Did you read part one of our interview with Grand Theft Auto art director Aaron Garbut? I thought so - I've always liked you. In which case you already know how a GTA game begins development, how characters and story are integrated, and why Rockstar prefers to be inspired by a real-world location rather than recreating one.
In part two, Garbut explains how GTA IV's development echoed that of GTA III in certain ways, and how development of a game as vast as this isn't as tightly controlled and rigorously planned as you might imagine. Enjoy, spoiler-free, and look out for our review of GTA IV as we close in on launch. The full game's out on 29th April.
Eurogamer: I've read before that you weren't seeking realism visually, but you must have discovered that the extra graphical fidelity afforded you the ability to give characters and locations more depth. Can you talk at all about how you filled that technological headroom?
Aaron Garbut: There were some similarities between GTA III and IV in the initial approach. It's always a lot harder to get your head around the possibilities of a completely new system. With Vice City and San Andreas we had a pretty good idea of what the PS2 was capable of. On each the engine was enhanced or we came up with a new way of doing something and that bought us a little more power, but generally from day one we knew what to expect and we had an engine to try things out in.
On GTA III and IV though there was a lot of guesswork involved, we had to make things up as we went along. A working engine doesn't appear till a fair ways down the line and even when it does arrive there's still plenty of guesswork since it will not be optimised till nearer the end. So we just make one guess after another and hope we're guessing right. It's always a compromise between memory, detail, lighting, AI, physics, streaming, numbers of characters and vehicles, missions, etc. There are no hard and fast rules since we can bias what's important depending on the area - one area may be lighting heavy, another physics, another memory and rendering intensive. To work successfully with these sort of unknowns, and this number of variables we try to work iteratively as much as we can.
It's only towards the end once we have an engine that's close to final that we can start to tweak with that in mind. And it's at those stages where we find ourselves getting a real sense of what is possible. Where we have to take things out and make compromises and where we end up adding detail. I think we have already pushed both consoles very hard but I'm really excited to see where we can take it next now we know what works and what doesn't, now we know how to play to the strengths of our new engine and the consoles themselves.
Eurogamer: It seemed to me from playing the game that individual neighbourhoods and streets didn't just have their own character, but that you were using them to direct an emotional reaction. Is that scale of mood something that consciously informs your location design, or is it a natural characteristic of New York that you're simply able to harness to fit the existing parameters of mission scenarios?
Aaron Garbut: The cities are never built specifically with missions in mind. We always build the cities first and fit the missions and stories into them. There are a few reasons for that. One of the main ones is practical and it's more pronounced on a new engine. The basic rendering parts of an engine tend to come online a lot sooner. The mission designers need a scripting language, fairly evolved physics and vehicle handling, the weapon systems, AI etc before there is much they can play with. Whereas the artists have 3D software from day one and the game can start rendering that quickly so we can get on with building the city right from the start.
So we've always treated the cities like a real place. We build them, we pack them with interesting things and then we place the missions within them at a later date. Obviously once a mission is placed and working we will tweak the area to work better, but essentially the processes are fairly separate. That's not to say there isn't a deliberate intention to evoke emotional reaction as you say. It's just that if there is one it's happening during the placement and pacing of the missions. I think having this massive environment available first gives a lot of opportunity to play with the missions and find what works best.
There are essentially two routes you can go down in making a game: you can do a load of pre-production upfront and plan it all out in advance or you can just dive in and be a bit more organic. The first option is the safest, it lets everyone know where they are from day one, it lets everyone know what needs done and it's the easiest to organise. But I think it tends to lead to fairly lifeless, soulless games, particularly when the games are more open like ours. We are a lot more organic, this is a conscious choice and it does lead to more difficulties along the way, it's harder on the team and it's trickier to keep track of but it leads to better games. It works because the core team know each other well and have worked together for a long time, we trust each other and know what to expect. Our whole ethos is to try things out, play with them, find what works best and move in that direction.
The entire game in some respects starts blurry and just slowly comes in to focus over the project. Some areas work better than others and the worst areas are always looked at and pushed forward. Missions start as experiments and are moved around the map until they work. The story drives some of this and weaves its way around the rest.
No one aspect of the game is the driving factor, we don't create a list of missions, build levels around it and stick a story on top, and we don't create a story and hang everything off of it. Instead we have a bunch of ideas, elements of the story, the characters, locations, the general tone, gameplay elements, technology, mission ideas, and we just mix it all up and see where it goes trying to steer it along the way. It's all a big scary tangled web. But it works.