On Banding SolutionsThe pixel artist that works with more than two colors will undoubtedly run into banding problems. Here are some useful approaches for clearing out banding in one's pixel clusters.
First let's talk about the problems a little.
A is the case where a mostly regular cluster has a few irregular edges. The instinct of many pixel artists is to 'follow the shape' with anti-alias so it's 'shaded' while still retaining its irregular identity. This impulse isn't the best solution. Squares are strong in pixel art. Pixel art does perfect lines better than any other medium we know of. Absolutely crisp and straight, 90 degree angles in pixel art as a strength. The good pixel artist subconsciously is drawn towards the platonic solids they can present in their art, so to 'fill in the form' is a possible solution. However we see in the example above that the artist has exacerbated the problem by introducing banding. Instead of a perfect shape he's left with a very visually striking reminder of the shape's imperfection. The blemishes at the edges have been made doubly apparent because
banding draws attention to the grid.
Let's talk about B last.
C is slightly more complicated combination of clusters where the artist is both trying to 'shade' the irregular contours of the shape and at the same time taper its end off into its own cluster to suggest a gradation or fade. To use contour anti-alias (single or double-pixel rows going around regular angles) that introduce a new cluster is a common reflex in pixel artists, however by introducing banding here what is instead achieved is a very strange visual illusion where the pixel cluster is blurred and doubled. I can't explain this any better, anyone who has spotted the perils of banding can see what I'm talking about.
B is a composite problem, where there's both a 90 degree boxy shape conveyed but there's also sloping and a free-form taper edge. Banding here has made every problem that much worse.
I want the reader to stop a while here and consider how they'd solve these problems. Go in your pixel art program and make do as best you can, commit to an approach on how to deal with the banding issues apparent in these examples. Then come back and read the rest and see how your choices parallel mine.
Here's a set of solutions.
On
A you'll see I opted to double the anti-alias vertically.
This effectively makes the antialias from contour anti-alias to separate pixel clusters on its own. This might seem counterintuitive at first but it looks much better than the banding. It suits however only the situations where the form can be elongated in either axis like that. Generally keep in mind:
When pixels of close shades touch in a systematic manner, there'll be banding. Although it seems counterintuitive, using clusters that are not one-pixel thin, less colors and therefore stronger segregation works better than banding because it suggests there's an orderly system of stacking that isn't dependent on the pixel grid, but on the number of available colors. If you have to choose between the two in pixel art, I posit that
it's better to draw attention to a small color palette than it is to draw attention to the grid.
On
C my solution here is to remove most of the contour anti-alias that would connect it to the separate cluster. Again this feels counter-intuitive but art with a grid has special considerations to keep in mind.
Best to deemphasize the introduction of a new color cluster through anti-alias *at the connection* between the two, otherwise between trying to anti-alias one form into another and introducing the new shape gradually, banding will occur.
On
B my solution is as expected, composite. I identified the perfect angle on the right slope of the form and gave it one-point anti-alias. I made the irregular nooks sharp and readable and I suggested a slight curvature at the bottom of the identified square with a single pixel's worth of anti-alias. I broke up the line on the bottom of the form with the anti-alias because I judged any other solution for that configuration of pixels would lead to more banding. That last bit is effectively, sub-pixelling.
Here's a different set of solutions:
As you can see here anti-alias is mostly removed. This is a different approach but a valid nonetheless. On
A note that the 'platonic square' that the cluster desires to be is again empathized and no banding is occurring. On
C not how the two clusters (and no aa, at this point) interlock as if they're shaking hands: that's an elegant solution to the edge between clusters issue. In your paint program, add a pixel on the top or bottom of the 'handshake' to see what I mean about it. On
B the minimal AA brings out the cluster shape loud and clear, but it also brings attention to the grid. It's a cost/benefit choice.