Pixel Spaceships is retro pixel pop art. Spaceship-like forms are algorithmically generated by the hundreds, even thousands! It is an exploration of generative algorithms, an examination of limited perception and the power of suggestion, and intended to evoke a nostalgia for the artistic sensibilities imposed by the hardware limitations of early arcade machines. Soon to be released as a long-form generative piece. Release March 11, 2024: fxhash Pixel Spaceships project page.
GENERATION / ALGORITHM
Generation is entirely from scratch, based on a pattern of bits, no pre-made assets are used. The work area consists of a 14x14 grid, so there are 196 total "pixels" available to consider, giving 2^196 possible combinations of purely random on/off monochrome pixels. 2^196 is roughly 1e59, which if fully written out would be a 1 followed by 59 zeros. For comparison, that number is more (far more) than the number of seconds the universe has existed since the big bang! That poses a problem of complexity. Not only is the number too large to conveniently explore fully during our lifetime, but the vast majority of those combinations would utterly fail to satisfy the principle criteria of "resembles a spaceship".
So it will be necessary to (vastly) simplify the problem. In so doing, some of the possible "range of expression" will be sacrificed in order to focus efforts on what remains as most-productive for these purposes. In the end, such an approach will not be capable of generating every possible combination that might look like a “spaceship”. But of the limited range that it can produce, there will be a much higher likelihood that they are "spaceship-like". This is an entirely acceptable, even desirable, trade-off. (spoiler: don't worry, even after range reduction, there still remain a ludicrous number of satisfactory results)
The first simplification will be to impose strict horizontal symmetry. This will reduce the area under consideratio to 7x14, or 98 pixels. A significant improvement, though still having roughly 8e28 possible combinations. The second simplification will be to impose a "mask" of potentially-useful pixels. The particular mask chosen (lovingly known as “the avocado") leaves 62 pixels still available for direct manipulation, capable of roughly 4e18 possible combinations.
This is now finally into the realm of somewhat "reasonable" numbers, where the entire possibility-space could be encoded within a single 64-bit value. In practice, additional "adjacency" constraints are further imposed during generation, using on average 2-bits per pixel to fully resolve. So four 32-bit values, or a total of 128 bits, are used for "structure" during generation. Another 32-bits are (optionally) used for coloring, and a dozen or so more bits from another 32-bit value are used to determine various additional styling. (aside: the color/styling bits are not used “optimally” - both could be resolved with a single 32-bit value,; it was simply “convenient” to split the functionality with separate values) All told, these 6 32-bit values completely define a single spaceship.
STRATEGY: COLORING / SHADING
Unapologetically: this is retro, this is low-rez, this is pop art. The use of color herein may seem “garish” to sensitive viewers. Please simply avert your eyes if it causes excessive pain! You’ve been warned. But for the more adventurous who remain…
Coloring and shading were implemented so as to mimic hardware limitations of early arcade machines. Ten separate color/shade combinations were created based on (roughly) emulating the concept of “bit depths”: 1-bit monochrome and a bold variation; 2- and 3- bit grayscale; 2-, 3- and 4-bit tones and colors. All color/shade modes are capable of self-adjusting to both positive (on-white) and negative (on-black) presentations. In addition, the black and white points auto-adjust if/as needed when other “decorations” (e.g. an overlaid pixel grid) are present that need some spare tonal range for their own styling.
STRATEGY: AUTO-CURATION
Even with a thoughful toggling of selected bits, there remain some possible output forms that may seem less-suitable than others. In order to reliably produce suitable forms in the quantities needed, the algorithm includes two stages of internal “quality assurance”. During generation, if certain “potentially problematic” sub-patterns are noticed, then extra steps will be taken to try to ensure that any such problems are mitigated. The “bug eyes” form shown here is one such “potentially problematic” form - it’s fine as a form unto itself, but at times may not connect well with the remainder of the form. A second step of QA occurs after generation, whereby the entire form gets a quick looking over. Particularly if the result is destined for “spotlight” use (as a “hero” or other large presentation) then it may be discarded entirely and regenerated from scratch all over again if found to be deficient.
STRATEGY: MASS PAREIDOLIA (wiki)
Merely by reading the title of the work: "Pixel Spaceships", the viewer has been subtly conditioned toward a preferred interpretation of the presented images. Whether or not any of the images actually look like actual spaceships is largely irrelevant - the viewer will nevertheless be slightly more-inclined to perceive "spaceship-like forms" once this suggestion has been conveyed.
This is the same effect whereby two dots and a line may be perceived as a "face". Human perception is already strongly biased toward facial recognition, so this effect should come as no surprise, even without additional conditioning. What is potentially more interesting, is the degree to which this "bias of perception" might be further manipulated through suggestion, even for themes that are not so naturally pre-wired.
LOOK! IT’S A PICKUP TRUCK! … NO, WAIT, IT’S NOT A TRUCK. IT’S HAIR AND EYES.
The smile from previous smiley was added as reinforcement. But now go back to the "just truck" image.. what do you see there now? Are you now at least somewhat conditioned to see it as hair/eyes instead of a pickup truck? At the very least, it perhaps leaves you more open to either interpretation.
The viewer’s brain will likely (i.e., just “by default”) try really hard to find “faces” amongst these symmetrical “spaceship” pixels, so at least an attempt will need to be made to help mitigate that to whatever degree possible.
"Pixel Spaceships" presents the ships en masse as part of a deliberate strategy to help reinforce this type of perceptual conditioning. It also helps make up for any occasional "dud" that might sneak through the generation process. That is, if one of the generated images doesn't quite convey "spaceship-like" to the viewer, the presence of many others nearby that do positively reinforce the "spaceship-like" perception may help ameliorate the deficient one.
The generation algorithm is pretty reliable overall, perhaps 999 out of 1000 properly read as "spaceship-like". (Caveat: Perception is very subjective, so I have no actual measure, so this is just a guestimate of the generator's success rate.) But that thousandth one may need the benefit of this extra level of assistance to help convince the viewer to see it as intended.
In traditional art, this would be akin to a landscape painting where near objects are depicted with extra detail so that distant versions at reduced scale can be conveyed with much simpler depictions. For example, the viewer may be conditioned to believe that an "orange dot" in the distance conveys the same meaning as a "detailed orange flower" presented up close.
Given the tiny format, most pixel-art relies on at least some degree of "seeing more than is there". The masters of hand-crafted pixel-art can often convey very different forms through extremely subtle manipulation of just a few individual pixels. "Pixel Spaceships" can't replicate such proficiency, but does aspire to at least be suggestive of it - the mass presentation aids in this goal.
In the end, it's all smoke and mirrors, and you are being “deceived”, but such is art - to a large degree it "is" whatever you "see" it is. "Pixel Spaceships" deliberately exploits this form of trickery and invites its further contemplation.
If you happen to find a truly remarkable “dud”, that utterly fails to read as “spaceship-like”, feel free to share it among friends as a curiosity, to perhaps test the extent that the (mis-)perception is shared. Perhaps tag with #pixelspaceships #shipornot. You might also try a “clever” caption to further impart an intended (mis-)perception bias. It could be an amusing and enlightening experiment! (I’d only ask that you please “mind your manners”… follow?)
STRATEGY: IMMENSE VARIATION
As alluded to above, despite strongly restricting the generator so as to be more productive, there remain an immense number of possible outcomes. Let us admit that many of the 2^62 bit patterns may generate superficially similar looking ships, so instead conservatively estimate the number of “usefully distinct” bit patterns by halving the exponent to a more-modest 2^31. Then multiply those by the twenty different color/shade strategies. Then multiply those by some ridiculous factorial of possible layout combinations (that I won’t even begin to try to calculate). Then multiply by other minor “styling” variations such as the type and style of border and et cetera. Then multiply by several other “secret” features that haven’t been discussed here. All told, the number of possible outcomes resulting from the combination and permutation of such “seemingly-simple” basic forms taken en masse is truly mind-boggling.
Only the tiniest fraction of this possibility-space will ever actually be produced by a limited minting, but that is intentional - to help ensure that the few that are produced at least have the opportunity to be quite different from one another. Granted that, in the largest sense, they’ll all be very much the same. That is, they’ll all be “collections of pixel spaceships” - it’s not like one will magically transform into something entirely different like a plasma shader, or circle-packing! But in the detailed sense, it should be exceedingly rare that any two could be similar enough in their particulars that one would ever be confused for the other.
HISTORY / MOTIVATION
This idea has been floating around in various forms for the better part of 40 years now! Earliest attempts were at 8x8 pixel resolution…
Circa early 1980’s: A computer lab had just been created at my high school, adjacent to my math classroom. My math teacher offered an open extra-credit assignment to “create a spaceship and be able to move it around on screen” as a way to promote the lab. At the time, I wasn’t as interested in the movement aspect, but the design aspect intrigued me - particularly given the “unweildy” character ROM graphics. I asked if the extra credit might still be awarded if the design aspect alone was explored, in depth. In the end, I turned in not only enough ships for the entire math class, but more than enough if even the entire school had participated!
The concept was later revisited at 10x10 resolution circa mid-1990’s, then at 12x12 resolution circa 2005. At each (seemingly minor) increase in resolution the generation becomes considerably more complex as the number of possible outcomes increases dramatically. This latest revisit (2024) will likely be the final stage of that evolution of algorithmic exploration and expansion (at least for a while!), though I may continue to explore further “presentation” possibilities with the current algorithm.
