3D Printed Zerg Zergling

[Zhall]

Hey everyone,

I've been working on a new Youtube series called 3D print lab, and one of the first episodes is printing the Zergling from Starcraft 2!

Check it out

 

[Heinvers]

hmmm, maybe try Kerrigan sometime

 

[Dr Super Good]

There must be some way to enhance surface detail using the normal map and such layers. Since they represent how light reflects off the surface they must be able to be used to generate extra geometry detail somehow.

 

[Zhall]

Well, I am only printing at half the max resolution, so that might help abit.

I can't imagine maps making extra geometry, but I suppose it could be possible?

 

[Dr Super Good]

I can't imagine maps making extra geometry, but I suppose it could be possible?

Due to resource constraints it is not possible to have a lot of complicated geometry to represent actors in real time graphic situations such a StarCraft II. It is also not suitable to simply treat the geometry as flat (what you are doing) since the results are ugly and unrealistic due to the simplicity of the models. A compromise is done where by the geometry is treated as flat however a special texture (normal map) is used to produce the illusion of fine surface details during the application of lighting (after all lighting determines what we perceive as flat or curved). Although the surface is still physically flat, the result after rendering does not appear to be as the interaction with lighting fakes fine surface details.

A good example would be a rough stone wall. In reality it is not at all flat with bumpy weathered stones and grout in-between. However in a game they can represent it as a flat surface with a normal map. From a distance away it will look realistic as if it was not flat however physically in the game it is perfectly flat. Another good example are tiles which are surrounded by a sunken border of sorts. In real life the surface is not flat however in a game it is flat with a normal map faking the illusion of it not being flat.

Here is some internet explanation for it...

One of the most valuable maps for a 3D artist is the normal map. Rather than having a color range of black to white, like a bump map uses, normal maps consist of red, green, and blue. These RGB values translates to x, y, and z coordinates, allowing a 2D image to represent depth. This way, a 3D application is able to fake lighting details based on the color associated with the 3D coordinate.

Normals are often generated in a process of model optimization whereby a high-poly source model is used to generate a low-poly final model with baked normal map retaining some of the high-poly detail. In theory it should be possible to go the opposite direction and take a low-poly source map and normals to produce a high-poly model (possibly using tessellation? I am not sure how one would do this). In real-time graphics this is a pointless thing to do as you can use the normal map in the first place. However in the case of 3D printing where you cannot use the normal map and model complexity is not a concern it would be very useful to do as it would add high-poly details onto the low-poly model and so improve the resulting product quality.

The main difference would be that a lot of the angular surfaces that are visible currently could become less pronounced. They are angular in-game however the normal map gives the illusion of them being more curved. In real-life such illusion does not exist so you need to physically make the model more curved.

I believe this applies to games like Warcraft III even except the technology used was different. Warcraft III and older games used vertex normals which I think operated similar to the normal map except without much fine detail. Again it should be possible to use these normals to add detail to model geometry via some processing so that surfaces appear more smooth than flat (like they do in the game).

 

[Zhall]

Right, bump maps and such effect the render.

I don't think there is a way to export the mesh after this effect...

 

[Dr Super Good]

A bump map can be converted into a depth difference map across a surface. Integrating a depth difference map gives you a displacement map. The displacement map can then be applied to the model once its mesh is sub-divided (eg by tessellation, something that can happen in real-time).

The problem is the normal map will produce only an approximation of depth difference across the surface so trying to integrate it at the Texel level results in considerable error. Processes do exist to perform such integrations and recover a depth map from a normal map. The results are not perfect (as can be seen in that presentation), however in the case of 3D printing I would imagine them to be invaluable. Even slightly incorrect surface detail is better than no surface detail at all.

This entire process can be viewed as surface detail recovery. Games do not need it as the normal maps can be used directly to give the appearance of surface detail. In real life physics does not allow such a shortcut. Currently you are discarding all surface detail which is why the product looks poor and different from what it does in-game. Recovering some of the surface detail (even if inaccurate) might greatly improve the quality/realism of the product as it will more closely match what people see in-game.

 

[Zhall]

Well, that's past my understanding of model extraction and art tools for Sc2.

Perhaps I'll come back at some point and re-print it.

I've also been told that there might be cinematic quality models available, and that might be a good option.