So I just found this interesting talk on procedural generation from the GDC 2017. The presenter is pretty good, but she's no Josh.

Josh's 4th employee maybe?



also, now i know why spore visuals sucked... they went to disney

I watched it, unfortunately she rushed though that vast topic (not enough time I suppose).

What would be interesting is to show procedural content generation as part of an artists workflow to speed up certain tasks. (example is terrain generation and autoplacement in leveldesign) to create more content with less work,
without totally relying on everything beeing procedurally generated.
A mix of manual and generated "layers" in the workflow.

They did that partly by making a white-list for well generated planets in SPORE. Where a human in the end picks the best results of the auto-generation to yield better results. (culling)

The problem of procedural content is that even though every pattern is uquique, it might appear repetitive to a human, as the patterns will sill have too many similarities when in a mass. Thats where a human artist still has the upper hand.

Damocles wrote: ↑

Fri Jul 07, 2017 4:06 am

What would be interesting is to show procedural content generation as part of an artists workflow to speed up certain tasks. (example is terrain generation and autoplacement in leveldesign) to create more content with less work, without totally relying on everything beeing procedurally generated. A mix of manual and generated "layers" in the workflow.

They did that partly by making a white-list for well generated planets in SPORE. Where a human in the end picks the best results of the auto-generation to yield better results.

Too bad they didn't take the next step and turn the human "pick the best examples" process into an expert system or neural network.

I'm waiting for the day when AI engines are a selling point. Game companies don't reinvent graphics engines or physics engines with each new game or even series of games. They often license them from companies that specialize in those things. AI is becoming so complex and specialized that it makes sense that companies would specialize in it and license it to others.

Arclite wrote: ↑

Tue Jul 11, 2017 11:58 pm

I'm waiting for the day when AI engines are a selling point. Game companies don't reinvent graphics engines or physics engines with each new game or even series of games. They often license them from companies that specialize in those things. AI is becoming so complex and specialized that it makes sense that companies would specialize in it and license it to others.

Thing is that AI is so dependent upon the specific game, and it gets harder and harder to do so as games get more complex as gaming grows and evolves.
So I doubt that will ever actually be the case. :/

Damocles wrote: ↑

Fri Jul 07, 2017 4:06 am

What would be interesting is to show procedural content generation as part of an artists workflow to speed up certain tasks. (example is terrain generation and autoplacement in leveldesign) to create more content with less work,
without totally relying on everything beeing procedurally generated.
A mix of manual and generated "layers" in the workflow.

There was another talk at GDC this year where one of the devs from We Happy Few talked about the procedural aspects of their pipeline. Basically, they make the road for the level and a set of 'template buildings'. The templates would then be scaled and warped to auto-build the city around the road. I designed (but never implemented) a scarily similar system on my previous project, so it was awesome to see it working so well in practice.

EDIT: Ah, sweet, it's streamable: http://www.gdcvault.com/play/1024239/Yo ... edural-Art

AdamByrd wrote: ↑

Wed Jul 12, 2017 9:17 am

Damocles wrote: ↑

Fri Jul 07, 2017 4:06 am

What would be interesting is to show procedural content generation as part of an artists workflow to speed up certain tasks. (example is terrain generation and autoplacement in leveldesign) to create more content with less work,
without totally relying on everything beeing procedurally generated.
A mix of manual and generated "layers" in the workflow.

There was another talk at GDC this year where one of the devs from We Happy Few talked about the procedural aspects of their pipeline. Basically, they may the road for the level and a set of 'template buildings'. The templates would then be scaled and warped to auto-build the city around the road. I designed (but never implemented) a scarily similar system on my previous project, so it was awesome to see it working so well in practice.

EDIT: Ah, sweet, it's streamable: http://www.gdcvault.com/play/1024239/Yo ... edural-Art

Deja vu.

Yes, i thought that template aspect of semi-flexible buildings was amazing!

Oh, Adam, that reminds me Have you seen Miguels work over at Voxelfarm ?
http://procworld.blogspot.com/search/label/Grammar <-- procedural castle (with interior), terrain, and possibly textures.

Yep. There's some good stuff in there.

Regarding city generation: there is a vast community who work on landscape and city generation for flight-sims such as FSX or X-Plane.

here one random article taking about the auto-gen of cities in X-Plane:
http://developer.x-plane.com/2012/03/wh ... or-cities/

Flight sims are of the most successful areas where manual editing of terrain features and procedural generation are mixed. (basically populating the whole area of the earth in variable fidelity)
,for use in a realtime 3d engine.

Important here is: the users dont try to let the computer generate "everything", but use a layered approach of combining manual definition of locations and shapes, with manually build structure-templates,
that then get automatically placed and modified by a generator. The quality of the scene gets better and more realistic the more template assets are provided to the system, and the better the algorithms of the generator work.


Another lesson from MS Flight simulator is, that even if the product is quite old, its retaining a large community by being able to be modded and run addons.
Its more of a rich framework that can simulate a world, than an encapsulated "game".

Circling back to Limit theory: I would rather generate ships and stations out of manually editied "prefab modules" than geometric primitives.
Eg: higher fidelity models of elements like trusters, hull segments, wings, cockpits, docking bays etc.. The modules are then connected using some standard seams or connectors, and get adjusted by material and color-scheme, and have some "stretchpoits" to adjust their scale cleanly.
These modules then can also be provided by modders to change and enhance the style.

The results would look better than just connecting metal shaded boxes and planes...

I'd rather not have manually created modules (at least not for ships), it's what everyone does and it feels kinda dirty to me.

The less you give an audience the more their imagination can run free, and I find the imagination is usually better than artists I find, even really good ones. But that's just a thought.

Basmannen wrote: ↑

Sat Jul 15, 2017 2:41 am

I'd rather not have manually created modules (at least not for ships), it's what everyone does and it feels kinda dirty to me.

The less you give an audience the more their imagination can run free, and I find the imagination is usually better than artists I find, even really good ones. But that's just a thought.

That's really true. For example, look at the ships people are making in Skywanderers. They are truly amazing in both detail and scale. Most of it is because of a hinge mechanic piece which allows objects to be offset at any desired angle. Without the hinge piece Skywanderers becomes very similar to Minecraft in both appearance and function.

Damocles wrote: ↑

Thu Jul 13, 2017 3:30 am

I would rather generate ships and stations out of manually editied "prefab modules" than geometric primitives.
Eg: higher fidelity models of elements like trusters, hull segments, wings, cockpits, docking bays etc.. The modules are then connected using some standard seams or connectors, and get adjusted by material and color-scheme, and have some "stretchpoits" to adjust their scale cleanly.
These modules then can also be provided by modders to change and enhance the style.

The results would look better than just connecting metal shaded boxes and planes...

A thought: Before we go down this path, can we see what Josh and the team can do with procedural generation of materials, textures, space-filling forms and connectors?

If the best they can do looks shabby, then generating nice-looking prefab modules is a reasonable alternative.

But just speaking for myself, I'd like to give them the opportunity to try out full-procedural generation of ships and stations. That was promised in the Kickstarter, and could result in a game world that has a distinctive look unlike any other game -- a useful selling point, in addition to being visually enjoyable.

Ok, lets see how the generated models turn out to be.
But I think there is an intrinsic difference between generating natural formations (eg clouds, asteroids, topography) and man-made objects.

Where natural objects adhere to relatively simple rules, that just get added up manifold to form those (fractal) structures, its relatively easy to make it look believable.

On the other hand: man-made tech is designed to have a very specific purpose. Its much less random (much higher entropy) than natural objects.
Just randomly applying a few simply rules, will result in an unrealistic outcome here.
To generate believable man-made structures, there must be plausible rules of what each modules purpose is, how that technical purpose is implemented, and what the known production processes and applied materials are.