The point of this page is to document the steps taken to create an entry I recently submitted to a contest on Fark. The contest was titled "Photoshop this lighthouse lashing" if you want to have a look at what everyone posted.

I've created a bunch of content entries following an undirected, time wasting, collage approach and that's what I did in this case. So I wanted to show the intermediate steps as the bits from the original image turn into an image that doesn't bear much (any?) resemblance to the source.

First I pick a part of the original image to work with, usually something with a geometric shapes, a repeating pattern, or some clear representation of movement. High contrast, or at least sharp lines in at least one RGB channel, seems to be helpful too. The photo proposed to start the contest was a very dramatic image of wave crashing into a lighthouse. From that image I pulled out this section,

Then the time wasting begins... Generally the process is to reduce the opacity of the source fragment a lot, stack together a bunch (say 6-20) copies arranging them so that they form a pattern. Maybe arrange them as if they were moving along an arc, maybe rotate them, or some other iterative mod. Sometimes I'll change the size, color, bright/light, or perspective as well. The idea is to make a series that follows some predictable transformation as they go along.

For this one I tweaked the opacity by doing what could be described as "stamping the alpha channel". And by that I just mean replacing the alpha channel with an image in order to map the opacity to some pattern. So instead of an area that's 100% opaque with the rest of the space needed to fill in the rectangle 0% opaque, the area that was 100% opaque is now a patterned gradient. An example is probably worth more than 1000 words here... So here's what I did for this one.

1. Starting the small image linked above, decompose the image to CMYK channels.

2. From the same image, decompose the image to RGBA channels.

3. Copy the "K" (black) channel and paste it as a new layer in the RGBA image.

4. (I skipped this step by accident, oops!) Usually I edit the new layer (copy of "K" channel) so that the 0% opaque areas in the "A" (alpha) channel are completely black. The reason to do that is to prevent details from "invisible" parts of the source image from coming back. Have a look at the RGB and CMYK channels of an image after decomposing and you'll see all the detail is still there... Anyway, this time I forgot that step and ended with some unintended artifacts showing up later.

5. Compose a new image from the RGBA window, but use the new layer (copy of "K") as the alpha channel. That yields this watery version of the source image.

Not much to look at, eh? But it's just a building block... And now the OCD'ish fidgeting kicks in. After some trial and error I ended up making a stack of 6 copies of the image above, arranged as if they were moving to right and a little down, compressing the size and rotating each copy a bit more. That ended up looking like this.

The unintended re-appearance of info from the original image is clearly visible as a hard bluish black edge (and corner) at the top left of each image in the series.

Next I duplicated the image, flipped it vertically and aligned the two copies to create a symmetric shape. The images generated by this "pile on" process tend to get very large, so I've scaled down the rest of the images on the page. The full sizes of all of them are available in this gallery.

And flipping a copy horizontally and playing around with arrangements, resulted a closed, oval shaped image. Some changes to color levels, brightness, contrast, etc... changes were also applied.

There was a lot of messing around that this point and most of those images were discarded. I mentioned that bit about this being an undirected process that wastes lots of time, no? The element that did wind up going forward was a simple change to the color palette and some amount of "zoom" motion blur.

Next the image opacity was adjusted selectively make it look more "skeletal". I think there are some other color, brightness, contrast, etc... adjustments as well. And a second copy of the image was added and mirrored horizontally. I've scaled the size of the image down to make it fit better on an HTML page. This "pile on" collage process tends to produce very large images.

Then two copies of that image are stacked, one rotated 60 degrees, and the second rotated 120 degrees. The result is a sort of transparent donut (or inter-tube?) shape.

And in preparation for more stacking, the transparency is adjusted down again. The second image is what it looks like against a dark background. Since the color levels are so faint, all the rest of images will be on a dark background to make them easier to see.

The next set of changes is long and tedious, so I think an animated GIF will explain it much better than text. At a high level, it's a stack of 28 copies of that skeletal yellow donut. Each one is rotated, shrunk and shifted up and to the left of the larger one underneath it. Like I said, the animation will help more than the text...

The end result looks something like a wireframe cornucopia, or a clear insect wing.

After more playing around, with many experiments leading nowhere, a simple stack of two copies of the wing shape looked promising.

And stacking mirrored copies of that image both horizontally and vertically, led to the basic framework for the final image.

That's pretty much the end of the unusual parts. The remaining steps are just selecting oval shapes and adjusting the colors to suggest an orange/red reptilian pupil, a blue/purple iris, and red patterns like bloodshot eyes.

And color of the background image was adjusted using the same oval selections to reinforce the suggested eyeball elements.

Another background color change was done to represent eyelids and the "bloodshot" area under the eyelid surface was adjusted as well. The opacity of the outer yellow frame was also adjusted to make it brighter.

The last two elements added to the image are two "zoom" motion blurred copies of the outer layer of the image, one is gold/orange color and a smaller one with a white/blue tint.

The final image is four copies of the finished eyeball with the perspective changed to soften the direct, dead-on, flat feel.

A larger version and a much larger (6463x5940) versions are available as well.