The pros, cons, and techniques of procedural asset creation (renamed)

[Scott_Arm]

I think Shift and Laa-Yosh are arguing the same point, but from the perspective of a programmer/engineer vs an artist, respectively.

I can see why Laa-Yosh will always think that procedural content would be sub-par compared to completely hand tailored content, and would prefer tools that improve efficiency while adding flexibility. Some examples that come to mind would be Carmack's Megatexture tools or Lionhead's Megageometry tools. From his point a hand-crafted room, or a hand-crafted character will always be far superior to a procedural version of the same thing, so a game like COD with a more directed (art-wise) experience will always make the open-world game with more procedural content look shabby by comparison. With what is currently available, people seem to prefer the former to the latter, based on popular opinion of which titles "look the best."

I can see both sides of the argument. From the opinions expressed by those in the industry that have worked for companies that investigated this topic, it does seem like procedural generation of content is a long way off because it is too expensive/complicated.

Edit: Although my recollection of megageometry is vague, and could some aspects of the tools be considered procedural content creation?

 

[scooby_dooby]

I tend to really agree with LaaYosh on this one.

Bigger is not always better, currently we seem to have this fascination with this concept of huge massive worlds, but is that really better? Borrowing LaaYosh's argument, what use is a house with no-one in it? Or the same repetitive dumb AI characters with nothing to say and no-one to interact with?

They're good for nothing, they may as well not even be there, as they provide absolutely no enjoyability from a gameplay perspective. The only purpose they serve, is to not break the illusion of the game world... but that only lasts about 10 seconds, until something else breaks that illusion. Obviously you need filler, but it shouldn't just be there for the sake of being there, it should just fill in the blanks with enough "real" content sprinkled in to create a believable world.

There are just so many other considerations other than artwork that just make this sort of thing impossible to do properly. Audio recording, script writing, TESTING! To me this is a holy grail I don't really want. Once you get into that sort of scope, everything becomes ridiculous, everything becomes a massive amount of work in it's own right...next thing you know making a game is a 5 yr process...

The other point, is that creating good procedural content sounds like alot of upfront work to do well, and those are man-hours than an artist could just be spending actually doing the work properly, and get better results in the end. So in order to be worth while, you need a massive scope, but with a massive scope comes a mass of other issues...

So, is it not better to just keep the scope of games in check, and develop smaller focused experiences that can be delivered in a 24mth development cycle?

There's a reason that Oblivion just fell apart after a while, I spent > 100 hrs on that game, I was completely enraptured by it, and it was a pretty amazing experience... but one day it just fell apart, the realization came that this entire world is actually lifeless, and pointless, everything was the same... and I literally never picked up the controller again. It was a really strange experience, I've never felt something like that before with a game, to literally just lose complete interest over night.

Thinking back on it, Fallout was better done in this regard, as it was more focused and had much more unique set pieces. Seems to be that is a strong testament to the value of smaller, more focused, more unique worlds.

 

[ERP]

While I agree with the general premise that bigger is not better wen it comes to games. In fact I've worked on a few that I think would have been better if the time had been focussed more on fewer elements.
I think as games become more social we are going to start to see games that need so much content that the direction is unavoidable.

If I remember correctly the original godfather had 50 unique building layouts with some number of variations on those (though it might have been less in the end), the art staff on that game was enormous.
I think the designers would have preferred a lot more.
On top of that streaming was extremely challenging, if you can store the description of the entire city in a fraction of your RAM, you can increase view distances, and increase visual quality in these types of games.

Yes there are test issues etc etc, but there are plenty of games where getting sufficient coverage is difficult, it's shouldn't be and rarely is an issue that affects design.

I think it's more a question of when and how than if when it comes to procedural content in games. Doesn't mean every game will do it, but if I were trying to build for example an MMORPG today I'd be looking and how far I could practically move in that direction. Again doesn't mean I wouldn't author content like quest and store lines.

The original PC version of Dune 2 circa 1992? used randomly generated maps, the designer would just keep pushing a button until he saw a map he liked, then apply changes to it to tune the gameplay.

 

[Shifty Geezer]

Let's scale this whole argument back to one single example and one simple solution so that people don't get lost in the huge scope of the full concept.

In my above post, I linked to a video of Dungeon Siege 3 - a moderate budget game from a big publisher. In it I highlight a point in town where there are steps, and these steps share the same repeating stone texture such that the result is jarring.

What are the solutions to this? My suggestion is to provide a larger, more generalised stone texture (or set of textures working in combination), and to displace it per step. This could be done offline by an artist changing UV coords for each and every step, or randomised offline, or randomised in game. I imagine with a suitable stone texture, it could be reused in architecture too, along with a set, resulting in the possibility of a lot of the repeated textures in this game being much better discussed.

What are the issues with this approach? What are the alternatives that can eliminate these repeating stones, and how do they compare?

 

[tabs]

Does a Minecraft world count as being procedurally generated? If so then it shows some games just wouldn't be possible without it.

 

[ERP]

Let's scale this whole argument back to one single example and one simple solution so that people don't get lost in the huge scope of the full concept.

In my above post, I linked to a video of Dungeon Siege 3 - a moderate budget game from a big publisher. In it I highlight a point in town where there are steps, and these steps share the same repeating stone texture such that the result is jarring.

What are the solutions to this? My suggestion is to provide a larger, more generalised stone texture (or set of textures working in combination), and to displace it per step. This could be done offline by an artist changing UV coords for each and every step, or randomised offline, or randomised in game. I imagine with a suitable stone texture, it could be reused in architecture too, along with a set, resulting in the possibility of a lot of the repeated textures in this game being much better discussed.

What are the issues with this approach? What are the alternatives that can eliminate these repeating stones, and how do they compare?

Procedural textures for things like brick/stone and wood have been around for over a decade, the primary reason they aren't used in games is shader cycles.
Certainly they could be used off line to improver things like the example.

Having said that i doubt that particular issue is a function of limited space or art resources, it's more likely a technical limitation of the engine, i.e. it's all tile based, probably with some preset tile limit and the artists do the beat they can within the budget.

 

[Shifty Geezer]

Procedural textures for things like brick/stone and wood have been around for over a decade...

I wasn't suggesting using procedurally created textures. They can be hand-drawn artists textures.

Having said that i doubt that particular issue is a function of limited space or art resources, it's more likely a technical limitation of the engine, i.e. it's all tile based, probably with some preset tile limit and the artists do the beat they can within the budget.

Looking at the game, I don't think this is so. The steps are pretty evidently a long cuboid polygon object with a texture on and the same object and texture is used repeatedly, and there's nothing about the game to suggest the designers wouldn't use this system. I imagine the choice for the steps is either due to a limit of RAM or artist budget, hence reusing the same asset over and over in the same immediate location. One single large stone texture reused would save RAM over lots of individually textured steps. It would use more than one smaller stone texture reused, but if that one stone texture can be reused elsewhere too, you get net savings and more variety.

 

[Arwin]

Does a Minecraft world count as being procedurally generated? If so then it shows some games just wouldn't be possible without it.

I don't know much about the game, but I presume yes though of course the most interesting bits were user generated. There was also a game called Love, which was created by just one guy and for which I think the entire game world was procedurally generated.

 

[MfA]

Wot's DA2 ?

Dragon Age 2 ... the game with 1 city and 1 copy/pasted dungeon.

 

[patsu]

I think it may be fruitful if we start with a list of of procedural contents that are already out there. I get the Borderland's weapon generation, SpeedTree (?). What else ?

 

[tabs]

Diablo dungeons
Minecraft worlds
Terraria worlds


I'm not sure how far you can go with it really, what about pedestrians in GTA games. They're not just random - there are rules governing the likelyhood of particular types spawning in particular areas. Is that classed as procedural content or not? I'd like to see a proper definition, if that's even possible.

 

[Ethatron]

Is that classed as procedural content or not? I'd like to see a proper definition, if that's even possible.

What means "procedual" -> "procedure"? There is your definition.

 

[Ethatron]

Project IGI 1&2's terrain is procedual created.

 

[hoho]

Many RTS/TBS games have random maps that can be qualified as procedural. E.g age of empires series and anno series have quite decent maps. UFO: aftershock/math/light also had kind of procedural maps but they just combined them from smaller chunks and things got repetitive relatively fast as the selection for those items was quite small.

 

[tabs]

What means "procedual" -> "procedure"? There is your definition.

Yes but everything is created with a procedure, that is the nature of computer software. My question was really what constitutes procedural content. Can it be something as simple as rules governing where entities spawn, or should it be something more than that?

I would guess most people take it to mean jobs that are traditionally done by hand, to be produced by some sort of algorithm, but my point was where do you, or is it even possible to draw the line?

The reason I want to know is when asked to make a list of games with procedural content, it becomes incredibly hard to say if I'm not sure what the criteria is.

 

[Ethatron]

Well, I think whenever a repeatable "procedure" has been created and runs on the end-user machine it classifies as procedual - that is basically an ideal expressed in code; the code will always produce results within the constrained idea. Autonomous AI in a broad sense is procedual behaviour, the opposite case is scripted/stored/replayed behaviour. That is storage vs. algorithm. You can paint a texture and store it, or you can describe the rules for the creation of a texture and store the rules. While from my PoV it'd be very weak procedual if the rules could just create a single variation of the texture, it still fits the definition (there is a blurred line between procedual and parametric). In general it's unquestionable that if the rules allow unconstrained recreation of, not quite but practically, infinite outcomes, it's called procedual. Programmers often work with "seeds", which is a little bit like a gene. IGI's maps had specific seeds stored for the maps, those were selected by the designers. The maps themselfs were recreated on the end-user machine.

 

[Prophecy2k]

I think it's funny how the state of the disscussion in here can be effectively sybolised by the thinking of bioware when going between Mass Effect 1 to Mass Effect 2; in particular regarding the procedurally generated planets that you could visit.

ME1:
- Bioware uses procedurally generated assets to create planets and terrain for the player to explore.
- Many complain that the procedural system for generating this content was insufficient and could not hold up to artistic scrutiny like the other more artistically directed sections of the game.

ME2:
- Bioware knows it can't have its artists sit and manually create every of the explorable planets in the game's galaxy because of budgetary contraints.
- Bioware decides procedural content generation is insufficient and abandons both it and the prospect of having explorable planets in ME2 altogether.

The result:
- Many gamers rage because though the procedurally generated stuff was quite clearly of a low quality than the none procedurely generated stuff, having more game content over less was always the better option.

I don't think that procedurally generated content in games is a bad idea. If anything i think that it could do wonders for being able to expand the physical world of games in a way that offers more to the gamer than what games currently do. Of course artists are needed and their roles are invaluable to creating what i like to call the core "directed gameplay experience". However procedural content generation can do wonders for expanding the content of games beyond the core experience, albeit at a lower quality, but that can be irrelevant if the tools are good enough to create content that is "good enough" for it's purpose.

Mass Effect is my case-in-point, because i don't know anyone who prefers planet scanning to even just bioware re-using the exact same already generated planets from ME1. If they had a better planet generating algorithym, which of course might require more artist created content to have a better result, it would invariably have been better in ME2 than none at all. Plus the various different extra content and asset types they would have had to make, on top of the existing ones they already had from ME1, for the precedural planet generation system would have probably cost them more or less the same in artist man-hours as it did to create all the individual 3d model planets and the planet scanning system anyway. At worst they could have just re-used the same ready-generated ME1 planets and had their artists tweak them a bit more.

I'd also throw in a suggestion to all those saying that setting up a solid procedural content generation tool is too complex and costly for game studios, that there's no reason why a game technology middleware provider like Crytek and Epic can't do something like this. Especially Epic given how popular their engine has been this gen. Something like this would be a huge gain for them and their ability to sell their engine tech, given that the whole point of engine middleware is to save dev costs (which goes hand-in-hand with the point of procedural content generation). There's no reason for Epic not to do it, and to be honest i wouldn't be suprised if that ends up as a big part of what UE4 is all about when it's released for next gen consoles & PC.

 

[Laa-Yosh]

Just because a vocal segment of the customers criticized ME2's approach it doesn't automatically mean that they are the a majority and they are also right in their criticism.

The only objective facts are that ME2 got better reviews and sold more copies.
I'm sure Bioware also has statistics on how many of the random missions in ME1 were completed and how many of the more hand-made ones in ME2, so they can draw more detailed conclusions and we'll see if procedural missions are back in ME3 or not.

Oh, and throwing in planet scanning as an argument is a low point, as it's completely unrelated...

 

[Laa-Yosh]

Also, there is no such thing as a "procedural content generation tool". Every case is vastly different and needs its own solution, that's one of the bigger problems. You can't even make a universal face generator...

 

[hoho]

Also, there is no such thing as a "procedural content generation tool". Every case is vastly different and needs its own solution, that's one of the bigger problems. You can't even make a universal face generator...

Basically what that means is you can shift some of the artist workload to developers and if you have really good developers you might end up with decent solution.