Wednesday, May 8, 2013

Game Programming and 2d Art

If you're a game developer, specifically working on 2d games, you will probably want some sort of graphics sooner or later.

Suffice to say, graphics aren't an easy topic for a programmer to really "get" aside from binary values for R G B A.  Yet, for art people, it's all about hues, lighting, perspective, and all that other non-programmer-technical stuff...

Why am I rambling about all of this?  Because if you want to incorporate 2d graphics, you will likely need to interact with artists (unless you're the artist, but you should still take heed!), and likewise, they will have to interact with you... and this is where things can get interesting.

Through some miracle, programmers and artists are working together, but how exactly does all this neat 2d art actually work and get in to a game?  Texture atlases.

Well, let's rewind a bit, the most basic thing is a single graphic.

Now, how you take a single graphic and display it in a game is dependent on your environment, game engine, and all that other stuff.  For the sake of this article, I will not be delving in to any technical examples of how to use any graphics libraries (you won't be seeing any code!) -- I'm more going to focus on *methods* of putting art in to a game.

Now, some programmers might think "methods" and think of programming -- well, reminder, not everyone thinks the same words mean the same things... "methods" can also mean the process of how you accomplish some task... like an artist might use a certain method to create your art, but that has nothing to do with your actual programmatic method used to display that art in your game... or your method of how to implement your method to display the art...

Anyways, back to graphics!

There are some technical issues that you and/or your artist should know right at the get go.  First, some terminology.

Raster Graphics - This refers to graphics that are pixel data.  Stuff like png files, raw RGBA binary data, bitmaps, things that are defined by pixels.  All of this stuff is collectively called "raster graphics".  This is as opposed to...

Vector Graphics - These graphics are a bit more complicated and usually not useful to implement directly in to games.  However, there are some pretty huge perks to having an artist use this format as opposed to raster graphics.  With vector graphics, the artwork is now a collection of mathematical shapes assigned various colors.  This means that, like all math, the original will always remain true, regardless of "zoom level" or how much you scale it; unlike raster graphics which use filters to "approximate" when scaling.

The issue of scaling graphics is complex and is worthwhile to research, but the skinny version is that raster graphics lose quality every time you scale up or down from the original until you will just end up with a blob.  With vector graphics, you can essentially scale up, down, shear, or do anything else to it without losing quality, as much as you want to.

Now, I mentioned vector isn't directly useful to games.  This is true, rendering vector graphics is a massively difficulty task for even today's hardware.  Sure, you could render a couple dozen vector graphics, but you need to realize that the graphic being displayed is generated by math, and if that needs to be refreshed (like during a screen redraw!), it could potentially need to be regenerated; so if the shape is complex enough, it will ruin your game's performance.

Ok, now that we got all that out there, it is really only practical to use raster graphics in games!  Graphics programming is all about using pixel data to feed directly in to some framebuffer, or use as texels for polygons, or things like this.  Someone will need to convert vector in to raster before it can be directly useful in a game.  Thankfully, most (all?) vector programs have an export feature and will allow someone to export vector art (which is flawless!) in to a rasterized version of that art.  What's even better is that you can essentially export to some given scale right away!

I was once confronted with the idea that an artist could "just give me a really large original" from photoshop and then I could just scale it down based on my need at the time.  While this may sound "ok", this is not ok.

Let's talk about graphics scaling some more to hopefully highlight why this isn't acceptable.

We're living in a day and age of thousands of devices, smartphones, tablets, computer monitors, etc.  All of them have different display sizes and capacities (what a nightmare!).  If you want to deploy your game to some/any/all of them, you will need different sizes for all of your art... meaning a LOT of scaling up and down.  But, how do you know what size art you need, or rather, what size would look the best on some given device?

The lazy answer is "just use a big version and scale down", but this actually isn't a good answer at all for 2 reasons.  One, you use excessive amounts of memory/cpu/gpu when you scale it in your game engine.  Two, the scaled version will look much worse regardless if you did the scaling in-engine or pre-rendered the scaled version (ie, you scaled it down ahead of time in gimp/photoshop).

The correct answer is you have to have an exact pixel match for your art to any given display size (read: not feasible).

Anyways, if those 2 lazy reasons don't scare you away, then consider this: scaling graphics up or down too much will always look bad regardless of the originals' size.  I believe the magic cutoff is about 50% up or down before it really starts looking terrible.

Example time.  Let's say I have a 5000x5000 original, which should be considered excessive.  Yet, it does fit the artists' idea of a very large original, and if we can just scale it down to fit our needs, what's the problem as long as we scale it down before we actually use it?

Think like an artist here.  a 5000x5000 original will look absolutely amazing, and the art going in a game should look amazing, so what's the problem?  Well, aside from the obvious memory issue (5000x5000 = 25 million pixels, at 4 bytes per pixel... 100 million bytes to even open that graphic in game!), the appearance will be very terrible if you have to scale it down to something like 128x128.  Go ahead and try it in your favorite raster art program, make a doodle at 5000x5000 and scale it to 128x128 and look what happens, especially around the edges!

Now, this issue is actually more problematic, not only will this scaled version look horrendous, it will actually likely get scaled again when you go to display it in game!  A scaled version of an already bad scaled version will look catastrophic at best.

We could take some less absurd examples, but hopefully you get the idea that a scale of a scale is bad... with larger scaling happening anywhere meaning exponentially bad results.

Vector art sort of fixes this (it's not a panacea though).  The original export will get you a more or less "perfect" raster version at your specified dimensions during the export process.  There isn't any scaling of a raster, there is just a mathematically generated piece of art.  This doesn't work too well with very small graphics (like exporting to say under 32x32ish), so if you're dealing with very small graphics, it is probably best to start doing pixel art.

This generated/exported art will still probably get scaled when it comes time to use in your game and this is the reason why you will likely want various versions of each art asset for different display densities (which is a whole other topic).  Long story short though, you will want to export new originals from the vector art at different pixel sizes rather than scaling your already generated raster art.  The result will likely be much sharper looking art in the end product!  I said likely because there are other steps that can mess up your art all along the way.

Next article I'll cover more about finally putting some of these graphics to work and texture atlases.

Check out part 2

No comments:

Post a Comment