Official Off-Topic Thread

[.TakaM]

Friends! A new and glorious season of Futurama has started!

[crab2selout.png]

...

Would those images with AA still be viewed at a multiple zoom?

[Photocopier]

Didn't anyone read the article?
"Despite coming out after the Atari 2600, the Cassette Vision is arguably less powerful, busting out a palette of only 8 colors and a killer resolution of 54×62 pixels. One odd quirk of the hardware is that these pixels could optionally be displayed in the form of right triangles — “half pixels,” as it were — and that’s why you see oddly smooth diagonals in the above video."

[Ai]

...

Would those images with AA still be viewed at a multiple zoom?

The output image matches the resolution of the input image, so whatever zoom was originally correct is correct for the output. (though I've also successfully combined this with eg Scale3x by adjusting a few parameters like Amplitude proportionately. GMIC implements Scale2/3x too, so it's easy)
Does that answer your question?

I'm aware it's currently too smooth. I suspect this method needs to be combined with my sprite vectorizer so that a balance between sharp and very smooth can be achieved.

[ptoing]

Didn't anyone read the article?
"Despite coming out after the Atari 2600, the Cassette Vision is arguably less powerful, busting out a palette of only 8 colors and a killer resolution of 54×62 pixels. One odd quirk of the hardware is that these pixels could optionally be displayed in the form of right triangles — “half pixels,” as it were — and that’s why you see oddly smooth diagonals in the above video."

I don't think that is how it works. The resolution for sure is not 54x62 pixels because on a CRT monitor that would be fucking tiny ass. I am sure it is 54x62 CHARACTERS, and those might be limited to filled and triangle stuff. And I am sure the diagonals are not super clean since again, NTSC can only display so many pixels. It's not like this thing shipped with a special monitor or anything.

[Helm]

Yeah obv. the sliced pixels are not vectors, but still, I'd like to look at some specs.

[Mathias]

So I was doing my usual periodic poke at the problem of automatic, respectful AAing of pixel art,
And recently began to understand how GMIC's 'anisotropic smoothing' function works. (gmic: http://gmic.sourceforge.net/)
. . .

Very exciting stuff, AI! I'm trying to get a coder friend to compile it for me so I can try it out.

[Ai]

So I was doing my usual periodic poke at the problem of automatic, respectful AAing of pixel art,
And recently began to understand how GMIC's 'anisotropic smoothing' function works. (gmic: http://gmic.sourceforge.net/)
. . .

Very exciting stuff, AI! I'm trying to get a coder friend to compile it for me so I can try it out.

Why..? There are precompiled Windows binaries. (http://gmic.sourceforge.net/gimp.shtml). Although you might get better speed by compiling your own, when running a 64bit OS.
Anyway, the GIMP plugin is the way to try it, definitely; running from commandline does not compare in terms of experimenting.

My latest iteration -- I'm fairly happy with this one. Process changed:

* Edit->Copy original image
* GMIC 'smart upscale' (also in the enhance section) with smooth=1, anisotropy = 1, sharpen = 40.
* downscale to original size with linear interpolation (okay, now we have the smoothed version. Time to restore the small detail)
* Edit->Paste, and turn it into a new layer
* Add a layermask to that layer
* Edit->Paste (into the layermask)
* Alt-click to see the layer mask (obviously this is only if you are doing it manually)
* GMIC 'Difference of Gaussians' (in 'Contour') variances 0.10 and 0.20
* Apply Curves with a contrast curve (points at coordinates 84,39) This will mask out the low frequency detail (lines etc), and enhance the high frequency (dithering).
* Export a PNG

^^ intentionally designed to be automatable.



Then I adjusted the layer mask a bit so that every pixel of water showed the smoothed result.
Then I smoothed the water in the underlying layer some more (amplitude = 2, iterations= 2, other settings as specified in my previous post).
Pretty happy with how that's working -- only a little touchup of the highlights is needed IMO.


Next on my list is to autodetect the usable colors for a pixel with reasonable reliability.
(that is, given a pixel and surrounding 4 pixels, choose an appropriate color ramp based on image color ramps.)
I think I can do this by scanning the image and constructing a weighted graph where the weights are
frequency counts: This pixel color shows up N times next to this pixel color. Then 'cutting' the graph where
frequency is too low, and also cutting where improbably large jumps in hue are made, then iterating over all nodes connected to the source pixel color,
to determine a list of possible destination pixel colors.
Then quantizing auto-AAed images back to the source palette may become generally easy.

[Indigo]

Just opened public beta for our facebook game, "Lucky Train"

The game will be updated quite a bit from this point, but feel free to play it up and post feedback to the fanpage.  The art was done by myself, Monsoon2d, and Nvision.

Screenshot:



Work-process for the trains that you may find interesting:


Roughed out a 3d model of the desired train in Wings.


Switched it to 100% ambience and painted the polys with solid color.  The multi-colored edge highlighting helps with antialiasing later on.  Also added an inverted-expanded model to act as an outline in the render.


Rough render at game-scale of model


Final Render and alternate styles.

[Jakten]

Those look fantastic Indigo. I'm not one to play facebook games but I'll definitely check it out just to see the work you guys did because it looks absolutely wonderful! Can you give a link to the fan page or anything? I wasn't able to find it.