How do you start an image from scratch?

Techniques of three artists:
Janet Parke, Damien Jones & Kerry Mitchell
posted to the Ultra Fractal Mailing List on Tuesday, May 18, 1999

Janet Parke

No one has really asked me this question, but I'm not going to let that stop me from answering it. ;-) Several people have commented lately on how difficult it is to start a new image from scratch. Since that's pretty much all I do, I thought maybe someone would be interested in how I go about it. If not, I apologize for what will undoubtedly be another long message.

I feel like I've said this before, but maybe not on this list -- when the first UF beta was released we couldn't yet import FractInt pars. Plus, due to technical difficulties of one sort another, I really didn't have any FractInt pars to import (about 2 dozen total in my lifetime -- none of which are the least bit memorable). Soooooooo... I had to learn to work from scratch and over the last 9 months or so, my strategy has evolved into this:

- I have set the Grayscale gradient in the Standard folder as my default. That way when I am first beginning a new image, I can concentrate on shape rather than be distracted or put off by color.

- I pick a formula and start looking for an interesting shape. If I'm working with one of the Mandelbrot/Julia sets which incorporate the switch feature, I often open the Mandelbrot one, hit the spacebar or the Switch icon on the toolbar and use the eyedropper to find interesting spots in which to zoom.

- Once I've found a reasonably interesting location, I start trying different coloring algorithms. Up to now I have worked mostly with Damien's formulas and with a few miscellaneous others from Ron's, Kerry's and Mark's collections. There are now so many from which to choose that one could spend days on this part! Anyway, this is not to say that other formulas are not good, I've just been concentrating primarily on Damien's lately.

- I like to have a layer available that has a smooth-ish coloring -- something like Exponential Smoothing or Smoother Iterations. Then I usually try (from Damien's collection) Cilia, Curvature Average, Lyapunov, Triangle Inequality Average and lastly, Orbit Traps. Every time I find a look I like, I keep that layer and add a new one. I don't worry about whether or how the layers merge, I just sort of gather a list of layers that have nice, cool, or unusual effects. I save the Orbit Traps formulas for last since I'm pretty sure I'm going to find plenty to work with in them.

- After I've got anywhere from 3-? layers that show potential, I start playing with color. I pick a layer, toggle the others off so I can see what I'm doing, open the gradient editor and press Ctrl-3 to give me a random misty gradient. I work in HSL mode only (press Ctrl-H with your gradient editor open) because it gives me the control I want. If I like the random gradient, I keep it. If I don't, I keep pressing Ctrl-3 until I find a color combination that I like and that has control points in vaguely the right places to show off the design -- all this gets tweaked later anyway. I don't ever save gradients because they seem to be linked inextricably with the layer and fractal for which they've been adapted.

- Sometimes I do a new random gradient for each layer and just see what happens, but often I copy that first gradient to several other layers and start playing with merge modes. I may or may not use all the potential layers in my list.

- Hopefully somewhere along in here some form or "look' begins to emerge. If it is another one of those darned flower-shapes that seem to be appearing so often in my images lately I try to resist it, but often at this point, the fractal is leading me and I try to enhance the direction in which it seems to be moving rather than try to change it. Is this getting too cosmic?

I don't really plan images ahead of time -- or at least not successfully. I think it's more important to just be able to recognize what has potential and what doesn't and then be willing to patiently work/tweak it into whatever it wants to become.

Anyway, that's what works for me.

Damien Jones

Oooo, this sounds like fun. Can I try it too? Can I? Can I? :-)

I always start from scratch. I haven't changed my default gradient to grey, because for me it is easier to see the potential for color if there's at least some color there to begin with. But I generally pick a formula, almost always a Mandelbrot type.

At this point I usually start playing with the formula parameters. Currently I'm doing a lot of exploring of the DoubleNova and DoubleHalleyNova types, so I'll often twiddle the exponent or scale values in this formula. During this stage I am looking for interesting shapes, things I might not have seen before. I may pop open the Switch mode here, and look through Julia shapes. I may do some minimal zooming, usually not too deep.

If I feel so inclined, I might tinker with some transformations. I'm not actually that big on transformations; I do sometimes use them for effect, though. Most of them change the shape of a fractal enough that I feel they should be applied before I really start exploring.

The next step is to experiment with coloring algorithms. I also tend to try out a few algorithms, although my choices are often directed by my limited understanding of the formulas. I usually try Triangle Inequality, Lyapunov, Kerry's Statistics, plus a few others on random impulse; if I can't find something that immediately grabs me, I'll fall back to Orbit Traps, which almost always can produce an interesting image with some set of options.

Unless the default options on a coloring algorithm produce stunning results--and they seldom do--I will experiment with the options for it. I do have some favorites I try, such as using "real(z)" for Statistics. If I find something with some promising detail, I'll continue to zoom at this point. If I find an interesting coloring algorithm or combination of options, I'll sometimes add another layer and keep exploring. I might delete the layer later. At some point during my experimenting with the coloring algorithm, I will start toying with random gradients (Randomize Misty is my favorite) to try to bring out new possibilities and features. I'm still looking for structure here rather than final colors.

Once I get here, I'm looking at a fractal with new formula parameters, new coloring parameters, and some random gradient. Possibly with two layers, although usually just one. I've seen some shape that interests me and I want to bring out the detail, or subdue the detail, to make it what I really want. I'm moving away from aimless exploration and towards refinement and coercing the fractal into what I picture it in my mind to be.

So now I'm going to reach into my bag of tricks, techniques that I know work fairly well, to try to tweak the fractal. I will often try cloning the bottom layer, then changing just one or two parameters in the coloring method, and combining the two layers (with the same gradient) with Hard Light or Overlay. This is especially effective if the parameter that is changed has a natural complement; for example, if one layer has real(z) as an option, I will try the other one as imag(z). If one layer uses distance coloring for orbit traps, I will try the other one as angle coloring. I don't always do this, because sometimes the combinations are bland or uninteresting, but these are things I try.

If I'm in a particularly patient mood, I might try a multi-layer monster. Generally these are ten or more layers, carefully blended so that each layer has only minimal impact, where a single parameter in the coloring is adjusted over a small range throughout all the layers. This is how images like "Ethereal", "Ephemeral", and "Gorgon" are done. If you're going to try this, you will need the following: plenty of RAM (close those other apps), a fast processor (since you're rendering ten times the normal amount), and patience. And expect to throw a lot of them away, as these are *very* tricky to make work.

The final stage for me is color tweaking and merge mode experimentation. If I'm reasonably happy with the colors as they are, I may just tweak a bit, here and there, as sort of a "touch up". If the colors are uninspiring but the shape is promising, I'll try some more random gradients, and tweak when I find something with promise. I almost never use a random gradient without adjusting it.

If the image has a lot of layers, or has layers with different gradients, and I only want to adjust the colors without replacing them completely, I will often add a "color adjustment" layer. This is a layer using the standard Mandelbrot type, 1 iteration, bailout 1e20, with the inside color set to (0,128,128) in the HSL color model. Then I change the merge mode on this layer to HSL Addition and Opacity to 100%. In this merge mode, the color (0,128,128) is "neutral", and has no effect. I can adjust the hue for the entire image by changing the solid hue, and I will often try all the hues in steps of 15 or so. I can desaturate by reducing the saturation of the solid color, and supersaturate by increasing it. I usually don't adjust the lightness with this layer, since that is normally part of what has attracted me to the image in the first place.

Another trick I might try is to invert the lightness without changing the hue. To do this, I use a hue-adjustment layer as described above to rotate the hue 180 degrees; then I add another color adjustment layer, solid white, and set its merge mode to Difference. This inverts the lightness and rotates the hue another 180 degrees (back to where I started). I do these with layers rather than by tweaking gradients because it is often difficult or impossible to achieve these effects by modifying several very different gradients.

At some time during the exploration, often before I begin tweaking colors, but sometimes after, I will change the framing of the image. There are several sizes and ratios I use, 4:1, 8:3, 4:3, 1:1, and the portrait versions of these ratios. The thinner ones are harder to get right, and it is often difficult to persuade a fractal to fit them well.

I almost never name a fractal until it is finished, although there are a few notable exceptions. (Not all of which have been posted.) In some ways this can be harder than creating the fractal. :-)

Kerry Mitchell

Here's what I do:

I also have a default gradient that I use. It's actually a fully saturated hue spectrum, with the green negated. (I find that I don't much care for computer green.) For me, the actual gradient doesn't matter so much, but if I use the same one over and over again, I can use the colors that pop up in an image to calibrate how the technique works.

My favorite images typically come from Julia sets, and I have a stable of Julia parameters that I use over and over again. Some of them are: (-.778,.201), (-.1,1), and (.28,.005). When I settle in to an image, I have hundreds more stored in text files (from Fractint usage) that I can cut and paste into UF.

For me, my images almost always have some "deeper" connection between the parameters and coloring schemes. For example, I may use the polar angle as a hue layer and the corresponding radius/magnitude for a saturation layer. I find that using multiple aspects of a given coloring scheme in concert can lend an element of cohesion to an image.

I often use many layers (8 or more). This can help reduce the effects of banding for those coloring formulas that don't have continuous versions. This also works out well with formulas like the "embossed" types, that don't necessarily have iteration bands but can still have sharp boundaries. One thing I like to do is to use several layers, each with black/white coloring, and average them into a continuous range of grays.

I tend not to use transformations too much unless I'm going for a special (read: premeditated) effect. The apsect ratios I typically use are 4:5 (for images that I might want to enlarge to a 16x20 print), 1:1 (square) and 3:2 (works well with photographing the monitor on 35mm film).

Advice for newer folks and anyone else who cares to listen: pick a few favorite tools (coloring schemes, calculation formulas, parameters sets) and work with them extensively to figure out how they work. Don't try to use and understand everything at once. Go back to the basics often. Play.

Back to Ultra Fractal Resources

Copyright © 1997-2009
J       a       n       e       t             P       a       r       k       e