Development Staff

Anime Style Character Modeling TIPS

Translated by @LeoBGKK via DeepL, Yandex.Translate.

About today’s talk
 Speaker Profile

Junya Christopher Motomura

Arc System Works Co.
Lead Modeler / Technical Artist / Various others
I was a modeler and wanted to write shaders, so I became a
technical artist.
Modeling, rigging, shader creation, giving talks, etc.
Currently in charge of technical art support and R&D.

Representative Works
⇒GUILTY GEAR Xrd Series
Lead Modeler / Technical Artist
⇒DRAGON BALL FighterZ (Bandai Namco Entertainment)
Director / Modeling Supervisor / Technical Artist
 Table of Contents
PART1: The Modeling Workflow of Guilty Gear Xrd
Based on a character model from the Guilty Gear Xrd series, I will
show the process of making a 3D object from a drawing.

PART2: Creating a face shape that won’t Break

Common mistakes beginners make and how to avoid them. These
are some tips on how to make anime shapes while maintaining
3-dimensional consistency.

PART3: Practical Normal Editing

How to control normals to get shadows of any shape. I will explain:
“What are normals, in the first place?”
 About this talk

Any type of 3D software is acceptable.

As this lecture is a sequel to the previous lecture on "GUILTY GEAR
Xrd", Softimage will be used for demonstrations. However, we have
chosen tips that can be applied regardless of the software.
Focus on creating mesh shapes.
I won’t be covering special shader techniques or how to draw textures.
the main focus of the talk is on how to shape the polygon mesh. There
will be some discussion on bones and skinning, but I won’t speak on
animation.
It’s about polygon modeling.
Because Guilty Gear Xrd doesn’t use sculpts, there will be no mention
of Normal Maps. Instead, I’ll talk about Vertex Normals a lot.
 Part１

The Modeling Workflow of

Guilty Gear Xrd
 Our Modeling Goal at ArcSys

How to draw out the appeal of a character

with a 3D model.

⇒I want to transmit the appeal of the 2D character design into a 3D model

⇒I want to create a model that is as versatile as possible by finding a
compromise between 2D design and 3D convenience.
What to focus on when starting

⇒The most important thing to start with is

recreating the silhouette
⇒Proportions can completely change the
impression you give
 The slightest difference will affect the feeling
Head Size -5% Head Size +5%

Simply scaling the head

by 5% can make a
character look more
mature, or more childlike.
 GGXrd Workflow

Modeling based on a picture

 TIPS : Put the original image next to the model!

By placing it right next to the model, you can see the differences at a glance!

⇒Place it directly in your viewport, not in a

separate window or something.
⇒You don’t have to put it on top of the mesh you are
working on. Displace it so you can compare both.

Huge benefits:
By placing them in the same window,
you’re able to directly compare the sizes
It’s a good idea to make the reference square, to and ratios of the figure.
ensure there’s no aspect ratio distortion.
Match the balance by comparing to the original
Example:

“Huh? Are my legs a little too long?” “Is this about right, based on the concept?”
→ Correction
 A suitable reference

⇒The view you will see most often in the

product, or the most demonstrative.
⇒For fighting games, this is often the
so-called “idle pose”. Good reference, you can easily tell the
length of the limbs.
⇒It’s important that the proportions of the
limbs, head and body are clearly
understandable.
→Avoid images with a strong perspective,
such as high or low angles.

A bit of a high angle, so not great as

modeling reference.
Workflow of modeling from reference
 Placing and scaling the reference

Place the reference as a Adjust the character size bsed

textured plane within the 3D on another existing character
viewport. model.
 Modeling from reference #1
When modeling from scratch without a base mesh:

Start modeling based on the Roughly outline the torso. Create limbs while referencing Place the limbs according to
reference. the original drawing. scale.

⇒If you already have a base mesh with a similar physique, you can start
your model from there.
⇒In that case, start by matching the overall physique from the reference.
 Modeling from reference #2

Detail Detail

Put the bones in during the Compare the pose and the Textures and details are added
low-poly stage and make the reference. Adjust the balance once the silhouette is set. Match
reference pose. Adjust the accordingly. everything according to the
proportions here. reference.
TIPS : Add bones as soon as possible.
TIPS : Add bones as soon as possible.

⇒In a T-pose or A-pose, it’s hard to tell whether

the length of the limbs matches the reference.
⇒Before finalizing the model, pose it the same as
the reference image, while in the very low-poly
stage.
⇒This preliminary rigging isn’t necessarily final.
⇒Ideally, you should be matching the proportions
from the reference just with the base mesh and
bone rotation, without having to use any advanced
rigging or scaling.
Summary: Modeling from reference
 Summary: Modeling from reference

⇒Basically, what we’re doing here is copying from 2D to 3D.

⇒All you need is eye-copying powers. Observation,
interpretation and reproduction, not tracing.
⇒However, the goal is not to 100% perfectly reproduce the
reference.
→”Perfect reproduction of reference” is not always the same as
a “High Quality Model”.
⇒If you’re too particular about a complete reproduction of the
reference, and neglect other areas, you will end up with a model
that isn’t good enough.
⇒What is important is that the complete model expresses the
character. In the end, the reference is only a guide.
 Part2

Creating a Face Shape that

won’t Break
Why do anime-style faces “break”, anyway？
Premise: Anime faces are broken to begin with.

Because the stylization of the face comes from exaggeration

and omission, it’s inevitable.
In fact, often the “inaccuracies” are the source of the appeal.
There’s a type of face that is less likely to break
For example…

(?)

The more realistic the face is, the less likely it is to break.
(“Realistic” in a sense, of course.)
The more deformed it is, the more likely it will have issues.
 There is no way to completely prevent breaking

An anime-like face = Extreme and “Surreal” deformation.

Therefore, it’s hard to completely create it as a 3D solid.

“Breaking is inevitable” So you have to use bones, morphs, etc.,

and modify them as necessary.

But does that mean there’s nothing we can do on the model itself?
To avoid breaking as much as possible, and to get closer to the ideal look
there are many things that can be done.
 Why is it so difficult in the first place?

It’s easy to reproduce the look from It’s possible to reproduce the look from 2
1 direction directions
(If you do it on a plane, it’s possible to match it (You can shift vertices back and forth to match
100%) the depth)
The difficulty of modelling a deformed face

As you try to define looks from several directions, the difficulty rises!

From this point on, you can’t just “trace” from three views. This is
the area where your skills as a modeler will be tested.
TIPS : First, let’s make a Diagonal View face
 TIPS : First, let’s make a diagonal view face

⇒This is the most common angle. Front and Side views are rather rare.
⇒There are many stylized parts that cannot be read from the front or side.
⇒Because it’s in the middle, if the diagonal face looks right, it’s easy to get the
other angles in-between.
 How to make a diagonal view work

“Making a diagonal view”

means:
Create a “three-dimensional” face

Achieving both anime-like deformation

and three-dimensionality to minimize
breaking.
What to remember when creating a diagonal view
 What to remember when creating a diagonal view

Cheek line Eyebrows and Eyes

Retracted Jaw line

mouth
TIPS : Adjust the cheek line!
TIPS : Adjust the cheek line!

→These lines are invisible in front and side

view.
→They greatly influence the feel of Sharpness,
Softness and Ruggedness of the character.
→This line is the lifeblood of “bishonen” or
“bishoujo” characters.

TIPS:
→Create a dedicated edge to
define this line, and split the
polygons accordingly.
→Do not spare polygons to get the
exact line you want.
Retracting the mouth in a diagonal view
Retracting the mouth in a diagonal view

→A common feature in the anime style is to

draw the mouth slightly retracted, when the
face is at an angle.
→This is important if you want a sharp mouth
→If you don’t do this, it can look sloppy.

This effect can also be achieved by using

bones, or morphs depending on camera
angle, but we can create this effect from
the start, by making some adjustments
while modeling.
 TIPS : Pull the corners of the mouth back!
Explanation
DEMO
PO
GOOD POV
NG V
Push the corners of the mouth
backward.
When you create depth across the
mouth, you affect the diagonal view.

Hypothesis: The style of anime may

itself be trying to allude
to this sense of depth.

TIPS:
⇒Push the corners of the mouth
back to create the illusion of
depth.
⇒Be careful to erase the contour
lines around the lips.
 Common Beginner mistakes: Mouth modeling
NG
⇒It looks unnatural, especially when the mouth is
open.
⇒Too tied to the symbol of “Lips”?

Connecting the corners to a single vertex.

BETTER TIPS:
⇒By adding thickness to the corners of
the mouth, the structural credibility is
enhanced.
⇒It further pushes the mouth back,
slightly.
⇒It looks more natural when the mouth
Add thickness to the corners. is open.
Three Dimensional Eyebrows
Three Dimensional Eyebrows

⇒An important element of a

“Handsome” face.
⇒Especially important because, being
around the eyes, generally the focus
point, the slightest mistake can cause
discomfort in the viewer.
 How to capture the eyebrows and forehead

⇒The distance from the top of the eye to the space between the eyebrows gives
the impression of “finely chiseled” features.
⇒It’s easy to think the forehead is flat, but it is only flatter near the center.
⇒The distance between the forehead and the eyes is deeper the closer to the
center of the face.
Shaping the Eye: The Key is a difference in width

A diagonal face looks more natural when the width of It feels awkward if the width is about
the eyes is different from closest to furthest. the same…

⇒ In anime faces, the width of the eye furthest away is dramatically

narrowed.
⇒ The style exaggerates the sense of three-dimensionality by changing the
width between both eyes.
⇒ Even a real life person changes a little, so this stylization makes sense.
 Common Mistakes when shaping the eyes

Top
View

Alright……?
Trying to achieve the width The width is narrowing nicely,
difference by placing flat at first glance.
eyes diagonally.

Sure, the width changes, but…

 Common Mistakes when shaping the eyes

NG
NG From a low angle, you get
droopy eyes, or the eyes seem
Even in profile, you can see the white of to be unfocused…
the eyes, which is unnatural…

This results in several problems for other angles!

So what should I do!?
 TIPS : Better Eye Shaping Techniques

It’s better to create a width

difference by shifting the
BETTER
corners of the eye back.

⇒Instead of slanting the entire eye, move only the corner of the eye back, to create
the illusion of depth and an eyeball-like roundness.
⇒Even large anime eyes can be changed like this, to some extent.
 TIPS：Better Eye Shaping Techniques
BETTER BETTER

TIPS:

3D Shape Interpretation is Key!

⇒By re-shaping the eyes three-dimensionally, they’re less
likely to break when viewed from other angles.

⇒ In the end, it boils down to “Let’s model it properly, while taking into account the
three-dimensional aspects of reality”.
⇒Even if you don’t model the eyeball, just considering the three-dimensionality of the
shape of the eyelids makes all the difference.
Jaw line
 Jaw line

⇒Along with the cheek line, this line forms

the contour of the diagonal view.
⇒The distance between these two lines
defines the width of the diagonal view.
⇒Like the cheek line, it greatly affects the
sharpness or ruggedness of the character.

In order to have a consistent look, we want to

align the width of the face from the front, with
what we see diagonally, to some extent…

This tends to be a problem…

Common Problem: The Diagonal View is too wide.
Huh? Something subtle…?

Wide?

The front and profile faces are just as I imagined them,

but when you look at it diagonally, the face is just too
wide and ruins the look.
 I want control over this width

Strangely chunky… I wish it was this wide…

 A common chain of mistakes in diagonal views

The diagonal view

looks too wide… The width of the face is
narrowed and the
problem is solved.
However…

I don’t want to mess

with the front view, so I
After all the squishing it
sacrifice the looks like a festival mask,
“low-priority” side view with no three-dimensional
and try to squish the effect.
face back and forth.
→As a result, the shape
quickly collapses in low
NG or high angle views.
DEMO
Analyze the cause: If you know it, you can solve it
Why does it even look extended when viewed from that angle?

Objects with clear sides increase in

width when viewed from an angle.
TIPS: Separate the jaw line from the contour line

⇒The root cause is confusing the frontal face contour with the jaw
line.
⇒This is a common pitfall when modeling by drawing vertices out from
the front of the face
TIPS: Separate the jaw line from the contour line

Adjust in the
X-axis

⇒The contour line of the face is the widest part of the face.
⇒The contour line of the face defines the ridgeline before the jawline.
⇒By narrowing the width in the X-axis, you can obtain the jawline you
want.
 A diagram of the solution

■Front Line ■Front Line

■Jaw Line ■Jaw Line

⇒The frontal contour line is seen in front of the jaw line.

⇒The width of the overall diagonal view can be adjusted by narrowing
the jaw line sideways.

⇒If you don’t do this, you won’t even have a front line defined at all.
 Comparison

⇒I was able to adjust the width of the diagonal view without

changing the appearance of the front or side views.
⇒The key to solving issues is to identify their causes without
being bound by preconceptions, and to think flexibly.
Summary: Creating a face shape that won’t break
Summary: Creating a face shape that won’t break

⇒ When you aim for “style above reality”, some kind of breaking is inevitable.

⇒The best way to deal with this is to use bones and morphs on a case-by-case
basis.

⇒However, this can be mitigated by planning the model itself.

⇒The objective is to have both stylized deformations and three-dimensional

consistency.

⇒Plan the three-dimensional consistency in a range that doesn’t detract from the
charm of the style.

⇒To achieve this, we use surprisingly realistic modelling techniques.

 Part3

Practical Normal Editing

What are normals, in the first place?
 What are normals, in the first place?

⇒The direction of each vertex is represented by a vector

⇒It is defined by three values representing each axis (XYZ)

⇒Put simply, the results of lighting greatly depend on the orientation of

vertices and the orientation of the light source.
 How are normals determined?

Usually, they are automatically calculated by the 3D software.

⇒They are calculated from the positions of connected vertices.

⇒This is why the shadow changes depending on how edges are
connected

An example of the shading changing

depending on how the faces are divided.
 So what is normal editing?
Normals are set automatically, but users can change them to any direction.
(Usually referred to as “Custom Normals” or “User Normals”, depending on the software.)

Auto Normals Custom Normals

You can directly affect resulting lighting by overwriting the normal directions!
→This is very useful for celshade styles which require stark shadows.
 Results of normal editing
Auto Normals
Lighting depends on the shape of the mesh (position of vertices). If
you want to get a good-looking lighting result, you are dependent
on the mesh shape.

Custom Normals
But by separating the shape of the mesh from the lighting
results, you can get lighting results that would otherwise be
impossible.

Important: The mesh and the resulting lighting can be separate.

Several methods to control normals
 Several methods to control normals

Controlling Normals via modeling

Control the direction by placing vertices during modeling.

Transferring Normals
Transfer normals from a mesh with a different shape.

Manual control over Normals

Manually edit the directions of normals.

You have to put all of these methods into use, depending on the purpose.
Controlling normals via modeling
 Controlling normals via modeling

Beginner Level: Control Normals without editing them.

“If the vertices define the normals, then I

should arrange them to get what I want!”

⇒If vertex placement defines the direction of the normals, then why not just
place them to get the results we want?
⇒I think a lot of people do this unconsciously, but if you understand the
theory, you’ll get better results.
⇒A very important method for models that doesn’t use normal maps.
DEMO : Controlling normals via modeling

Controlling the
DEMO
shadow of a box.

Controlling the shadow

DEMO
on the wristband.

Controlling the shadow

DEMO
of a lock of hair.
Controlling Normals via modeling: The Theory
⇒The positions of adjacent vertices affect vertex normals.
→By placing new vertices on an existing edge, you can get a normal that’s
perpendicular to the edge.
→You can adjust the resulting lighting depending on where on that edge you
place the vertex.

⇒You can improve your lighting by using this on the edges of

clothes and the tips of hair.
⇒At ArcSys, we call these vertices “suppression” vertices.
⇒Since it’s a useless vertex in terms of silhouette, it’s a
disadvantage in terms of geometry “cost”.
→ But when it comes to the style, it’s worth it.
Transferring Normals
 Misconception that has to be cleared up.

Normal Editing ≠ Transferring Normals

Transferring Normals is only one of the

available methods.
 Transferring Normals

Intermediate Level: Transferring normals from another mesh

“Your normals are

looking good. Can I have
‘em?”
⇒By transferring the normal information from a different mesh, you will
change the lighting, based on that original shape.
⇒Not really suitable for finer control over shadows. Often used to get
smoother shadows overall.
⇒It is often used to prevent unnatural shadows in “card”-based geometry,
such as vegetation.
DEMO : Transferring Normals

Transfering normals to
DEMO
hair.
Manual Control over Normals
 Manual Control over Normals

Advanced level: Edit individual normals manually.

“It there’s a specific shadow shape I

really want, I’ll just do it myself!”

⇒ To get fine details and specific shadow shapes that can’t be obtained
with transferred normals, you need to edit the normals individually, by
hand.
⇒Mainly used for areas where you want a striking style, such as the face.
⇒It would be an unrealistic amount of work to use it throughout the whole
body. (Combine the other two methods for this)
Manual Control over Normals

DEMO Getting rid of messy

shadows quickly

DEMO GGXrd-style normal editing.

Modeling with manual normal editing in mind
Shading planes（simple Divide geometry according to these
example） areas

⇒Because normals are specified per vertex, you should model with that in mind.
→You can’t reshape a shadow if you have no vertices.
⇒You’ll have to split polygons to achieve the desired shadow shape.
⇒Some parts can’t be solved well due to the flow of topology, in these cases,
you have to use other methods.
Pros and Cons of Normal Editing
Pros of Normal Editing
⇒ No messy shadows. You can create the
shadows you want.
→The artist’s style can be incorporated. You
can make your own “illustration”.

⇒The shape of the mesh and the resulting

lighting are independent.

⇒You can escape the “curse” of topology. You’ll

have more freedom when splitting polygons.
 Cons of Normal Editing.
If you rotate a facial bone, the
normals rotate and we get a
messy shadow….
⇒ Bone deformations will rotate the normals,
which can break your shadows. A problem when DEMO
creating facial expressions.

The lines are unintentionally

⇒ When creating your inverted hull outline, you thin….

need some ingenuity.

(Can’t extrude it along the edited normals.)

The normals are

recalculated
⇒Can’t be used on subdivided surfaces.
(Because the normals are recalculated.)
→Could be a problem in non real-time rendering？¹

TN¹: Unsure about the translation on this. 映像系で問題になるかも？

Summary: Practical Normal Editing
 Summary: Practical Normal Editing

⇒ There are many methods to control normals, and you can get
better results by mastering them within your preferred 3D
application.
⇒ Although there are a lot of rules and it’s quite a bit of work,
controlling shadows is worth it.
Especially effective for real-time celshading.
⇒ Vertex Normal editing is a very powerful tool, but there are still
few examples of its use, and insufficient research.
We expect several uses and editing tools to show up in the future.
Lastly
 About different techniques

Character models are created through a series of small

innovations such as these.
However, what I showed today are only a few of many
techniques.
Anime-style character modeling has a very short history, and
there is still room for improvement.
Today, someone, somewhere, may be discovering a new
technique.

Or maybe, you’ll be the one to find it next.

 How to be a good character modeler

⇒Character modeling isn’t completed just by creating a mesh.

⇒By acquiring knowledge of other fields, such as rigging, shaders and
animation, you will be able to create better and easier-to-use models.
⇒Especially when working with a large team, it means nothing if only the
character model is good. Contributing to the quality of the entire project is a
must for a good character modeler.
⇒With a broad perspective, let’s aim to create models that improve the quality
of the entire project!
 Staff Recruitment Info
I want to work on models that look like this!
I want to animate models that look like this!
I want to make backgrounds!
I want to make effects!

Come to Arc System Works!

E I NFO
MOR https://www.arcsystemworks.jp/official/company/whatisarc/
Other GGXrd 3D Materials
 Follow me on twitter.

If you appreciate this translation and wish to donate

or something, here’s my paypal.
Other presentations will be linked here, as I finish
translating them.