The project has outlines

I’ve busied myself with… producing text and looking at images. Much time has been spent browsing ffffound.

Here follows my first sketch for outlining the Recombinable Vector Graphics library and what the project should achieve. Pompous, eh?


To create a functioning prototype, or proof of concept for the recombinable vector graphics library that includes at least:
Shape creation
Shape combination
and a test functionality for a semantic categorization system

Additionally a test method for automated composition generation could be included, but this is work-intensive and can risk sinking the project.

With a functioning prototype I suggest that the system should be usable for interesting visual output, so that different applications can be tested, before creating a Nodebox (or general-use) library of the material. The technical functionality and reliability can then be improved for release versions.

The visual interest of the results are for me at this stage very important. It would be a good goal to try to get some attention to the project from outside the generative graphics scene.


A. Shape creation
The structure or method for creating image content for the database.
How can these be created in a flexible manner? It is interesting to consider wether the visual products of the RVGL should be more signlike, or more like unified compositions. Signlike shapes can just be assembled on a canvas in different groupings and need little modification in themselves, but more elementary shapes provide for a wider variety of compositions and more visual interest.

Different possible approaches include:

1. using imported handmade shapes. Relating to this, one can for instance look at the well-drawn images by Gerd Arntz’s for Isotype. (from’s page on Isotype).


This is a very signlike approach. In the Isotype system one (theoretically) only needs a sufficient repertoire of shapes to illustrate any concept. However, the execution of the parts is very much the result of Arntz’s excellent draftmanship.

2. modular shapes or “atoms” for building larger aggregates, i.e. basic shapes (blocks) that form a figure or part of it. The illustrator Charles Harper used a characteristic geometric way of abstraction for his animal figures. In the tile mural below, we have animals constructed from simple geometric shapes, in turn made from square and triangular “pixels”.

The classic puzzle game Tangram challenge you to create different shapes using a set of geometric pieces. Here’s an egg-shaped version and birds created from it:

3. What other possibilities coul there be? Shapes consisting entirely of outlines / contours that intersect and branch would be a concept perhaps worthy to explore. This approach would mean that all shapes consist of a loop or a line, a wire if you like. The inspiration for this comes from wire sculptures like those created by artist Alexander Calder. The picture below is a portrait by Calder of Ferdinand Léger, the famed “tubist” painter.

In the thursday meeting we discussed the idea of modifiable shapes – something that would bring additional interest to the project. Each shape could have a domain of possible transformations, i.e. – multiple master shapes.This means the transforms would be built-in for each shape, and produce meaningful, dynamic transformations. This is the method used in Will Wright’s computer game Spore for creature building, and it is taken to an impressively advanced level. The creatures have symmetry that define how parts behave, depending on where they are attached. The final creatures are procedurally generated and can be described with a short string of code, the creature’s DNA if you will.

One important question regarding the shapes is what the smallest shape unit should be? Does it have a semantic content in itself, or is it a building block, a shape atom or pixel? In spore, the basic shapes are individual bodyparts that are in themselves modifiable, and thus allows for an enormous amount of different critters through combinations and modifications. However, creating new basic pieces for this system is not easy, and generally beyond the capabilities of the users.

* * *

B. Shape combination

This means the way larger shapes or figures are generated from the discrete parts – say, a bird from a set of wings, head, beak, legs. More than one shape will probably be used in a composition, regardless if the smallest unit is complex or very simple. The shapes need to “know” their own buildup and have properties that guide how they click together like the Spore creature pieces. For those working with illustration (or a general design audience), this is perhaps the most interesting part of the project. Many illustrators / designers create their own reusable graphics pieces, but dynamic shape combination methods is something new.

What possible ways are there then to combine the artwork?

1. A modular approach allows for shape combination to use a similar method with which the first-level images themselves are created by clicking pieces together edge-to-edge; but the result will also look modular, somewhat like Fontlab’s FontStruct that allows you to create grid-based fonts using different geometric pieces. I made a simple test image using Tangram-like tiles in illustrator (below). This approach makes for easy recombination of the created pieces, and the user can make new pieces with relative ease.


2. For handmade shapes, fixed skeletons can be assigned that will be used for composition. At a simple level, the shapes would just be added to each other and not combined visually, like in cut paper animation. A more advanced process would modify the objects to fit them together seamlessly. Path modification does have the additional drawback of being quite computationally intensive, and thus slow.

A skeleton which dynamically modifies the “skin” requires handwork to function correctly (if the system should work for animations), but the skeleton can also be fixed in regard to the shape and just function as a guide for attatching. The usefulness of the skeleton is particularly that it would form a framework for basic shape recognition, based on a few key points without having to analyze a complex, arbitrary shape.

3.The shapes could attach directly to each other using something akin to surface tension, melting into each other. Shapes could be given polarities, zones of preferential attachment that attract each other. This would be more a experimental technical concept than a practical method of combining shapes.

4. For combining line shapes, the connection should be simple: a line can either intersect or join another.

* * *

C. Generating compositions

Compositions are structures created of several shape combinations.

1. If the shapes are based on some grid structure, a composition is essentially similar to a single shape or a combination of shapes; the scale just increases.

2.In case of pre-cerated blocks, the composition needs to be either free-form, or conform to a grid defined independently from the shapes – in which case the shapes are attached via their skeletons to grid points.

3. An approach using lines result in a similar construction to a modular approach: a composition is just a larger aggregate of lines.

The compositional system might need to have rules that stop it from creating too dense or otherwise unworkable compositions if used automatically. For user-controlled compositions, no grid-snapping might be necessary most of the time. Users are capable of positioning their creations on the drawing board, and a grid can be more of an annoyance – think PowerPoint’s invisible grid!

An interesting alternative using the modular tile-based approach is to have the entire canvas in essence to form a tiling pattern, that is then modified to include the created figures! To create an effect like that by hand is very painstaking, so it is good use for some computing power.

* * *

D. Semantic structure

1. The semantic framework behind the shapes that links them to classifications using tags etc. relating different shapes to each other in a semantic web. I.e. the shapes need to be assigned meanings when they’re added to the database.
2. Depending on the build-up of the shapes, a network for shape resemblance can be implemented. (Meaning, rather than shape).  Different methods of classification loaned from those used for organic systems could be useful; such as sorting by symmetry types.
3. The semantic framework opens the possibility of generative compositions and compositions created from natural language input, by linking it to WordNet.

* * *

General issues and questions

The project can be divided into semantics and visualisation, that can be seen as independent. Depending on the solutions chosen, shape combination and composition generation can also be somewhat independent from visual style.

Before starting to create shapes it is important to decide wether to work with a fixed perspective or viewpoint. For instance, are faces always drawn in profile? Constraints on visual style could give focus the project and perhaps also thus improve visual quality. The images will be two-dimensional, but for instance an isometric perspective could be considered (Eboys use a large and growing library of isometric pixel figures, that can easily be assembled to enormously complex compositions).

I personally find that the prototype could even be highly idiosyncratic, since this can have more visual interest than a general-use neutral visual style. Canadian comics artist Marc Bell creates images that are at freewheeling, psychedelic and unhinged, but at the same time very constrained. Ultimately, one would want to have a recombinable artwork library that can create images as rich in complexity and texture, using generative methods. This is not to suggest a replacement for handwork, more like a target for what should be attempted.

Another issue is to decide wether the system will integrate color on a basic level (i.e., is color a property of the shapes?) or if color is applied separately. It is presumably relatively easy to use semantic tags for coloring, but to get natural shadows etc. a more advanced system would be needed. Because the project has so many possible paths, it is important to watch out for time-consuming sidetracks!


Post a Comment

Required fields are marked *

%d bloggers like this: