 |
| ||||||||||||
I probably have a different mental picture of what banding refers to (do I?), but at this point I see it simply as the visual effect of a line created where two borders color. Ofc stuff like staircasing will amplify it in places that makes no sense. I'd say your approach bands more, though maybe only as a technicality.
True. I think my approach is the exact opposite -- "AA everything that doesn't need to be sharp." -- I avoided adjusting the AA on 'taut' parts of her legs for that reason.
Anyways, you can guess I'm not a color space geek, for better or worse.
Your edit is intersting. While there were definitely a couple of spots I missed as suggested in the description, you went with a slightly different approach to AA to one I use, one that doesn't concern with 'AA banding'. Though to be fair, the banding pixels are so low contrast the visual effect is practically non-existent.
My view of things as to why my AA approach is the way it is, while I was experimenting with AA with Ptoing's approach in that one tutorial, I try positioning an edge the closest I can to having AA that practically resembles an approach that avoids AA banding. If I position a line just right, the darkest AA pixel and the lightest AA pixel can be rounded off to not use them at all. I find this advantageous as I realize that every AA pixel results in unavoidable blurring. In some places in this piece I add AA mostly to help maintain a consistent level of blurring.
I work with the HSL color space graphicsgale gives. I just referred to the number for the Lightness lightness slider because I have no better word to refer the number as. (A value that describes a value of a color? lolol)
The thing about your ramp was that while I was color picking your ramp, I was using the HSL color space of Graphicsgale. I currently have no known access to LAB. They gave me value-values/luminescence numbers that increased rather linearly. This was counterintuitive because the point of gamma-correction is to be aware that a linear change of power input (or whatever monitor geeks refer to that) does not mean a linear change of perceived brightness.
Lum values? Which values do you mean? Luma (Aka Y of YCbCr) with rec.709 coefficients, Luminance (one of several non-equivalent formulas, , Lightness (of HSL) or Lightness (of LAB, also often incorrectly called luminance)? They are all different,. Hopefully you are using a program that doesn't label it simply "L" :)
My ramps were calculated as a linear gradient through LAB space, which is a colorspace designed so that a linear change in values produces a linear change in 'perceived' brightness/hue/saturation. The result is that the RGB values usually change nonlinearly, although for a low-contrast enough gradient they may turn out linear.
Anyhow here's my quick AA edit.
http://i.imgur.com/c2hwCsA.png
You really have a lot of curves in there. Anyway I mainly checked the diagonal + contour AA here (ie. every 45degree line, and whether a given AA color has 'continuity' as you move along the edge of the contour.)
Putting personal taste aside -- eg. whether this is 'over-AAing' -- , I haven't really got much more to say about the AA in this piece, my edit says it all.
Interestingly, what I saw on my monitor in the suggested ramp was not a straight line, but a curve going the wrong direction... pretty much what I expect to see with a ramp that assumes a gamma of 1...
...and for some reason I picked colors on your ramp and saw the lum values increase linearly?
There's a missing pixel on the right (two yellow pixels in the upper, only one in the lower) and most of your AA ramps are missing the final(least opaque) pixel. Also to be 'strictly equivalent' you would keep the color info and only modify the brightness.
(there's also the complication that you're rendering curves, not triangles, in some of your AA, so not all progressions should be mathematically linear.)
I'll work up an edit for that.
BTW the side-by-side colors in the calibration chart you made look very close in brightness.
Anyhow I avoided giving the impression 'and that's bad' WRT boosted contrast - it's an aesthetic choice which pops the lineart more, which is fine if that's preferred to mathematically accurate AA. It should work well on curves but less well on straight line segments.
Huh... I was about to think you're definitely on to something until I tried applying some of the changes suggested. Have you tried changing the AA on top of the upper mane section to the suggested ramp?
http://img14.imageshack.us/img14/2887/87057773.png
Someone else who gives a damn about gamma correctness
This is quite accurately AAed -- probably as good as you can do with a fixed palette.
EDIT:Or at least I thought so -- see the edit below
I also like the color choice -- specifically that green/purple contrast makes her eyes really pop.
With the help of a palette editor/color picker that does gradients in Linear RGB or (preferably) LAB, you can easily generate near-'ideal' gamma-correct gradients between two colors. I use GPick when I do this.
EDIT: here's an edit showing many of the transitions you use in comparison to the calculated 'ideal' transition:
http://i.imgur.com/o7xcTzO.png
It shows that for shorter/lower contrast gradients, your gradients have a close match, and for longer/higher contrast gradients, in the dark end you tended to pick colors that were noticably brighter than their 'reference color's, causing boosted contrast in the dark end of the gradient.
Nice Shading style here, you make me love ponies.Huhu
I don't get THAT many comments on dA so I have the time to read comments. Besides, comments are fun to read :p
Plus, I'm getting myself into the game art industry sooner than I expected, so I better start building more connections anyway.
Oh wow, you took my suggestion! Didn't even think you'd read the comment :p
Yay! I'm so glad to see you joined PJ, I'm sure your work will be appreciated here.
I dunno, this is really good at least in my eyes, but there will always be someone better then you and someone better than that person who is better than you.
"Somebody of my talent?" I dunno, that sounds hella weird considering that I know that much more talented artists hang around here. Just my point of view, haha.
Glad to see someone of your talent here, I'm really excited to see more of your stuff.
Lovely art! You did a great job man.
I give max rate to this (maybe because i love those ponies art... not the cartoon)
-
What exactly your definition is seems unclear: by the words you wrote I could consider my method as intentionally *adding* banding.
The way Helm explained it, banding occurs where 2 or more bars/strips of color are both adjacent and parallel or near-parallel, creating a 'block' or 'stuck together' effect.
Eg
1134
2211
3022
the 11,22 pairs band vertically (each creating the impression of a 2x2 'big pixel').
30 and 34 don't band with anything.
or a staggered layout like
11000
21100
22110
22211
in which each
211
221
Makes up a large 3x2 'pixel'
or simply
102
010
201
in which the two
10
01
's each band, and so do the
02
10
and
01
20
AKA why 45degree AA looks bad.
Note though that this is different from the 'diagonal AA implying curve':
200000
130000
014000
001400
000130
000012
In which the AA only bands at the areas
14
01
and
40
14
I agree that for higher numbers of AA colors, it gets harder to escape subtle banding, though it -is- avoidable: in this image, only the
12
21
block of pixels at the top left, provided for comparison, bands.
(This image I made is rather typical of the oversmoothed aesthetic found in many old Amiga demoscene pictures)
Edit: BTW, if you use Krita and create an image with 'scRGB' colorspace and 16bit color depth, you can see exactly what gamma-correct AA rendering looks like (just use the selection tools to build up a shape, then fill it).