and can sometimes override the value of the intended practical solution.

The fundamental value of generation in creative disciplines, such as visual arts and design, is that it enables creative exploration. Sometimes called ‘looking for inspiration’, this process involves the observation of artefacts—either pre-existing or purposefully generated—in search of elements and/or characteristics perceived as conducive to satisfying the expectations that the new artefact is meant to address. In traditional design and arts, the ability that a creator has to to generate and explore a landscape of alternatives derives from their talent, and is what supports the heroic narrative behind creativity. Generative systems have been used for this purpose by designers, artists and makers since the times of Aristotle (for more on this see Mitchell 1977). Generally speaking generative systems can be described as sets of building blocks and explicit rules that are used to generate new things. The rules determine how, when and for how long the building blocks are manipulated, aggregated, deleted, decomposed and/or recomposed, resulting in a new artefact. Classic examples of generative systems are John Conway’s Game of Life and Stiny and Gips’ Shape Grammars (the latter are actually the foundations for a whole family of generative systems).

Setting up a generative system is not an easy task. It requires thorough understanding of the components and rules, and in turn designers and artists have to partially give up control over their creative process, as what comes out of the system is often hard to predict.

The upside of this compromise is that through the use of generative systems, it becomes possible to explore lineages of alternatives, rather than being limited to those one can imagine directly.

Additionally, the knowledge of building blocks and rules developed through the definition of a generative system enables the quantitative evaluation of characteristics of the artefacts it produces —e.g. the statistical distribution of individual building blocks and the detection of patterns or motifs, just to name a few— which can potentially lead to the development of methods to guide the exploration of vast spaces of creative possibilities.

For these reasons we believe that generative systems—specifically digital ones—are a well suited method for creative exploration and to search for innovation in design and visual arts (for more on this see this article). 

Growing 3D Prints

For the past year we have been experimenting with a generative system that uses the constraints of fused deposition modelling (FDM) 3D printing as the conceptual basis for a 2D bio-inspired dynamic life simulation, which is then used to generate 3D printable objects. Growing 3D prints is not a fully fledged design driven generative system, as the objects it produces are not necessarily meant to address practical problems. However, it enables us to produce a vast landscape of alternatives where we can search for the ones that best take care of conflicting goals: printability and aesthetic interest, where the former is purely practical, and is determined by the limitations of the manufacturing process of choice, and the latter is bound by ideas of style, complexity, pattern association and other subjective attributes. Our aim is to explore the tension between the artistic and practical attributes of physical objects.

This short animation illustrates one organism (the squiggly polygon) in its interactions with the environment. Here, each small circle along the perimeter of the organism is a cell and the arrows coming out of them show the result of the forces acting upon them: attractions, repulsions, and their own desire to acquire nutrients (represented by green circles)

The building blocks of Growing 3D Prints are 2D cells; abstract entities that have a position in 2D space, velocity and acceleration that define their movements, and attraction and repulsion forces that define their interactions with other cells. These entities can also hold energy, and have a desire to survive, represented by their constant food-seeking attitude.

Each cell is connected to neighbouring cells by edges that behave as springs. This allows them to distribute their excess energy, as well as to maintain structure, which in this case are closed polygons that we call organisms.

Organisms live in a 2D environment; a homogeneous medium where food sources grow in random positions and diffuse food for a limited period. We measure the system’s time in discrete time-steps.

The way in which our organisms behave is determined by their genetic attributes:


  • Metabolic Rate (MR) determines how much of the food consumed by a cell is converted to usable energy, and how ‘energetically expensive’ it is for a cell to be alive.
  • Cell Velocity (CV) limits the velocity of movement in exchange for a lower energy consumption.
  • Energy Capacity (EC) limits the amount of energy a cell can hold in exchange for more freedom of movement.
  • Spring Coefficient (SC) determines the flexibility of organisms.
  • New Organism Energy Ratio (NOER) determines how easy it is for an organism to divide into two.
This organism has high values for MR, CV and NOER, and low values for EC and SC. This results in a configuration that remains as a unit and produces a small tentacle to reach for nutrients.
This organism has high values for MR and SC, a medium value for EC and low values for CV and NOER, which develops into a community of smaller organisms searching for food independently.

Based on these characteristics, cells and organisms will have different capabilities to move around, split into colonies of organisms, acquire food, and ultimately  maintain sufficient energy levels to survive for as long as possible.

These short animations to the left illustrate the behaviour of organisms with different genetic makeups on an identical environment.

In these animations we illustrate how the history of each colony is arranged to generate 3D printable objects (these colonies correspond to the ones shown in the slides above).

As can be observed in the images above, the dynamics produced by seeding the generative system with organisms with different genetic codes can produce a wide variety of behaviours. For the generation of 3D objects we capture the state of the organisms at every time-step, and then stack them from bottom to top.

The series of paths produced in this process is then translated into G-Code, and ultimately fed into a Prusa MK3 3D printer to generate objects that exhibit vastly different characteristics in texture, shape and topology, from plain to intricate, as shown in the images below.

Early 3D printing tests.

Navigating the generated design landscape

Experimenting with Growing 3D Prints has shown us the potential of digital-to-physical generative systems. On the digital side of things, the capability the system has to produce diversity within a lineage of objects (the constraints of the system, based on 3D printing principles, give way to objects with similarities, yet vastly diverse) makes it a fertile platform for design and creative exploration (see image below). However, diversity alone does not guarantee that expectations will be met. We need to continue working on methods that allow us to steer the search trajectory through this vast design landscape based on intent or purpose. Additionally, on the physical side of things, fabricating generated objects using 3D printing has enabled us to physically manipulate them, as well as to compare them to their digital twins, and in doing so we have gained further understanding regarding what works aesthetically, such as what the limits and tolerances of the fabrication technique are.  This is often counterintuitive and overall, in silico and tangible representations of theoretically identical entities will show discrepancies. This gap needs not just to be acknowledged, but could also be successfully exploited for creative purposes.

Wireframe renderings of a selection of diverse generated 3D objects.

Dr Camilo Cruz is a research assistant at SensiLab. Read Camilo’s recent research publications here.

Growing 3D Prints is part of SensiLab’s research into Generative Materiality.