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

May be more interesting to steer the discussion towards the 'hows' rather than 'whether' (How good, how terrible, how do they work).

For auto-gen vegetation, do the tools follow fibonacci sequence when generating the branches and petals ? Or just fractal math that looks mostly correct ?

For human related modelling, do artists and tools observe the golden ratio for appealing main characters ?

(Laa Yosh should be able to school us better now. :p)
 
May be more interesting to steer the discussion towards the 'hows' rather than 'whether' (How good, how terrible, how do they work).
I have tried! I've asked for ideas from to doubters but they just want to say nothing's going to work. Laa-Yosh only took one attempt to explore what needs to change to improve procedural content. One problem is convincing people it can be done, as once convinced they'll start looking into it. Although typically progress comes against common belief, and it'll be some fringe development that breaks new ground against common wisdom.

For human related modelling, do artists and tools observe the golden ratio for appealing main characters ?
AFAIA the golden ratio has been pretty much debunked, as the actual variance with proportions makes for quite a wide range of ratios fitting within the so-called 'golden ratio'. Beauty is in the eye of the beholder. ;) A good artist can go by eye. A procedural system could take a default human proportion and scale it accordingly, like EA's sports games.

I think a major problem is how to get procedural content to include the performance optimisations an artist can make. An artist's model can have just the right vertices in just the right places, whereas a mesh intended for deforming would have to be more densely populated. i think cars are a good example here. The fundamental design of any hatchback is the same, just shifting the window shapes and positions a little, adding a crease here or there, etc. But a single generic car that can be deformed to a dozen different models and still preserve texture fidelity all over and have the crease in the right place is going to be hard. As is joining models (a subject brought up earlier this gen regards high order surfaces) at seams.
 
Anyway, I'd be happy to see any good procedural tools that really work well, I really hate 12-hour days and 6-day weeks. But there's a level of quality we're expected to reach and it is impossible to do with these tools
Expected by whom exactly? Quality/quantity is a balance ... the industry is not producing the quantity I expect them to reach either (DA2). If quality has to suffer a bit, so be it.
 
. i think cars are a good example here. The fundamental design of any hatchback is the same, just shifting the window shapes and positions a little, adding a crease here or there, etc. But a single generic car that can be deformed to a dozen different models and still preserve texture fidelity all over and have the crease in the right place is going to be hard. As is joining models (a subject brought up earlier this gen regards high order surfaces) at seams.

I think any solution that tries to treat the problem as just Mesh generation is doomed to fail.
You need to understand the semantics of the components that make up a car and their relationships to each other, if you can come up with a compact definition of a car, that's rich enough to build a model from, the rest is more or less a solved problem, given a large enough sampling of what good looks like.

There are a lot of research papers on all of this, I read a paper 10 years ago on a "language" to describe buildings, and using it to generate procedural cities. I wanted to play around with the idea for a PS2 game that was starting development at the time, the idea being to pass the description to the VU and have it build the model, but the game was canned before we looked at any real technology.

The biggest issue when you look at all these approaches is they are too specific, the cost becomes engineering or training models to produce what you want instead of having artists generate them. It's unclear to me today that for most games, it's a cost saving and you pay the penalty of having no artwork until your done with the engineering/training.

I don't think it's insoluble, I just don't think the obvious financial justification is there to justify the investment.

Someone will do it, because they have to and everyone else will copy.
 
I don't think it's insoluble, I just don't think the obvious financial justification is there to justify the investment.

Someone will do it, because they have to and everyone else will copy.
Should we be looking to a third party engine, like Natural Motion for animation and Havok/PhysX for physics? I'm not sure how a system could be generalised to work with loads of different art types, but I'm sure some cleverclogs can work it out - there's always some cleverclogs who reduces everything into a simple equation! ;) I think datatype will be key. Perhaps the concept of a vertex
needs to be reconsidered??
 
Should we be looking to a third party engine, like Natural Motion for animation and Havok/PhysX for physics? I'm not sure how a system could be generalised to work with loads of different art types, but I'm sure some cleverclogs can work it out - there's always some cleverclogs who reduces everything into a simple equation! ;) I think datatype will be key. Perhaps the concept of a vertex
needs to be reconsidered??

I think the first movement you'll see on this is likely to come from a developer working on a product that needs it rather than core tech companies though I could see some 3rd party coming in and trying to sell a solution.

The reference earlier to need an AI is exactly right (albeit a very specialized one).

I think this problem is completely independent of the rendering format. Until you can describe an object and how it's parts relate to each other, along with what is good within that ruleset, your pissing in the wind.

This is fundamentally knowledge representation.

I can draw a picture of a car because I know what constitutes a car, once you get that far you have to be able to generate good looking cars. That's in practice is "just" a classification problem, your just carving out a volume in some very high dimensional space. Once you know what that volume is you can just randomly pick points in it.

I should write a paper and file a patent for semantic models ;)
 
If designing cars would be that simple, there wouldn't be star designers in the automotive industry...

I also don't get why it's necessary to push for quantity instead of quality. Do we really need thousands of different rooms and hundreds of different cars and even more variation of people?
And most open world games suck on the simulation front anyway, as I've indicated elsewhere we haven't really seen anything beyond what Ultima 7 had nearly two decades ago. Maybe the new Deus Ex game will have some more complex world simulation, but evne then the underlying systems aren't usually complex enough; for example you still need to write lines of dialogue and record them. What good is it to have ten thousand citizens if you still can't get into conversations with them? Entering an apartment should have its gameplay related consequences too (confrontation with inhabitants, police investigation, etc) and have various items to pick up, from clothing through hidden cash to whatever. Otherwise it's not worth the effort to model them, right?

Today these things are more or less balanced, what's modeled is usually also at least to some extent a part of the gameplay. Adding interactivity to vast procedural worlds would require a lot of additional resources and allow less to be spent on the core gameplay. Unless the core gameplay is roaming around in a large procedural world, which I personally don't really find that interesting ;)
 
...Unless the core gameplay is roaming around in a large procedural world, which I personally don't really find that interesting ;)

I think the idea that they are presenting is more along the lines of having procedurally generated content which fills the void of the main story. In the current context of most games, this would lead to having other houses which are interact-able that are filled with procedurally generated content instead of a locked door as is usually the case.

Their concept seems to be filling the typical void of non-interactive environments with somewhat interactive, procedurally generated content to give the worlds a bit more of a realistic tone.



IMO, I think the concept of generating new "worlds" and content for every game is a ludicrous waste of time, effort, and money.

More importantly, it's financially unsustainable.

Especially next-gen where expectations are increasing.
 
The reason for nearly no asset sharing between various games, even between the same studios, is that there are large differences in the rendering tech and art direction among the various titles. Even games like Uncharted and Infamous are too far apart to do that.

Still, later on it should get more common, if nothing else then the more and more common use of external contractors for graphical content should make it happen. If it's the same studio building stuff for a dozen games, they may be able to re-use assets to some extent. Of course publishers should allow this to happen, but lower costs are in their interests, too.
 
If designing cars would be that simple, there wouldn't be star designers in the automotive industry...
They're designing cars for people to buy, and not to fill up a virtual city as background noise. The people who designed the cars in Infamous wouldn't be winning any design awards. Until you can see the bigger picture, that a lot of content doesn't need the best art, you'll never understand the part procedural content can play in helping create better experiences, whether providing more content like random rooms, or more varied content like different people, or cheaper content like automated asset building.

And most open world games suck on the simulation front anyway...
The problems you outlined cannot be fixed. They are too vast. That's not reason to give up improving other aspects of the games! As I've said before, we've pushed for better and better lighting solutions despite them remaining very far from realistic. But because we've pushed, accepting lots of rough compromises along the way because they've moved the art forwards a bit at a time, we're actually getting to a point where next-gen, the shortcuts might be pretty convincing, and the appeal of games will be increased. Thank goodness when someone suggested prebaking GI into lightmaps, no artist said, "don't bother. It's unrealistic, dynamic objects won't influence the lighting," and when someone suggested SSAO they weren't knocked back by an artist saying it was unnatural and not worth bothering with. ;) The fact that I cannot go up to a pedestrian and engage them in a natural conversation, or there are no consequences for me looting chests, and so these games are fundamentally unrealistic, doesn't strike me as a reason to give up trying to improve lighting to make games look more realistic, or improve the variety of people to make things look more realistic, or mix up textures a bit so they don't keep repeating themselves to make the world a little more realistic. Every improvement is welcome, even if a true virtual recreation of our world is possibly impossible for any tech for generations to come, if ever.
 
Until you can see the bigger picture, that a lot of content doesn't need the best art, you'll never understand the part procedural content can play in helping create better experiences, whether providing more content like random rooms, or more varied content like different people, or cheaper content like automated asset building.

Procedural content does not provide better experience on its own, period.
Also, please don't school me like that.

I'm a bit fed up with this discussion though, because you keep dismissing every point I make... "blahblahblah I'm right and you're not"
 
Procedural content does not provide better experience on its own, period.
Also, please don't school me like that.

I'm a bit fed up with this discussion though, because you keep dismissing every point I make... "blahblahblah I'm right and you're not"

To be fair Laa-Yosh, and please don't take offense, but that could equally describe your contribution to this topic as well.

I've lost count of the number of games where the virtual city is this closed box, where every door (but a select few) is locked and every window is opaque. I'd welcome something that solved that. Sure nobody's going to go into every apartment and check out all the crap, but it'd add a lot in that particular genre to simply know it's there.

I agree with ERP though, nobody is going to go down this avenue until there's a need; until it becomes a necessity for a particular game.
 
Procedural content does not provide better experience on its own, period.
Never said it did.

Also, please don't school me like that.

I'm a bit fed up with this discussion though, because you keep dismissing every point I make... "blahblahblah I'm right and you're not"
But you've done exactly the same, "blahblahblah I'm right and you're not," and not agreed with anything I've said either! I'm not going to agree with you just to be polite! :p If we disagree, we disagree, and it's quite apparent neither of us is going to get the other to change their perception.
 
okay, you are working to a very different definition and thought process to me, which is why this discussion is going in circles. Procedural/algorithmic means getting a computer to work out how something should be instead of a person manually telling the computer. Uncharted animations are procedural. It's an example of a computer mixing a selection of parts (animation cycles in this case) and blending them. It's exactly the same idea as taking a number of character parts and blending them to make a character.

It is only art if the artist dont control and know the final result in those case the artist know how things will look in the end, it is like making a circle in corel, not art just a tool.

It will be art when the artist say I dont know what will happen or how it will happen but it is part of the look of the game anyway.


Anyway I think we see game in a different light, for me a game is just that, you seem to like that at least a few games to be a open world simulation.

What I like in games is that they are a human crafted experience, not a simulation with boring stuff to fill it up.


As such, procedural content covers a huge area. From assembling premade parts to create a whole, to adjusting a model or texture on the fly, to creating a new model or texture from scratch. From Dust's environment's are a mix of procedural content, as the levels were created by artists and yet can be created and changed in game, rather than being a baked level. The only difference between a level building tool and procedural levels is whether the same tools for raising/lowering/colouring land and adding objects are executed via a human or AI.


I am happy with todays gfx, if dont get me new games in situations that I will actually play I dont care that much.




There's a WORLD of difference between a good artwork and a bit of background. how often do you stop in the street to admire a bit of trash in an alleyway? Or stop in a pine woods and look over each and every tree and appreciate it's beauty? How jarring would it be instead to walk past alleyways seeing they were all exactly the same?! Or noticing every tree in the woods you were in was one of three models? If you saw that in real life you'd freak out because it's unnatural. Life has more variety, which isn't yet modeled in computer games.

But we do in Borderlands or other games, at least let it be finely done. Again a game is not a simulation, that is way it dont need to look like the real life, I even think I prefer it is not like real life, there is a lot of stylized stuff this days, even in movies, and I really like it.


But it is. Instead of the user having to select parts of the photo, duplicate them, extend and blend them following scene contours, a computer performs those processes. The computer does nothing a human operator wouldn't do, only much faster. Of course, without any artistic appreciation for what's happening, so it may not produce results that'd be good enough for final without a human's touching up. But it is procedural, AI content, and another example of how the technology of computer's creating content is improving.


It is not AI, it is good trick done in smart algorithms, probably based in already existing tools.

Great work none the less, really good for artist ;):D.
 
Procedural content would be nice if used well, as long as artists provide decent assets and situations to start with, although it would never replicate the myriad expressions of life and other things, it's a fine alternative.

Some smart thinking might be required after all, something that computers can't do, but we can provide them with "intelligence", this way small or large percentages of a world could be calculated for the purpose of variety.

You could provide people with tasks, dependent on the context and atmosphere of the game, not overly complicated stuff like someone working two jobs to put a roof over their head, but for instance the same person mowing the lawn on weekends, then cleaning the house, cooking, making school lunch for kids...

Someone doing holidays at home, putting up the tree, do all the shopping and cooking, like fried Spam and boiled potatoes, etc. Calling up at night from where this character is working to see if their kids did the homework, or some people doing anything because some tasks bored the crap out of them.

That would be for the AI.

As for the environment of games, random events are always welcome, like when in a racing game a car goes out of the track, or an old man who, while talking to you and in the middle of conversation, spits out dregs of a cigarette, because his cigarette doesn't have the filter tip..., etc etc.

Random map generators like Diablo 2 are useful examples of procedural content which can work fine to some extent:



Also, a game like Viking: Battle for Asgard would greatly benefit from a more varied set of clothes. Massive battles look very repetitive in that game.

In similar situations or games, a character in jeans, another wearing a skirt, a t-shirt, a jacket if it's chilly. a jmerry.., wearing things like that, would help people to probably recognize them in the scene, as individuals, not being the same clone repeating, like garlic repeating on you.

That's just my view on everything, because speaking from (the lack of) experience...it's tricky.
 
The reason for nearly no asset sharing between various games, even between the same studios, is that there are large differences in the rendering tech and art direction among the various titles. Even games like Uncharted and Infamous are too far apart to do that.

That may mean we need to create more art using both human and procedural art in the new games. It doesn't mean procedural content is unwelcomed universally.
 
Procedural content does not provide better experience on its own, period.
Also, please don't school me like that.

I'm a bit fed up with this discussion though, because you keep dismissing every point I make... "blahblahblah I'm right and you're not"

No I agree entirely with a lot of your points.
Artists aren't going anywhere, and neither is hand created art.

Procedural generation of acceptable assets is
a. A hard problem
b. Not financially viable at this point

There are a number of reasons I think procedural content creation will move forwards.
Firstly I see a lot of computing and storage moving to the "cloud" in the long term (my vision of the cloud isn't the dumb terminal one, but that's a different discussion), and I think procedurally generated content makes a lot of sense when it comes to pushing assets over a wire with limited bandwidth.

Secondly it's hard to move backwards in the games industry, if someone produced a game in which every room in a city could be explored, it would become the expected norm for every open world game there after.

My own point of view is that I think it's a really interesting technical problem and if you constrain it sufficiently, it's at least partially soluble with enough time.

I'll add it to my list of good startup ideas incase I ever do that again.
 
No I agree entirely with a lot of your points.
Artists aren't going anywhere, and neither is hand created art.
But the topic has never been about replacing artists! All along the advocates of procedural content in this thread have been viewing it as a technique for applying the artists assets in more versatile ways. I dunno...maybe the whole issue here is coming from the title since the topic was spawned? Maybe I should change it to a more generalised discussion rather than a versus?

I've agreed with Laa-Yosh about the need for artists and their essential contributions, but I'm also seeing opportunities to turn processing power into game content as a natural evolution of game engines.
 
Back
Top