Ramon Llull and Image Idea Generation
The following paper was given at the Generative Art '98 Conference in Milan, Italy. I have had vague thoughts and fragmentary implementations of my ideas for new image generation for many years. Having to write this paper crystalised my concepts. So, if you are interested in how Repligator and Gliftic started, read on!
Our minds follow familiar well trodden paths, and we sometimes need mechanical help to find new, undiscovered thoughts and ideas. Ramon Llull was a Christian philosopher who used a mechanical way of investigating truth, a set of concentric wheels on which words were written. By rotating the wheels, the words were combined in unexpected ways. A similar technique has been (re-)invented, randomised and changed by Edward de Bono and is called Lateral Thinking. Can we apply this technique to image idea generation? This paper describes a first attempt at building an image idea generator for use by artists and non-artists alike, and sets out plans for a new a more powerful version. An image idea generator takes no (or very few) instructions from the user and creates a set of images from which the user chooses those he likes best. The purpose of an image idea generator is to easily create previously un-thought of images, to easily explore the 2D world of color and form.
1. The Mechanical Investigation and Creation of Ideas
Ramon Llull (1232 - 1316) was a pleasure seeking Spanish knight who liked to love the ladies, take part in tournaments, and do battle. Apparently he was accomplished in all three, earning high renown among his peers. One evening, Ramon awoke to a vision of Christ that changed his life. He took up Orders, became a monk, and dedicated the rest of his life to writing and trying to convert "heathens". He wrote on love, war, alchemy, religion. Legend has it that his long life was ended by stoning; he had been attempting to bring Christianity to a Moslem village. He invented a mechanical way of combining words, thus ideas. A set of concentric wheels or disks are inscribed with words. He thought that by rotating these wheels and reading off the combinations of words he could reach every possible knowable truth. Reading about these wheels it occurred to me that a similar technique could be used to generate image ideas. Not "every possible two dimensional image", but at least millions of them, some of which might even have artistic merit. Now we can do this using the graphics of a normal PC or Mac. Keeping with the model of Llull's wheels it would even appear easy to an untrained user of the program.
1.1 Three Llullian Wheels
So let us say that images are combinations of the positions of three Llullian wheels: color, form and interpretation. Beginning with color we can imagine a wheel for the color on which many color schemes are arranged around the circumference. But how do we represent a color scheme? Three histograms of relative distribution of red green and blue pixels is a possibiity. This simple method has the advantage that existing images can be analysed for their "Llullian color wheel position". Reducing every image to a standard 256 by 256 anti-aliassed version we have a way of comparing and analysing color schemes from images of different sizes. The Llullian wheel for "form" is harder to imagine. By "form" I mean area and/or line, but no color and no instructions on how the lines/areas are "interpreted". I investigate somes ideas about "form idea creation" in Section 2. A sort of wheel for "interpretation" I have already made. It is called Repligator and was my first attempt at making an image idea generator. I will discuss Repligator in Section 3.
2. The 2nd Llullian Wheel : Form
By form I mean line and shape but not color. This is in order to keep things simple for the user, though I realise that an interaction between form and color would be interesting. How can we encode and/or create forms? The following sections give some suggestions, which could be used together or combined into a single method. 2.1 Primeval Forms There are some forms which seem to immediately appeal to the eye, mandalas, Persian carpet designs, Celtic art patterns, "sun" images, cathedral windows, backgammon boards. Figures 1 to 4 show some examples of what I mean by "primeval forms":
Figure 1. Rose Window
Figure 2. Celtic Spirals and Tiles
Figure 3. Classical Sun Figure
Figure 4. Backgammon Board
If we could create a single function which, with the appropriate inputs (or genes or parameters) was able to re-create these primeval forms, then the same single function may, with other parameters, create as yet undiscovered "primeval" forms.
I think that it is important that the function is a single one, and not a collection of say, a mandala function, a celtic spiral function and a church window function. It must be a single function which allows, in some as yet undefined, undiscovered way, infinite variation from a rose window of a church to a backgammon board.
2.2 Alexander's "The Synthesis of Form"
In  Christopher Alexander describes a "mechanical" method for the synthesis of form. He was writing about the layout of towns and cities, but the same ideas could be applied to pure images. He points out than "self concious" (i.e. "modern", "western") design involves an architect who wants to (even subconciously) apply a certain style or vision on the building or village he is designing. This limits the ideas he comes up with. On the other hand "primitive" ("traditional", "unselfconcious") design, after evolving with many generations of crafstmen, comes up with good (if not perfect) solutions to the problem. Figures 1 to 4 are good examples of tradition coming up with solutions, figures 1 and 3 being solutions to artisic problems, and figures 2 and 4 being solutions to artistic/practical problems.
In the case of "unselfconcious" design however there is the problem of "tradition" which limits the exploration of new methods. In this book Alexander puts forward the idea that a purely rational system of design would come up with the optimal solution, and with no limits on the outcome (which are normally set by "primtive tradition" or "egocentric style"). This echoes Llull's idea that a purely mechanical device could come up with all the truth in the universe. Alexander's method is iterative, initially setting up certain areas (car park, industrial zone) and weighted connections between these areas (the car park should be near the industrial zone). Every iteration creates a selection of solutions, and the best are selected and used for the next iteration. There is a fitness test which can be applied to each solution generated. Alexander points out that another consequence of using his method is that the "designer" cannot predict the outcome.
It may be possible to use these same ideas for the creation of two dimensional images, but many questions need to be answered. What is the "fitness" of a given design? Who invents the rule that "the small square should be near the large crescent?" . Here the "artist" would need completely different skills than he has today.
2.3 Shaminal Forms
A shaminal (SHape ANIMAL) is a form or shape which reacts to other shaminals. It knows where it wants to go and it knows what shape it wants to be. We can imagine putting all the correct shaminals of a mandala in a square space at random, and, with each tick of a computer clock, each shape moves slowly into its proper place and shape. What would happen if we mixed mandala shaminals with backgammon shaminals?
3. Repligator, an image idea generator.
When I first started writing Repligator I came up against a big problem. I wanted to generate image ideas, but what do I start with? If a user wants to generate images how is she to specify what she wants? How can a program start from nothing and yet be guided by the user? Now I know I should investigate the three Llullian wheels, but initially I had no idea. Today Repligator is a program which takes an original image (and this is the seed, the user input, for the image ideas generated) and "interprets it" by applying various effects. Here is a screen shot of the program:
Repligator is the last wheel of the three Llullian wheels of color, form and interpretation. It does not yet really understand "form" or "line", that is for the future. The use of Repligator by "the world out there" has undergone three phases:
1) Designed initially to be an "image idea generator" the users did not understand how it could possibly be that, and so they used it as an "easy to use effects" program.
2) Responding to the user's feedback I started calling it "Repligator : Graphics effects with ease". I was trying to get image ideas generation "in by the back door" so to speak. I left in the idea generation functions.
3) Now I get email from users saying "I use it to come up with new ideas when I am stuck or need fresh input", they are using it for what it was originally designed for.
Repligator as an idea generator works as follows:
1) Bring in your original image, which may be a scanned photo, or an image created by a paint program, or whatever.
2) Select the options to "let the Wizard choose". Inside Repligator the Wizard is a slightly crafty selector based on a random number generator. The wizard's name is Randy.
3) Hit the F7 key (or the appropriate icon in the tool bar) to get the Wizard to generate a new interpretation of the original image.
4) Every time F7 is hit the Wizard chooses what effect will be used on the original, the settings for the effect, and if the original and transformed image will be mixed together in some way.
5) All the images generated are saved and placed in an "image idea sequence", which can be reviewed at any time. No generated images are lost, this is like having an unlimited "undo".
6) Once you have a few images you can refine the Wizard's settings of the effects using dialog boxes
7) One last main manipulation: you can use Repligator's "internal copy/paste" to transform an already transformed image, thus taking it even further from the original. The transformed image become the "original" in a new sequence of image ideas.
The important thing to remember about Repligator is that the mouse is never used to draw anything.
There are three image quality levels, the lowest having no anti-aliasing, the highest having good anti-aliasing. A long sequence of images can be quickly generated at low quality, and then the most pleasing ones regenerated at a higher quality, which takes longer of course.
What makes Repligator different from a graphical effects program? Well, in fact, it can be used like that, and is used like that, but there are differences:
1) It is very very easy to use. There is no need for ability with a mouse or tablet.
2) It is not a "plug-in" but a stand-alone program.
3) With single key strokes the internal Wizard chooses transforms to apply to the original image. This stops the user getting stuck in a rut. (By the way some users have requested that I allow then to exclude effects, but I am resisting this because that becomes a (personal) tradition in itself, which creates limitations in what can be generated.)
4. Future developments
I am currently working on the first of the Llullian wheels: Form. I have to create (or find or evolve) that single function which will give me plaid designs and mandalas and rose windows and all the variations between. The color wheel will come next, and I do not see too many problems with that. Repligator itself will have to be re-written from scratch to take advantage of the new form and color programs. I would like it to be able to interpret lines as rows trees for example, and areas as ploughed fields.
Thinking about things from a commercial point of view, despite the rush into sexy three dimensional graphics, I still think the computer has at least as much potential in generating two dimensional images. I would like to get to the stage where a non-artist can design her own unique carpets, textiles, wall-paper, tatoos etc. Where she can say: "Show me a Tibetan buddhist mandala, interpreted as in the illustrations of the Celtic 'Book of Kells', but using the colors of Cluade Monet's 'Waterlily Pond'" It may even be possible in the future to have these carpets and textiles woven and printed at your local supermarket.
1. "Notes on the Synthesis of Form". Alexander, Christopher. Harvard University Press 1997.
2. Repligator, from Ransen Software, http://www.repligator.com/