Procedurally generated insect wings using curved lines and Voronoi patterns.
Overview #
This proof of concept is an editor for creating images of insect wings. It uses algorithms to draw the two types of veins: primary veins and cross veins. A series of curved lines represent the primary veins and Voronoi polygons are used to draw the cross veins.
Parameters control the placement, length, and curvature of the primary veins as well as the density of the cross veins. The parameters can be randomized at the press of a button to explore a wide variety of wing shapes. The sliders in the UI allow the parameters to be adjusted on the fly.
Details on Randomizing Parameters #
Randomizing the parameters doesn’t always create aesthetically pleasing wing shapes but two features speed up the process of exploring the huge variety of possibilities.
First, the editor automatically rejects wings which don’t meet certain constraints. For example, if the primary veins cross at any point the wing is rejected. There are also constraints on the size of the wing.
Second, the editor randomizes different groups of parameters separately. The workflow might look like this: the user randomizes the parameters that control the base of the wing—the starting positions of the primary veins—until a suitable shape is found. Then, the user randomizes the curvature of the primary veins without affecting the parameters for the base of the wing. Finally, the user randomizes the parameters controlling the edge of the wing—the lengths of the primary veins.
The more parameters being randomized at once, the longer it takes to find a case where all their values line up to produce the desired output. By breaking up related parameters into groups, the user can focus on one part of the wing at a time and retain the values of one group while randomizing the values of another.
Future Directions #
Exploring the Parameter Space. Since there are so many parameters, it would be interesting to create a map of the parameter space. A scatter plot where each point indicates whether the user accepted the wing shape could help find regions in the parameter space that produce more interesting wings and as well as regions to be avoided. Machine Learning techniques could help exploration rather than blind randomization.
Branching primary veins would make the wings more realistic and open up the possible wing shapes. This would be relatively easy to implement and the code was originally written with this feature in mind. Maybe the most interesting problem here is choosing a set of parameters to control the angle of branching and the curvature of each branch.
Recombining primary veins. Insect wings have a lot of instances where the primary veins recombine. Modeling this would be tricky. A tree will no longer work so a graph structure might be necessary. But determining a set of parameters to control the curvature will be hard. In its current state, the editor uses a series of lines so it’s easy to get neighbouring primary veins to curve in a similar direction. That would be much more difficult with a graph structure.
Reaction-diffusion. Incorporating some kind of reaction-diffusion system would probably greatly improve the primary veins. Branching and recombining would no longer break the model. On top of that, the very nature of reaction-diffusion systems would eliminate some of the issues with the current model. For example, primary veins that drift too far apart will automatically branch to fill the empty space. Primary veins would also automatically respond to their neighbours, allowing them to maintain equal spacing between them even when the curves are difficult to describe mathematically.
Tuning the reaction-diffusion parameters to get the desired effect would be a tricky but rewarding challenge. Such a program would also be more resource intensive than the current vector-based model. On the other hand, I’m sure it would produce some amazing animations.
The image above was generated using Reaction-Diffusion Playground.
Post-processing. The images of insect wings look a lot better when filters have been applied. In some of the images here, the corners have been rounded and shading applied. This can be taken further. The editor exports images but it could also be used to produce a set of points in space. These could be used to create 3D models.
Devlog Entries #
Want the raw and gory details? The wrong turns, naming struggles, and evolution of the project? Find it all in the insect wings devlogs.
Source Code #
https://github.com/nielmclaren/InsectWings
