Stupid Shit You Should Avoid Doing

Don’t model everything in one mesh (split into components when you get the chance)

If the model is broken up into different parts irl, make sure the model is broken up into those parts (will make your life much easier)

Don’t bevel every edge, let the normal map take care of some of it
N-Gons and Triangles are okay, even if your dad beats you for using them

Smart Shit You Should Start Doing

Unwrapping tips

UV: use relax, straighten cylinders, straighten edges (avoid normal edge aliasing)
UV: mirror and reuse geo
On SP, everything out of the 1:1 uv space will not get baked

Texturing tips

Don’t put fingerprints all over your texture
Don’t over-rely on generators/newer texturing software - avoid SP’s default smart mats like the plague
Google image search > by resolution > high res

Use hotkeys (for the love of god)
How to properly bake

Rely on references, lots of them
Make a high and low poly (sadly even this should be mentioned)
Soften bevels on high poly (consider practical scale model will be seen in)
Consider the function of your models (figure out what details and why, avoid making pointless shit up)
Optimize topology for low-poly (delete pointless loops, silhouette is important, explain polycounts for modern games, etc)
Don’t be afraid to use geo (enough sides on cylinders/curves, knowing when to model detail vs put it in the normal map)
Distribute topology intelligently (no huge meshes with faceted silhouettes but tiny details with a million polys)

UVs

Unwrapping tips

UV: use relax, straighten cylinders, straighten edges (avoid normal edge aliasing)
UV: mirror and reuse geo
Vertex normals/smoothing groups to UV islands

Texturing

Physically correct values
Moderate albedo values
Use reference
Mix of generators & custom work

Modeling (General)

Make a high and low poly (sadly even this should be mentioned)
HP

Don’t model everything in one mesh (split into components when you get the chance)

If the model is broken up into different parts irl, make sure the model is broken up into those parts (will make your life much easier)

N-Gons and Triangles are okay, even if your dad beats you for using them
Soften bevels on high poly (consider practical scale model will be seen in)
N-Gons and Triangles are okay, even if your dad beats you for using them
Consider the function of your models (figure out what details and why, avoid making pointless shit up)

Don’t bevel every god damn edge, let the normal map take care of some of it
Optimize topology for low-poly (delete pointless loops, silhouette is important, explain polycounts for modern games, etc)
Don’t be afraid to use geo (enough sides on cylinders/curves, knowing when to model detail vs put it in the normal map)
Distribute topology intelligently (no huge meshes with faceted silhouettes but tiny details with a million polys)

Preface

This list is not exhaustive. It attempts to cover the most common mistakes new artists tend to make on essential aspects of 3D asset creation for videogames.

In an attempt to remain succinct, the guide may not provide an exhaustive answer on how to correct every single of those mistakes. A more elaborate explanation on the techniques presented to solve these problems is often available with a simple Google search, or by consulting CG forums.

Sometimes, the challenge in fixing those mistakes is simply to gain awareness of them in the first place. This is what this guide seeks to aid with.

Essentials

(The really important ones)

High Poly to Low Poly

The base of game asset creation is to produce a high poly first, then create a low poly from it and use the high poly to bake a normal map. Not only is it easier and quicker to model and bake certain types of details than it is to paint them manually into your maps, it’s also the best way to get clean bevels onto your edges (without having to chamfer everything). This mean you’ll have two models - one with a high polycount that’ll have all your details, and a more optimized one that will go into the game.

There are several other techniques used for specific purposes (trim textures, mid-polys, blended materials), but high to low poly is the first one you should master, especially if you’re creating unique assets (such as a weapon or piece of furniture). You will then be armed with the necessary knowledge to easily branch out into other game asset creation techniques.

Use Hotkeys/Keyboard shortcuts

If there’s one thing newbie modelers don’t do enough that they should absolutely be doing, it’s to learn proper hotkeys. If you’re gonna remember one tip from this list, it should be this one.

Basic hotkeys to switch between translation, rotation and scaling gizmos (W-E-R in most programs) and to switch between different selection modes (1-2-3-4-5) are immensely helpful. Keep in mind to model efficiently, you should hotkey every single action you find yourself doing on a regular basis (such as extruding, chamfering, cutting, welding, etc.)

Using hotkeys will literally save you hours or days of modeling time, so it’s the first thing you should learn (it’ll make learning and practicing everything else after it much faster).

If you’ve been using your modeling program for a while and find yourself still not knowing the hotkeys for it, watching an “Intro to X” tutorial is a great way to learn them, because roughly 50% of what these tutorials cover is hotkeys and hotkey setup.

Use Reference

Your 3D creations should always look believable. If they don’t (even if they’re stylized, which does not count as a way to cheat!), the main reason is often that you didn’t use enough reference.

Reference is important in every stage of asset creation, from the initial blockout to the final textures. Especially as a beginner, you should not attempt to create things purely from imagination, but rather try to break down the design and characteristics of real objects and integrate them into your models.

Things such as the width of a door, height of a doorknob, or the engraved patterns specific to an art movement such as Rococo or Baroque all follow specific conventions that should be researched and applied. Doing so will convey your models a greater sense of realism and belonging.

Remember, if it looks off, use more reference!

Modeling - High Poly

Separating a model into components

In real life, people don’t make objects out of solid blocks of matter. And neither should you, unless you’re sculpting a statue. Almost every single manmade prop you can think of, from weapons to cars and furniture, is made of many small parts that are easier to create on their own.

In many cases, trying to break your models into components won’t only make them truer to life, but will also make the modeling experience significantly easier for you! In your high poly, having separate components gives you a place to terminate your edge loops (so they don’t run across the whole model if they don’t need to). In your low poly, it gives you a great place for texture seams. And overall, it makes modeling complex objects more manageable by dividing them into simpler components.

Bevel Thickness

When modeling for a game, you want to give your objects thick, beefy bevels. Generally, your bevels should be bigger and softer than they would on real life objects. The reason for this is simple: thin bevels do not hold up well in games when seen at a distance.

Another reason for using thicker bevels is to help them read properly on your normal maps, especially as those might be downsized in-game due to mipmapping.

Modeling - Low Poly

Avoid putting tiny bevels everywhere

Not everything needs a bevel. If you’re fairly new to 3D, and making props that rely on a normal map (using the HP to LP workflow), keep in mind you don’t need to bevel every edge in your LP. Your normal map is perfectly capable of depicting beveled edges on hard corners (as long as you remember to do your smoothing splits). Leaving the LP free of unnecessary bevels will make for an easier modeling, unwrapping, and texturing experience.

Bevels in the LP: use them or not?

Your high poly should have bevels (this will usually take care of itself if you use smoothing techniques). For your low poly, that’s a different matter.

Basically, if you’re baking a standard HP to LP asset with unique maps, the choice to add bevels in the LP or not is up to you. It’s not a necessity, and has some pros and cons. If you’re using a workflow with tileable materials and trims (think Star Citizen), you need bevels in your LP to properly round your edges.

Know how many loops to use in your bevels. Note how there is little difference between the first and second edge (unbeveled and beveled) once normal maps are applied to both.

Normal maps are fairly efficient at representing soft/rounded edge, and as a general rule, any edge too small to show a visible curvature when seen as a silhouette should simply be left to the normal map to handle. Very large beveled edges can be left as LP geo (though they should rarely require more than two or three edge loops). For anything in-between, the choice is up to the artist, taking in consideration the following pros and cons when beveling:

Pros:

-Edge is represented visually in the silhouette of the model

-Allows continuous textures over edge (no need to split UVs because harsh normal angles are reduced)

Cons:

-Can make your model more difficult to unwrap

-Destructive (difficult to get the original edge back should you need to do modifications, unless you use a reversible modifier)

-More time and effort than leaving things up to the normal map

The performance impact of beveled vs hard edges in a LP are negligible - as a hard edge will require a UV split, it will have a doubled vertex count and thus the same overall cost as a beveled edge (with continuous UVs).

Even without a normal map, you don’t need more than 1 edge loop of bevels to get a smooth corner in a LP.

Appropriate polygon distribution

Basically: give bigger/more visible things more polys, and smaller/less visible things less polys. Consider the distance at which your model will be viewed; you wouldn’t give more polys to a small coffee cup or a screw on a door than you would to an entire chair or desk.

One good example of this is weapon models in First Person Shooters. As the sights/scopes tend to be put directly in the player’s face, they’re given a higher polycount than the rest of the weapon. Cylinders on the sights might be 32 or 48-sided, while the barrel of the gun itself might only be 16-sided despite being bigger, as it won’t be in direct view of the player. If it needs to hold up to close inspection, give it more geometry. If not, save that geo for elsewhere.

This also applies within your models themselves. Don’t put a tiny 32-sided cylinder onto a huge 8-sided one.

There are two main areas where spending extra polygons is worthwhile. The first is on your curves: try to give your curves and cylinders an appropriate amount of polygons so faceting is as hard to see as possible. The second area is on smaller details that are too significant or silhouette-breaking to leave entirely to your normal map. If you have a tiny screw, it can probably be left as a normal map only - if you have a big red button that pushes out of the silhouette of your mesh a lot, it’ll need geo, as seeing it reduced to a flat texture will look unnatural.

What is an appropriate polycount?

Simple answer: it depends.

More complex answer: Here’s a chart of polycount estimates for various platforms, in 2017.

Even more complex answer: You should aim within the figures in that chart, as long as you need them. If a table has a 5k polygon budget but you can pull it off with 2k and have it look the same, all the better. What matters is for your in-game geo to be optimized. This means no unnecessary edge loops, no high polycounts on tiny details, no turbosmooths on in-game models (even if they still fit within polycount).

For portfolio pieces, you are mostly free of polycounts, as long as your models are intelligently optimized. If what you’re modeling needs 50 000 polys to look nice, so be it. But every single of these polygons should be justified. Extra polygons to remove faceting from a curved shape is fine. Extra polygons because you didn’t bother to delete unnecessary edge loops is bad. The main thing employers will care about is how intelligently you use your polys, not absolute polycount.

Keep in mind a low polycount isn’t an artstyle, but a technical limitation.

N-gons and triangles

They’re fine on a low-poly. No, seriously. If your model isn’t going to be deformed in an animation, you should feel free to use tris in it.

Any model that gets sent into a game engine gets triangulated. This means that all your n-gons and quads are turned into tris. Thus, your LP having tris, and even n-gons (as long as you know how they will triangulate) is no problem at all.

This is not necessarily true for “bake-less” (no HP) models (Star Citizen-style). Badly placed tris on those might cause smoothing issues. But as long as you bake a normal map from your HP to LP, any smoothing deviations caused by triangles in your LP should be compensated by your normal map.

Unwrapping

Maximizing UV space

The primary goal of an UV unwrap is to maximize the useable space on a texture. Textures with empty space are bad. While certain meshes with a low number of UV shells might be difficult to unwrap efficiently, appropriate techniques should always be used to ensure as much of the texture space as possible is utilized.

Some objects with very long UV islands (for example a column, hockey stick, or telephone pole) are better unwrapped with a rectangular aspect ratio (i.e. 1024 x 2048, 512 x 2048) rather than as a square.

An improper unwrap leaves many unused gaps (the holes in circular UV islands, and gaps between irregular-shaped islands that are not fitted together properly).Lack of mirroring also leaves several unique UV islands that could be turned into duplicates.

The same texture here with proper packing, symmetrical islands shared/stacked, and with a 2:1 aspect ratio (which made it easier to pack islands such as the long strip in the bottom left).

Mirroring/reusing faces

Whenever possible, identical elements on a model should share the same UVs to minimize the space they take. Many symmetrical objects can have their UV space reduced by 50% simply by unwrapping half the model, and duplicating the UVs to the other side. Reuse of texture space can also be applied to smaller repeating elements on a model, such as screws and bolts.

On this model, the faces in green have been mirrored, as they exist as perfectly symmetrical copies on the other side. Some areas, such as the optics and upper receiver cover, have been kept as unique on both sides as they are meant to receive text or unique patterns which cannot be mirrored.

Mirrored islands may be offset by 1 UV tile to avoid overlap - offset polygons will still display the texture in-game, but won’t cause errors during the baking or texturing process.

Note how the islands of the right tile in the image above correspond to identical islands in the original texture space, to the left.

Straightening edges

You should try to align the edges of your UV islands to a 90 degree grid as much as possible. This has two main advantages: it reduces aliasing on normal map edges (normal map bevels on diagonals will always be aliased), and it simplifies your packing by giving you straight edges that can easily be aligned with each other.

Straightening edges of non-square polys can introduce distortion in your texture - this distortion can be seen on text and patterns that are applied in texture space (not object space), but will often be negligible on smaller islands, and islands that don’t require a lot of straightening to become rectangular.

Texturing

Design/Misc.

Stylization is not the easy way out

Stylization is harder to learn than realism. A common misconception among beginners is that stylized 3D is easier to pull off than photorealism. While it might be easier to produce low-quality stylized assets, making good stylized 3D is not easier. In fact, it’s likely harder.

Efficient stylization involves understanding then simplifying reality while exaggerating its most striking features. To pull it off requires a deep understanding of the subject you’re planning to stylize. Lack of understanding creates lazy or “off”-looking models. If you’ve ever seen someone try to draw anime or cartoon characters without having a proper understanding of anatomy, this situation is roughly similar.

Just for the same reason there’s good-looking and bad-looking cartoons, there’s good-looking and bad-looking stylized 3D. The stylized work of companies such as Disney and Blizzard requires an equal amount of work and mastery as any “photorealistic” production.

The difference is stylized work requires you to master realism first, while realism doesn’t require you to touch stylization very much. Starting with realism will actually make your learning easier, especially if realism happens to be your final goal anyway.

Don’t design “randomly”

Link to full image:
https://puu.sh/wAX93/00c18ad1c3.jpg

Artists are often guilty of filling their models with “meaningless” detail. While greeble isn’t a bad way to fill in some of the blanks on a model (and is often necessary), you can’t make everything in a model be greeble. A good artist should think of the function of their design.

For sci-fi hard surface models, this can involve searching for images of various machinery parts that are appropriately related to the prop (it doesn’t actually have to be functional, just look like it could be). For current-day or historical models, making sure the look and function of the object matches appropriately its real life counterpart. Fantasy or stylized models are not immune to this, as they, too, tend to be solidly ingrained within real life design trends.

Something as simple as a generic sci-fi door can present many design problems to solve and integrate in your model. How does it open? Does it pivot? Slide? If so, does it have visible guiding rails? A handle? A locking mechanism? Can the user interface with it in different ways, for example by a control panel? How is it powered? Mechanically? Electrically? Through pneumatic energy? Are there visible cables or pipes to deliver power? Do all of these elements have a real-life counterpart that could be used as a starting point for reference?

This calls us back to a previous tip, which is to use reference! The most talented concept designers don’t make everything on the fly - they rely on an extensive library of design shapes they’ve built up and memorized over the years, and often do their own custom research for every new modeling project.

Scripts and When to Start Using Them