Digital Modeling

William Vaughan

This book covers the very basics into introductions of 3D modeling and briefs over some more intermediary concepts.

In the foreword, the author stresses the importance of thinking positively. Thinking in terms of “I have everything I need and know all I need to know now to do what I want.” It's important to not get into believing there's always a secret technique or tool to make things easier and faster. It may be true, but you'll discover it in time as you work. Just do the best you can now with what you know now!

I know I tend to get bogged down into thinking “This must be a horribly slow way to accomplish what I want, there must be a better way?!?” and sometimes there just isn’t. And sometimes there is and you’ll discover it as you push on! Sometimes there is but you don’t discover it until much later.

Production Pipeline

Although not about modeling, it's interesting to learn about what modelers are expected to do on a creative project. What other roles are, know the job names.

There are 3 main phases - pre, pro, and post production. A single model job doesn't have to go through the whole process, but some projects might depending on what's needed.

Preproduction - a dirty creative phase where ideas are rapidly tested and changed. Focuses on uglier assets to get the feel of something instead of spending weeks perfecting something that may change.

Write up a story
Craft visual designs - drawings, paintings sketches, clay models
Storyboard - helps define which parts of a scene get attention and pacing.
Animatics - a movie is rendered using really cheap dirty assets described by the storyboard and quick audio assets. Meant to give a rough idea of how everything will fit together before production.

Proproduction - Stuff actually getting made. A layout is set, work that needs to be done is defined, time to make stuff.

Modeling - A 3D model is made. Either by starting with a box, or starting by sculpting (which will need to be retopology-ed into a better mesh for UVs and rigging). Purely, this stage doesn't do any UVs or rigging, but should prepare the base mesh for those stages. Understanding those stages helps you know how to prepare the models.

Rigging - Bones and other deformation tools are applied to the model so that controls can be made available to the animators. The only goal is to create a rig that's flexible enough for the intended animations.

Scene Setup - Prepares the 3D scene by filling it with assets and props. Determines if new tiny models are needed to complete a look and feel. Takes a holistic view of the scene in progress.

Texturing - Materials and textures are applied to the model. Including ambient occlusion, bump maps, color, roughness, etc. Importantly, also does the UV unwrapping.

Animating - Everything so far comes together briefly and the animators try to create key sequences and poses. A suggested set of stages:

Layout. Animations done using stand-ins.
Blocking. Animations made with low quality roughs of the models or segmented models (which are way easier to pose without causing tears and problems)
Facial animation and lip sync using final models.
Final pass. Secondary animations, offsetting the timing, double check for visual errors.

Audio - Audio artists work with the animators to figure out what sfx should be going on based on the animations, the model's materials, everything.

Effects - Focuses on things categorized to VFX (explosions, particles, magic effects) and things like hair and fur.

Lighting - Taking everything so far, lights are placed into the scene to produce the right (literally) lighting and feeling/mood. Helping setup lights for the rendering process.

Rendering - Not just hitting the render button! Thought goes into placing different assets into different render layers and having multiple passes for the same scene so that compositors can assemble them with as much control as possible. They have access to multiple rendering hardware devices and need to split up the entire rendering task efficiently to make best use of the machines.

Postproduction - Slap it all together and #ShipIt.

Compositing - Putting together all the bits from the render layers stage and altering parameters to produce the best results for the needed design goals. Depending on the workflow, can be a more simple assembly or layers or more complicated where light colors are changed during this time.

Audio - Good audio is so important. Adds at least half of what you experience in anything.

Preparing for Modeling

Reference. Everything. Learning how to use references helps out a ton, and the best way to learn how to use them is to literally just start trying to use them. Finding references, figuring out which work well for what you need and which don’t. Even made-up imaginary things can be built based on real things, and it’ll only help the entire piece make intuitive sense.

Observation. This kinda ties in with learning how to use references. Refs are not meant to be copied 1:1, learn to infer from them. Study what about the reference gives the content that feel and sense you’re looking for? Break it down. Like when studying a game to break down its game design, it’s a poor idea to simply say “I like this game, I’m going to copy it.” What do you like about the game? What mechanics support the fun parts? Which detract? When it comes to 3D modeling, ask questions about how the object exists. Does it have hidden gaps? Is it all one piece, or segmented?

Warm-ups. Prior to modeling, visualize the object you’re trying to model (or take a simple object) and visualize how you’d model it.

Gathering References

Actually play with/see/experience the thing you’re trying to model. If it’s nearby, go take some good looks at it. Medical (real) models for creatures are cool too.
Take photos!
Measure dimensions. Either bring a tape measure or some other standard sized item to record the physical size of the item.
Sketchbook.
Online tips. Craigslist and other online sellers often take many photos of the same exact object!
Avoid… small, blurry, dark/bright images. Copying from someone else’s 3D models and illustrations (referencing them to figure out how to achieve a certain look/feel is okay).

A special tip for when you’re lucky enough to take reference photos literally for modeling. When you have many photos of the same object, toss them into GIMP and rotate/scale them all so they all line up (if it’s like a profile and side kinda shot). Creating a collage makes for super handy 1-image to display reference sheets.

Model Fundamentals

There are 3 core types of models. Polygonal models are made up of vertices, edge, and polygons. NURBS surfaces are networks of curves with smooth surfaces between them. Subdivision surfaces are like a hybrid, made up of polygons, but take advantage of some NURBS smoothing tools.

UV Maps allow for a 3D model to have 2D textures applied to it. Each vertex gets a UV value. You can create multiple UV maps for the same model by segmenting the model out into multiple materials.
Weight maps are a single value per vertex, per bone. The value determines how much the vertex moves when the related bone moves. Either a 1:1 or something like 30% so it only moves a little.
Morph values store an alternate X, Y, Z location for a vertex. An entire model can be given different morphs, each with a specific pose in mind (like a raised eyebrow, or forming an ‘o’ shape with the mouth). Models can blend between morphs. SUPER handy for complicated meshes where doing the deformations by bones would be very difficult.
Selection maps simply store groups of vertices that were selected in the editor.
Edge weights can increase or decrease the sharpness of an edge (2 connected vertices) when that mesh is subdivided smoothly. Usually finicky and instead just use additional mesh.

Nurbs Stands for Non-Uniform Rational B-Splines. A smooth mesh that’s defined by a series of connected splines (polynomial (math) curves). Transformed into polygons at render time. Used for CAD software, and long ago for organic forms (now subsurface meshes work better).

Splines are curves in 3D space, defined with 2 points (a start and end). Bezier curves are a form of spline. Control points along the spline can modify the curve easily. Splines can be patched together to form a 3D mesh plane, extruded, or cloning/duplicating objects along the spline.

Modeling Methods

Build Out - Processes of building 3D mesh quite literally and directly. You direct where each polygon will go and the entire mesh is built starting from one starting polygon/point. These tools are really good for very custom designs or retopology.

Point by point is a method where you first place vertices and then connect them to form the polygons and mesh. It’s a good system for creating letterforms when a font isn’t available.
Edge Extend creates a mesh by extending/extruding from a single primitive plane/edge and then build the rest of the model! Apparently used frequently for modeling faces and heads.

Primitive Modeling - Combine or start with multiple primitives (boxes, spheres, cylinders, cones, etc.) and modify them until done. Very handy for hard models which are often very easy to break down into these primitive shapes.

Box Modeling - Start with a primitive (usually a cube) and extrude from it to form the mesh. Similar to primitive modeling, but the difference is extruding modifying from the one primitive instead of just slapping on new primitives. Popular and used for organic shapes pretty frequently (with a subsurface modifier).

Patch Modeling - Using Nurbs/splines to create mesh. The mesh is formed from the splines combining together to form the surface. Popular for CAD, but it’s a cool tool and can be great for clothing and characters and all! Worth looking into.

Digital Sculpting - A sculpting tool (sculpting in Blender, ZBrush, etc.) is used to create an extremely high poly mesh (millions of polygons). It’s gained a lot of popularity because the artist doesn’t have to fret about the topology of the polygons - sculpted meshes are unusable for games and rigging because of the polycount, so topology doesn’t matter. Additionally, the HD mesh can be used to bake textures in a nearly automated process and then applied to a normal mesh (made using other methods) such that the details from the HD model will carry over!

3D Scanning - Real world cameras captured data visually and interpret it into a 3D mesh. A contact scanner is a tool that has to physically touch a surface to interpret it into 3D. Noncontact scanners work visually.

Textures and Animations - Mesh can be made via textures (using heightmaps) to generate terrain or surface deformations. All instances move vertices and it’s not limited to visual effects like bump mapping, which simulate a high detailed, high poly mesh by adjusting how it gets rendered. Ack! You can use bones too! Bones are normally just for animation, but instead you can use them to deform a mesh (like a hose) much more naturally.

Dynamics - (Dynamic refers to any simulated 3D system.) Using simulation systems to deform a mesh, save that deformation, and use it for other purposes. Like modeling a flat cloth, draping it over a table using a simulation, then saving that mesh with its natural looking bends and creases.

Professional Practices

Just a collection of modern professional goals, terminology, and guidelines.

Organization tips.

For a given project’s root directory, there should be a single folder for all 3D content.
Separate content out by types like animations, objects, scenes, etc.
When working within a group, using a shared directory/location helps a ton. Allows anyone to find and access whatever files they might need.
File naming and saving conventions:

When you first save a file, append a version number 000.
When you hit a milestone, start a new save with another version number later on. Keeps it easy to tell which is the latest version, easy to revert back if needed.

Clean Models

Tips on how to create clean models - models that have a low polygon count and a good topology.

Polygon Count - refers to the number of polygons in the mesh. The fewer polygons used means fewer vertices, which means it’s much easier to paint weights, rig the mesh, have it deform in a more predictable way, assign UVs, render, and more. Real-time rendering (like games) especially prefers lower poly counts.

Poly count really refers to the model’s triangle count. Although graphics processors used to be designed to handle a variety of face formations, there’s been a push to move to just using triangles. Blender now-a-days will count actual triangles of the mesh even when the mesh has 4-sided polygons. Avoid creating n-gons (5 or more sided polygons) which forces the modeling system to determine how to break it down into triangles which might be done poorly. In general, define the triangles yourself. Subsurfaced faces will also multiply the mesh and quickly balloon polycount numbers.

How much is a good poly count? Depends on how the model will be used. Basic rule of thumb is “just enough to have the mesh work for what it needs.” Things that tend to require higher counts - close up, well lit (not dark or shadowed), how much it will bend and deform, and how many angles will the object be seen from. For example, although hands are smaller than a forearm they will deform in many ways and require more faces.

A consideration is to keep several versions of polycounts of the models on hand.

Topology - (aka polygon flow or poly flow) refers to how the mesh is constructed. Good topology is easy to select, manipulate, build, and sculpt.

In general, focus on giving a model edge loops. Edge loops are edges (connecting lines from one vertex to another) that form a single continuous loop. Makes selecting easier and tends to deform well, especially for facial muscles which kind of work like loops too. It’s not a one-all fix and too many edge loops can cause pinching. A line of thought is to limit the model to only using quads which will naturally encourage edge loops.

Polygon flow refers to how polygons connect to one-another across the surface of the mesh. This isn’t something to watch out for, but rather a design decision if you realize you want some polygons connected that currently aren’t (to form good edge loops). A tool that spins edges around to change the flow can be helpful! The cool has tons of different names like SpinQuads, Flip Edge, Spin Edge, Rotate Edge, etc.

Four Sided Polygons are a constant goal over using triangles especially for organic models. They deform better than triangles and especially n-gons. However you will come across times when the mesh basically needs a triangle in order to work at all. A solution is to create a diamond-shaped polygon. One vertex points north and the 3 others are all south.

Rule of Three refers to giving joints 3 edge loops/segments - 1 right on the deformation joint and then 2 surrounding it to support the mesh. Ideal for things like knees, hips, fingers, elbows, shoulders.

Preparing for Production

A guiding list of principles to go over before releasing the model into the world or the rest of your team.

Resize the model to the object’s expected scale.
Model centered at the origin, resting on the ground plane, and facing forward.
Review the model’s geometry one last time.

UV Layout suggestions.

Give high detailed parts of the mesh the most UV area (like the face).
UVs follow the same general direction as the mesh (all point down towards the ground)
Hide UV seams where the model is not going to be seen from (like the inside of the arms)

Rigging suggestions.

Have the character in a standard post (T, A, or “relaxed”)

Polygonal Modeling

The hardest part when given an object you want to model is figuring out the most streamlined way to create the mesh in an organized, clean manner. The book goes through a case study of creating a special soccer ball used for the 2010 world cup which has a really unique seam pattern. Intuition suggests starting with a sphere primitive or modeling the seams from curves, but both of these will end up being difficult to create a clean mesh from.

Mr. Vaughan discovered an unorthodox approach where he started with a tetrahedron (like a 4-sided die), suggesting that great solutions can come from trying new things. The steps went as follows:

Breaking down the shape into its pieces, observing it, seeing how the object works.
From a starting tetrahedron, split it into the 2 main bits: the tips and the sides (the ball is made of 8 shapes, 4 sides and 4 tips). Subdividng the mesh can help create equal sides.
Clean up the subdivisions and reduce the mesh back to simple/few faces. Still very flat.
Replace triangles at the tips with diamond 4-sided polygons, introducing new vertices and leaving a center vertex.
Break down the sides similarly, connecting the tips together with freshly cut edges.
Add new edge loops around the tips to increase the number of polys used by the sides so that it roughly matches the size of the polygons on the tips.
Assign the sides to separate vertex groups and the tips to 1 group. This will help make selecting them later so much easier.
Subdivide more, and use a special spherical deformation so that the mesh tries to fit a spherical shape.
Assign more polys to the tips so they share a roughly similar area to the faces.
Create the seams on the ball by beveling each side and tip.

Very interesting! My takeaways are listed in the cool tips section.

Subsurface Modeling

Although the last chapter also used subsurfaces, this chapter uses them more?.. Seems like the distinction is modeling with a mesh with a subsurface modifier on it (this section) as opposed to subdividing to create new polygons only when needed.

A common issue when working with subsurface modifiers is that every bit of mesh is smoothed meaning desirable hard edges and corners are totally lost. The solution really is to introduce hold segments - edge loops surrounding the nearby corners (on both sides) so that the subsurf smoothing doesn’t over-smooth those corners! I always worried this would introduce an unnecessary amount of new polygons, but I guess it’s standard professional practice! I do wonder what the thought behind doing this over using edge creases is.

Remember to practice preserving edge loops and forming quads. This may mean introducing new cuts to bridge corners instead of making a bunch of triangles.

Modeling a Realistic Head

The top 2 popular methods is either box modeling or edge extending. I prefer box modeling - working from a rough head shape then into details, as opposed to working details-out via edge extending. Edge extensions are a teeny bit easier to use only because it’s obvious when verts are in the correct spot or not as opposed to working more artistically with defining details.

When using a reference, mirroring it down the middle is a good idea for building the initial mesh, but your creature should have asymmetries by the end of it all.

Eye Tips

Use a sphere primitive to start (UV sphere, fix the triangles into quads)
Bevel the front to create the iris and pupil (backside of the front/inner eyes.
Create another sphere with bevels out to act as the cornea/lens.

General Tips

Some parts, like the eyes and mouth, work best modeled separately and then attached. Focusing on good poly flow for just those elements.
Lots of edge loops are good for areas that will have lots of animation/deformation.
You can ignore areas that won't deform as much, like human ears and the back of the head. Still go for quads, but relax on edge loops.
To create good edge loops around animated bits, build the loops as you build out from that object. Easier than trying to subdivide them in sometimes.

Character Building Tips

Start with the most interesting parts of the character, usually the head because that defines the character. The rest supports it.
To create good topology...

Utilize edge spinning to ensure good edge loops.
Edge loops around the mouth, eyes, one for the jaw.
Eyelids also follow the suggestion of having 2 nearby loops to help aid deformation.

4 point polygons over triangles

Sculpting Tips

Really don't worry about polygon flow. If the model is to ever work outside of sculpting, it will need a retopology work regardless.
Get and use brushes! Learn how they work, what they can do, how they are effective in which ways.
If working from a geometric primitive, instead of a rough base mesh, you will need to do lots of pulling to create the basic form.

Inflate tools can help form muscles and things like knees.
When things get a bit too wonky or weird, the smooth brush will help even it out.

Work large blocks down to details. Get the overall idea and shape first, then add details.

To help with this, turn the model shadeless fully black (like a silhouette) to better see important shapes.
Sometimes avoiding necessary detailed spots like hands and feet can be avoided during the blocking phase.

Masking helps a lot when the program starts to lag.
Swap materials (shaders) on the object regularly to make sure the model will look right under a variety of lights and materials.
There are tools that apply fine details to a surface, like skin or leather, via a brush's alpha texture. ZBrush calls it Drag Rectangle.

Overlapping alphas help create a more asymmetrical natural feel.

Turn off symmetry when done with the medium details and make asymmetrical changes with rotate/transform tools.

Don't do this for assets you want to retopology later!! It will double your workload. No advice for when to make asym changes though.

The clay strip tool can be good for simulating layering muscles.

Retopology Tips

The Edge Extend method of modeling works exceptionally well here since you really just need to line up the vertices.
The number of polys to aim for depends on how the object will be used. The book aimed for 6k for a Unity game crrature asset.
Otherwise, it really is just a lot of work.
Once done, slice it up to make a UV map.
Paint in 3D to create the color map. It should be possible to paint just colors as you would paint physical textures like earlier. Apply some of those detail textures (skin, fur, w/e) to the color map too.
Convert the color map to a bumpmap (height map) by making it greyscale. Anything automatic will likely look nicer and work better than manually re-painted.
Bake a normal map using the HD mesh.
Bake an Ambient Occlusion map also.

Cool Tips

Things to do when not at a computer

Some fun ideas and exercises to practice when you’re out and about.

Break down a physical object and visualize how you’d model it. Go through the full procedure - how would you start the model? How would you refine it and add details? What’s the texture of it like and how would you replicate that? Etc.
Sketch books! Trains you to be better in observation, how to break down an object into the parts you like, replicating it so it looks good, etc.

Modeling Tips

Encourage edge loops.
Avoid triangles, try to convert them into 4-sided polygon diamonds.
Assign vertex groups to key bits of the model for easy selection later on.
Bevels help a lot on clothing and hard objects.
Instead of always starting from a cube, think about what other primitives can help form or be transformed into the mesh you want. If it will work, it’ll work really well pretty quickly!
You can build good mesh from drawing vertices and starting with an N-gon by using an inset tool, then deleting the internal n-gon. The result is a series of quads forming the border!
Don’t worry about poly counts during the initial building process unless the project is specifically low-poly or for a low quality asset. Objects can be easily made lower poly from an HD mesh, but it’s hard to make a low quality model HD.