Taming The Dragon: Next-Generation Asset Creation for PS3

The second floor of Sony's Metreon Center in the center of San Francisco was bustling with talk of Katamari Damacy and Donkey Kong as game industry professionals and a few not-so-professionals funneled past a pair of turnstiles into the three-screened Action Theater for the latest IGDA meeting and mixer. Matthias Worch, designer and technical art director for Factor 5, was the speaker of the evening. He was armed with an updated version of his technical lecture from the Game Developers Conference this past March, one that now discusses art creation for Factor 5's still somewhat mysterious PlayStation 3 title Lair.

Fundamentally, this was another opportunity to explain the value of digital maquettes and to demonstrate the rendering software Worch is most fond of. This seemed to go over fairly well, as Worch's tools are powerful (and indeed elicited constant gasps of admiration from the audience) and he has a number of sound arguments for at least considering maquettes as an alternative modeling technique.

This IGDA event all falls under the general topic of adjusting to the technical needs of next-generation technology. One senses some serendipity, in that these needs seem to have come about at just the right time to justify Worch's arguments – possibly not that the methods were set up to address those needs. But, as Worch seems fond of arguing – whatever works!

Resolution Conflict

The general goals, as Worch would have it, are in broadening the options your art department has to call upon, in "opening up the normal mapping workflow" (to be explained below), and – relatedly – in training.

As far as maquettes go, the benefits are notable: physical props are invaluable for communicating art direction; there's a relative availability of traditional modeling talent, compared to purely digital modelers; that the prepping of maquettes can be a valuable training tool for new artists or to help the existing art department in adjusting to the demands of next-generation modeling; and that if done correctly, maquettes are just as cheap and fast to produce as purely digital models.

Worch explained the difference between three models of scanners: the handheld Polhemus FastSCAN, which should be decent for general use; the Cyberware M15 that he spoke so highly of in March, and the ultra high-res XYZ RGB proprietary scanner. As he compared the results of each in sequence, the raise in quality was obvious. The XYZ scanner, according to Worch, picks up detail down to 100 microns in size, which he described as smaller than the point of a needle. Whereas the Cyberware was sufficient for the last generation, Worch explained, the next generation is at the point where the added detail from an XYZ scan will actually come in handy – at least for the initial model.

The Worch of Both Worlds

The benefit of a purely digital model, Worch said, was that, due to the way they're built up from nothing (much the opposite approach from scan prepping), one generally has a low-res version already handy; with scans, you have to build one. The good thing is that this is a simple process, which will serve well as a tutorial for adjusting members of the art department. Worch then demonstrated the downscaling of high-res scan models to game models using Cyberware's CySlice program. As part of the process he captured "normal maps" (basically dithered-down wraparounds of the high-res original) to apply to the low-res model, to preserve as much detail as possible.

For some contrast, he next loaded up Zbrush to demonstrate the manner in which digital models are built up and refined. For an added zing, he showed how some mapping information he acquired from a maquette could be applied to a purely digital model, to blend things together and add in some extra detail.

Watching the Show Dragon

The thing most people probably came for, though, was the dragons. Worch had little to say about the actual game he has been working on – now titled Lair – though he used the models all throughout the lecture. One of the key models was a maquette (built by Peter Konig at Massive Black, the man behind the Dragonheart models); another, a digital render. The final version used in the game, Worch said, was a blend of the best parts from each: a practical body and limbs, and a digital head, tail, and wings.

He played a recent extremely high-resolution trailer in real-time, occasionally pausing to swing the camera around or turn on or off various effects. To be fair, the scene in question was clearly a cut-scene, calculated to show off just how many polygons the PS3 can throw around; it's still a lot of polygons, though.

Each model, Worch claimed, contained somewhere between 100,000 and 170,000 triangles. Each had a bunch of other special maps and lighting applied, and the main character was built up with "over ten textures". He compared this to an estimated 10,000 for characters in Gears of War and other recent high-res games. The high-res models, meanwhile, that got dithered down to produce the in-game models, ran up around 5,000,000 triangles.

Lasting Impression

It was abundantly clear, from this IGDA-hosted presentation, how art departments will need to adapt for the PlayStation 3 and other next-gen consoles, to compete at this level of detail and sheer muscle power. Strictly on the level of brawn, models like this are basically unprecedented outside of feature films. This does not, of course, address the issue of how much detail is actually necessary for (or will even be visible to) the player, or what concepts this type of power will enable that were impossible to realize before. But it's certainly pretty.