3D modelling is a creative act in which it uses computer to make digital copies of everyday things, such as vehicles, machinery parts and so forth, for the purpose of visualization, communication and replication (for 3D printing).  The practice of 3D modelling is an abstracting craft where it allows practitioner to make things without confining to the physical world, as often the process of digital model making can be easily performed with a laptop or tablet that comes with preloaded 3D software.

Theory of Computer Graphics in displaying and rendering a 3D Polygon cube

Rendering Pipeline for 3D cube


All computer-generated (CG) 3D models are actually an array of different polygonal-planes that being coordinately bonded and rendered with object space techniques, such as the front-to-back display algorithm by Gordon & Chen (1991) for better preserving the computational resources by culling away the non-camera (or viewer) facing surfaces.  The pipeline to generate a shaded 3D model can be conceptually illustrated as the animated figure of the above.  3D models are seen to be used in wide variety of ways for different industries, which can spring across from medical visualization to accident reconstruction and visualization, as shown in below.

Applied examples of scientific visualizations with 3D Models

Figure 2 Applied examples of scientific visualizations with 3D Models. Image Credits: 01 Hitachi Medical Systems: SCENARIA http://www.hitachi-medical-systems.nl/products-and-services/ct/scenaria.html#Clinical-images-4 02 KINETICORP Accident Reconstruction & Visualization http://www.kineticorp.com/ *Disclaimer: In the Act of Fair Use, all the referenced visual elements are purely for nonprofit educational purposes. The demonstrated visual elements are copyrighted to its own respective owner.

Figure 2 Applied examples of scientific visualizations with 3D Models. Image Credits: 01 Hitachi Medical Systems: SCENARIA http://www.hitachi-medical-systems.nl/products-and-services/ct/scenaria.html#Clinical-images-4 02 KINETICORP Accident Reconstruction & Visualization http://www.kineticorp.com/ *Disclaimer: In the Act of Fair Use, all the referenced visual elements are purely for nonprofit educational purposes. The demonstrated visual elements are copyrighted to its own respective owner.

Although the use of 3D models are commonly seen in scientific field, its main usage is still lie with the realm of creative entertainment such as movies, animation and video games.  In spite of the fact that most 3D models are built for entertainment purposes, game models are fundamentally different from those cinematic models that used in the movies and animation.  For instance, consider Tolkien’s legendary character – Gollum Smeagol from The Hobbits, its visual representation is highly distinguishable if compared with its cinematic and video game renditions (See Figure 3).

Smeagol Gollum from LOTR The Hobbit EA Game

Smeagol: A comparison study between a cinematic and game model.
Image Credits:
01 Gollum; The Hobbit/ Photo © 2012 Warner Bros.
02 Making of – Gollum – The Hobbit An Unexpected Journey by Weta Digital
03 The Lord of the Rings: The Return of the King, 2003, EA http://www.ea.com/uk/lord-of-the-rings-return-of-the-king
*Disclaimer: All the demonstrated visual elements are copyrighted to its own respective owner. In the Act of Fair Use, all the referenced visual elements are purely for nonprofit educational purposes.

The differences between a game model and cinematic model are mainly segregated by its functional purpose, for the cinematic model is built to entice its viewers’ to suspend their disbelief momentary, and accept what had screened to them as reality for achieving the purpose of entertainment (Read more about: the suspense of disbelief).  In order to persuade the audience to cognitively embark on such novel experience of believing what’s not possible, the passage to the suspense of disbelief lies within realism.  Thus, it quite natural for the built of most cinematic models to have extremely high polygon resolution that could be easily chalked up to millions, for the process of modeling details would require as many polygonal faces as possible to emulate all nuances that found on a real world surface.  In term of computational resources, most cinematic models are very expensive to be rendered and even displayed in the 3D viewport; it would take minutes of time and many cycles of processing power just to generate one frame of lifelike graphical effects that has the quality of “hyper-realism”.

On the other hand, the game model is developed to interact with the player; it has to be light and efficient to be rendered in real-time of sixty frames per second (60 FPS).  The notion of 60 FPS can be explained as: “all the on-screen visual elements have to redraw itself by sixty (60) times within a second by the graphical processing units (GPU).”  Such intensive generative action can also be simplistically compared as the refresh rate, and it is crucial for any real-time graphical system such as digital games and 3D application alike to be able to rapidly update and redraw its transformation in milliseconds for catching up with the players or users’ action without lag.  Therefore, cinematic model is never fit for real-time 3D rendering purposes, as it would take more than seconds to be rendered or drawn by current game engine and consumer level of GPU technology.

In the construction of a game model, the primary concerns are always about the polygon count and the availability of texture space (graphical memory).  As such, game model is also known as “low-poly” model, for it is a highly optimized polygonal model that low in density and fast to be drawn if compared to the cinematic model.  Although the topological resolution of a game model is quite low, it would not lack of any vivid details if compare to a cinematic model.  For instance, by referring back to the Gollum again, the low-poly version of Sméagol had only 3039 triangles, but it still retained a very good approximation to its cinematic counterpart, which weight more than few hundreds thousands of polygon.  This is because in the art of game model creation, there are some advanced normal displacement techniques that help to add some lifelike details onto a low-poly model.

Gollum Smeagol 3D Model Wireframe Comparison: Weta The Hobbit & EA LOTR

The 3D model of Gollum Smeagol with shaded Wireframe display: From The Hobbit & LOTR: The Return of the King (EA, 2003).01 Making of – Gollum – The Hobbit An Unexpected Journey by Weta Digital http://www.cgmeetup.net/home/making-of-gollum-the-hobbit-an-unexpected-journey-by-weta-digital/
02 The Lord of the Rings: The Return of the King, 2003, EA http://www.ea.com/uk/lord-of-the-rings-return-of-the-king

Besides there is a science behind the process of making a real-time game model, there are also a series of principles or “the arts of doing”, for how to craft a high quality AAA-game-model that abides to the requirement of real-time rendering.  Through this article, I would like to share with you about these four (4) principles of 3D game art modelling which being developed and evolved over the time by the veteran practitioners.  Please do not confuse between the terms of game art modelling and the theory of game modelling (Zagare & Slantchev, 2010).

Principle #1 – Model after the reference

References are very important in the production of creative art, and regardless it is for entertainment or edutainment purposes.  As what Picasso said, there really is no abstract art: “You must always start with something” (Barr, 1974), and that “something” is reference.

In the context of digital model making, reference or the act of referencing it is actually a form of visual research whereby you search deeply about the subject that you are about to emulate or replicate.  The source of reference can come in myriad of forms as shown in Figure 5.

Mind Map of Visual References for Creative Production

Mind Map of Visual References for Creative Production

Although references can be easily gathered through the search engine like Google, a good visual reference does not equivalent to beautiful images!  So what makes a good reference then?  A good visual reference must have the following traits:

  • Good resolution; clear and sharp;
  • Provide dimensional data, such as length, position, angle, proportion & scale;
  • Supply texture information;
  • Describe the utility functions of an object;
  • Has good silhouette that could translate into primitive-shape;
  • And, it has to be relevant!

Tips: Never attempt to model anything until you have a clear reference or idea of how the subject should look like, otherwise you will mostly end up re-doing the work.  To learn how to use a reference during modeling, you can refer to this article How to reference a reference.


Pablo Picasso : Simplification and Visual Research - The Bull

An infamous quote from Picasso about visual research (reference) and simplification.

 To learn more about Picasso, please visit https://www.artsy.net/artist/pablo-picasso


Principle #2 Knowing the limits

“Every cool technology will have its own hiccups.  Learn to know them.”

There are always limits to be found on the platforms for game development.  These limits can be categorized into two forms: technical limits and game play limits.  The referred technical limits are actually the interdependent limits that found between the 3D hardware and software that set to host the game.  For instance, the old Nintendo DS-Lite has a hard limit of polygon count, whereby it can only render about 6144 vertices or 2048 triangles per frame at maximum (Wikipedia, n.d.).  This limitation mainly comes from its 32 bit ARM processor that has only 67 MHz clock speed and there is no way for game developers to tweak it other than replacing the soldered processor which deem not practical and near impossible.

Besides knowing the technical limits of a given platform, a game artisan must also pay attention to the game play limits that being outlined in the game design document before the actual production begin.  For instance, should a 3D model meant to be just a static decorative prop that fixated at a distance, then it should be low in term of resolution by using less texture space and polygon count.

The Limits for Platform of Game Design & Development

In summary, to build a playable digital game, the developers ought to know the limits of their targeted platform and respect that limit before designing it.  Such technical understanding is also critical to the 3D game artisans as well; you must get familiar with the frequent limitations of polygon count, texture resolution, and the required level of details for a particular type of game play.

Principle #3 Simplification

“Less is more, and things never looked bore with textures.”

After done acquiring the references and knowing the limits for building a game model, you ought to practice simplification, or simply simplify the geometrical shapes that representing the object that you are crafting. Do not attempt to build every single detail from the reference with polygon.  Instead, learn to substitute all the micro details of the references with texture maps that packed with normal-bump information.  For instance, let’s take a look at the following example.  From a distance, these two pieces of Sci-fi flooring panel would look just like any other modular game object.  They are almost identical in terms of its visual appearance should its wire-frames were not being revealed.  For example, by exposing their wireframes, Panel-2 is considerably an “expensive” 3D model as it had 10 times more of polygon counts if compare to Panel-1! 

Sample of 3D Game Art Sci-fi Floor Panel Low Poly by Lin Chou, Cheng

Sample of Sci-fi Floor Shaded Wireframe display by Lin Chou, Cheng

In the standard of 3D game art modelling, Panel-1 is a better example of game model that adhered to the principle of simplification through the use of bump mapping techniques for saving polygon count.  By and large, Panel-1 can actually be furthered simplify into a mere deflated polygonal-box, if the related game play limit does not required Panel-1 to be fixated on such high level of geometrical details.

Sometimes, the act of simplification can also be generalized as the process of optimization.  According to Merriam Webster (n.d.), the term of optimization meant:

a process, or methodology of making something (as a design, system, or decision) as fully perfect, functional, or effective as possible”.

This definition of optimization codeshare with the value of simplification whereby one of the central practices of 3D game art modelling is to reduce all the superfluous details, such as extra creases, scratches and etc. that being carved out with thousands of polygon faces, yet the process of the reduction must not lose sight of the original visual granularity that a surface once had.  A skillful game artisan will find his/her way to preserve and store all the relief details into a texture map.

Modular Low Poly 3D Building Block

For instance, let’s referring back to the above sample model again (which it was derived from this YouTube tutorial).  The demonstrated building block from the tutorial can actually be optimized into a much lower polygon count by removing all the excessive edge-loop without disturbing its original form as a solid building block.

Tips: Although the practice of game art champions optimization, please do not triangulate the model too early.  Do keep them in quads at your best.  After all each quad are equivalent to two (2) triangles.

Principle #4 Topological Perfection

Besides the differences in polygon count, a game model would have its unique topological requirements if compare to a cinematic model.  These requirements are:



  1. N-GON;
  2. Holes;
  3. Non-planar surface;
  4. Non-Manifold creation;
  5. Stack-and-Intersect meshing (except LOD).

 photo TopologyRequirementforPolygonMesh.jpg


The referred N-GON (see above) is another interchangeable term of polygon, whereby it can stand as a many-sided object or surface.  In real-time rendering, any polygonal faces that have more than four-sided is poised to cause rendering problems such as incorrect shading interpolations (especially on curvy surface), and some severe deformation issue during animation playback.  To an extreme, a rendering system that runs on the older hardware would crash if it is being loaded with a 3D model that plagued with N-GON.  In practice, it is mathematically acceptable and operation-able to have N-GON during the process of modelling.  However, N-GON surface is not efficient to be real-time rendered, as it does not bode well with the generalized subdivision operand, such as the Catmull-Clark algorithm (Catmull & Clark, 1978).  For instance, an N-GON surface would loss its original topological flow when it is being subdivided with the smooth function (See the following podcast).

This could explain why most 3D models that had N-GONs tend to cause operand errors when it is being imported into digital sculpting software such as Pixelogic’s Zbrush and Autodesk’s Mudbox (Spencer, 2011; Naas, 2013: p.38).



The notion of hole or the gaps are usually mistakes that caused by unmerge (un-weld) vertices between the edges of two newly combined-objects.  These unmerged vertices will create a chain of open-borders and it is quite hard to detect it by human eyes. It would make a model more costly to be rendered in real-time as the model would have extra vertex count than usual.  In addition to that, meshes that had holes will cracked-open when it is being “smooth” out with the generalized subdivision function.  Thus, do always to remember to merge the vertices of a half-mirrored object.


Non-Manifold Creation

The non-manifold creation is an interesting modelling error where it can be explained as, “by having  two opposite-facing polygon faces and collapsing them into inseparable one; it would end up just like a piece of paper or simply become just a line in the viewport.”  The term of non-manifold was derived from CAM (Computer-aided Manufacturing) which means non-manufacturable.  A model that has the thickness or thinness of piece of paper (as demonstrated on the above figure) is simply not “printable” by tooling-machine or the 3D printers of today.  The creation of non-manifold model is quite a common beginner’s mistake that one can learn to avoid.  The official Autodesk Maya User’s Guide has clearly explained to us what is non-manifold:

“Non-manifold topology polygons have a configuration that cannot be unfolded into a continuous flat piece. Some tools and actions in Maya cannot work properly with non-manifold geometry. For example, the legacy Boolean algorithm and the Reduce feature do not work with non-manifold polygon topology. The image below shows three examples of non-manifold topology polygons.

  • In the first example (the “T” shape), more than two faces share an edge. This is known as multiply connected geometry.
  • In the second example (the “bowtie” shape), two faces share a single vertex without also sharing an edge.This shape is also possible where two three-dimensional shapes share a vertex (such as two cubes meeting at a single point).
  • In the third example, a single shape has non-contiguous normals (without border edges). That is, the normal on each polygon face points in an opposite direction. This is a less obvious example of nonmanifold geometry.”

Non-planar surface

 photo planar.jpg

In order to understand what a non-planar polygon face is, one must know what makes a planar face!  A planar surface is just a perfectly tangent (90 degree) flat-surface without a dent.  And, as explained by the Autodesk Maya User Guide:

A polygon face is planar when:

      • all of its vertices lie in a certain plane. For example, a triangle face is always planar, because its three points define a plane.

A polygon face is non-planar when:

      • it has more than three vertices, and one or more of those vertices do not lie in the same plane. When a polygon mesh is comprised of quads or n-gons it is possible to have non-planar polygon faces.

“In general, you should try to create planar polygon faces on your polygon meshes whenever possible. Non-planar polygon faces may render incorrectly in your final images or when exported to an interactive video game console. For example, a visual shaded line or an unwanted spike may appear on the polygon mesh where it has been subdivided for rendering because it is non-planar.”

No stack-and-intersect modelling approach

Besides the topological requirement of no N-GON, Holes, Non-Manifold and Non-Planar Surface, there is another best 3D game art modeling practice which is to avoid using the “stack-and-intersect” approach when constructing an art asset.  For instance, by recalling to this YouTube tutorial again, the demonstrated 3D model is a classic example that uses the stack-and-intersect approach (See below).  This gazebo structure aka “The Mushroom”, was a combination of several primitive forms that being stacked together.

 photo GazeboMushroom-intersection.jpg

There is nothing wrong to use this approach when crafting a standard 3D Model that meant for the purpose of linear cinematic playback (in animations or movies).  However, such stack-and-intersect method is not an optimized approach for crafting game model that meant for real-time display.  Why?  It is all about the z-fighting and the depth/ transparency sorting issue again.  For instance, by referring to the above example, it would be a waste of the machine’s computing resources to keep validating which polygon faces to be shown or culled in sixty (60) times per seconds (60 FPS).  These sizable amount of “hidden bottom faces” would eventually eat up the allocated polygon budget and UV-spaces (in the texture maps).  In addition to that, a model that being created with such stack-and-intersect approach will consume more vertex count other than the count of polygon (tris).

Thus, to craft AAA quality of game assets, the best practices is to restructure all the separate-able polygon mesh into one solid composition, and below is a short demo for how to effectively optimize the base-mesh of the previous mushroom structure as one solid piece.


In Summary

To craft a quality 3D game asset, remember to go by these four (4) principles ya!

Principle #1 – Model after the reference

Principle #2 – Knowing the limits

Principle #3 – Simplification

Principle #4 – Topological Perfection

To cite this article:

Cheng, L. C. (2014). Principles of 3D Modelling for Games. Game Art Workbook.  Retrieved from http://www.gameartworkbook.com/game-art-theory/principles-of-3d-game-art-modelling/


Autodesk. (n.d.). Two-manifold vs. non-manifold polygonal geometry. Autodesk Maya 2014: Introduction to Polygons.  Retrieved: http://download.autodesk.com/global/docs/maya2014/en_us/index.html?url=files/Polygons_overview_Twomanifold_vs._nonmanifold_polygonal_geometry.htm,topicNumber=d30e151600

Autodesk. (n.d.). Planar and non-planar polygons. Autodesk Maya 2014: Introduction to Polygons. Retrieved: http://download.autodesk.com/global/docs/maya2014/en_us/index.html?url=files/Polygons_overview_Twomanifold_vs._nonmanifold_polygonal_geometry.htm,topicNumber=d30e151600

Barr, A. H. (1974). Picasso : fifty years of his art (1st pbk. ed). The Museum of Modern Art ; Boston : distributed by New York Graphic Society, New York.

Catmull, E., & Clark, J. (1978). “Recursively generated B-spline surfaces on arbitrary topological meshesComputer-Aided Design 10 (6): 350.

Gordon, D. & Chen, S. (1991). Front-to-Back Display of BSP Trees.  IEEE Computer Graphics and Applications, 11(5), 79-85.

Nass, P. (2013). Autodesk Maya 2014 Essentials. Autodesk Official Press, p.38.

Optimization. (n.d.). Retrieved April 24, 2014, from http://www.merriam-webster.com/dictionary/optimization

Spencer, S. (2011). Zbrush Character Creation: Advanced Digital Sculpting. Wiley.

Wikipedia. (n.d.). Nintendo DS: Technical specifications. Retrieved from  http://en.wikipedia.org/wiki/Nintendo_DS#Technical_specifications

Zagare, F.C. & Slantchev, B.L. (2010). Game Theory and Other Modeling Approaches. The International Studies Encyclopedia, 2591-2610. Retrieved from http://slantchev.ucsd.edu/incollection/pdf/IRCompend-W02F.pdf