4gamer - The Secrets of GGXrd's Graphics

The secrets behind
Guilty Gear Xrd -SIGN-’s real-time 3D Anime graphics
This new series, "Examining Game

Graphics" focuses on the graphics of
specific titles, and explains the
systems and technology behind them.
I have been developing "Nishikawa
Zenji's 3D Game Ecstasy" as a series,
but since the scope of coverage has
become too broad, I would like to deal
with technical explanations of specific
titles in this new series.
This 1st commemorative issue will
feature GUILTY GEAR Xrd -SIGN-, a
fighting game developed by Arc
System Works and rocking arcades
since 2014/2.
Original Article 1 / 2 .: Writer_西川善司/Zenji Nishikawa ; Translation and slides by .: @LeoBGKK on Twitter

Photographer_佐々木秀二 /Shuji Sasaki , 2014 Original rough translation by Chev on Polycount Boards
Guilty Gear in
Full 3D Graphics
Guilty Gear in full 3D Graphics
There are various theories on the origins of fighting games, but Capcom’s
Street Fighter series is probably the de facto originator of the real-time,
face-to-face, side scrolling system, where inputting lever directions plus
button presses controls the system.
Eventually, the 2D fighting game turned to 3D graphics in the early 1990s

and came to be called "3D fighting game". Virtua Fighter being the original
3D Fighting Game. What's interesting is that not all fighting games
transitioned to 3D graphics after that. As a result, the traditional "drawn"
and "pixel" games were given a new name, "2D fighting games," and
survived.
Street Fighter II © CAPCOM U.S.A
While 3D fighting games have evolved in line with the evolution of GPUs, with more realistic graphics, better
animations, and flashier effects, the evolution of 2D fighting games has been more subdued. It's no wonder, as 2D
fighting game graphics are ultimately a process of hand-drawing character animation frames one-by-one, so the
benefits of GPU evolution are not significant.
That's not to say that there haven't been attempts to revolutionize the graphics of 2D fighting games, such as the
Capcom vs. SNK series, which uses 2D graphics for characters and dynamic 3D graphics for backgrounds. In recent
years, there have also been cases such as The King of Fighters series, where the poses of the characters are
generated in 3D graphics and then traced by the drawing staff. Incidentally, this rotoscoping technique is commonly
used in recent TV animations.
In recent years, Capcom has created 3D graphics for Street Fighter IV that don’t break the pixel atmosphere of
Street Fighter II, a big hit in the early 90s. A new style of "3D graphics but 2D game system" was born.
2D fighting games have been exploring new techniques of expression by adopting 3D graphics technology in one
form or another, but what about GUILTY GEAR Xrd -SIGN- (GGXrd)? After knowing that this is a 2D fighting game, I'd
like you to take a look at the movie below. It's a digest of the gameplay scenes edited by Arc System Works, in the
1920x1080 resolution that will be available on the PS4 version. Check it out.
It may look hand-drawn, but in reality, it's 100% in 3D.

In moments like super special moves, the camera moves boldly from the side view of normal battles to show the
dynamic actions of the characters, which is truly a benefit of 3D graphics.
It’s Unreal Engine 3
GGXrd is based on Unreal Engine 3 (UE3), a game engine developed by Epic Games.
UE3 is well known as an engine that excels in realistic 3D game graphics, as seen Gears of War and Mass Effect. Of
course, game engines are basically designed to be able to handle a wide variety of game production, so from a
technical standpoint, it’s reasonable to say that it's possible to achieve anime-style visuals with UE3. However, from a
fighting gamer's point of view, the fact that Arc System Works, a company that excels at hand-drawn anime style
visuals, has created a new Guilty Gear game using UE3, which excels at so-called "Western" visuals, is hugely impactful.
UE3 has been increasingly adopted by major game studios in Japan. Above, CyberConnect2's "ASURA'S WRATH" , and Square Enix's "Chousoku
Henkei Gyrozetter"
ASURA'S WRATH © CAPCOM
Chousoku Henkei Gyrozetter © SQUARE ENIX
Why UE3? We first asked the development staff

about this, and got the following answers from Daisuke
Ishiwatari, the general director, and Takeshi Yamanaka,
the director.
Daisuke Ishiwatari— If I had to give one obvious
reason, it would be that we didn't have much time to
spare (laughs).
I'll talk about how we came to the decision to use 3D
graphics later, but in order to complete the game
using 3D graphics within the allotted production time,
we needed to use an existing game engine.
We had previously developed "Guilty Gear 2:
Overture," a real-time strategy-style game with 3D
graphics, using a from scratch in-house game engine. Daisuke Ishiwatari (General Director) has
However, we didn't have the time to improve on this been involved in the production since the
first title in the series, "Guilty Gear" for the
engine and start production. original PlayStation, and is the general
director of the series. In addition to
directing the entire project, he is also in
charge of visual production and music
composition, making him Mr. Guilty Gear.
Takeshi Yamanaka— Epic Games has announced

Unreal Engine 4 (UE4) as their next generation game
engine. We started looking for an engine right around
that time, and with the introduction of UE4, UE3 was
discounted, making it affordable for small and
medium studios like ours. This was one of the major
factors that led to our adoption.
Also, Epic Games, who had seen the prototype
version of GGXrd, gave it high marks for "bringing out
the new potential of UE3," and we were able to receive
multifaceted support from them.
GGXrd was an arcade title. Didn't they consider UE4? Takeshi Yamanaka (Director). He supervised
the storyline, wrote the script, and created
Or what about other engines, domestic or foreign? the sound effects for the game. He has also
Takuro Ieyumi, the lead programmer, answered this been involved in the creation of the
worldview for the Guilty Gear series with
point as follows. → Ishiwatari. In the past, he’s also been in
charge of the direction of the BlazBlue
series.
Takuro Ieyumi— Of course we considered it. But for

GGXrd, we knew that we would be adding and modifying
functions to the existing game engine, so we needed to
access to the source code and have a functionally mature
engine.
UE3 was the one that met these requirements. For an
engineer like myself, it had a lot of appeal.
It’s an engine that has been in use since around 2005,
and the feedback from developers who have worked on it
before us has made it functionally very stable. We don't
want the whole system to become unstable when adding
our own functions.
Also, if such a situation were to occur, it would be easier to
identify the cause if the engine side was stable. That's why it
was also important.
Takuro Ieyumi (Lead Programmer), who

Hidehiko Sakamura, art director and chief animator, and was in charge of overall program
development for GGXrd. His previous works
Junya Motomura, lead modeler and technical artist, gave the include "Battle Fantasia".
following evaluation of UE3 from the artist's perspective. →
Hidehiko Sakamura— The UE3 UDK (development kit), or the

"free version" of UE3, is available to the public, and with it, artists
like myself can create a sample FPS using our 3D models. It was
quite easy to use as a tool.
Junya Christopher Motomura— Our development team
doesn't have many programmers, so UE3 was a great tool for us
to be able to proceed with development without bothering them.
The artists can do a lot of work on their own.
In a fighting game, various adjustments are made during
development, but if the data is exchanged between the artists
and programmers each time, development efficiency would
suffer. With UE3, the artists were able to set up the game on their
own and see what they were creating, so we were able to avoid
that problem.
In the GGXrd project, we used the tools provided by UE3 for
basic content creation, but for other tasks, particularly the
creation of the core 2D fighting game, we used the in-house
Hidehiko Sakamura (Art Director and Chief
tools that Arc System Works has used in past projects. Animator) was involved in the character
Specifically, the in-house tools included a 2D fighting game animation design for not only the fighting
collision detection tool and a script tool for controlling character but also the Story Mode of GGXrd. He was
also in charge of animation for Guilty Gear 2
actions. Overture in the past.
The UE3 Tools

The game logic part of the 2D fighting game was not created anew, but
was ported to UE3 from the proven tools that had been developed
in-house.
The company's familiar tools not only have a stable track record, but also
many people who can use them within the company. Considering the
need to maintain development efficiency while adopting new
technology, we can summarize that the policy Arc System Works chose Collision setting tool. As with previous 2D fighting
for GGXrd was pretty realistic. games, the collision is a 2D rectangle.
← The shader adjustment

screen in Softimage, where
previews allow the user to
work while checking the
visuals in real time.
The shader adjustment

screen in UE3, where you
can see that the material
tree has been created to
produce the same
calculations as the shader
created in Softimage. →
As of the time of this report, the development

team for GGXrd consists of 25 full-time members
from within Arc System Works alone.
Excluding management, there are 4

programmers, designers, and 12 artists. In addition,
there are outsourced staff from Japan and
overseas, so the total number of people involved in
the project is around 100.
The idea for the game itself was conceived around

2008, and it was launched as a concrete
development project in the spring of 2011. At that
time, it was almost decided that 3D graphics
Representatives of the GGXrd development team
would be used, and a pilot movie was made for
internal presentation.
The development of the prototype version began in the latter half of 2011, and the decision to use UE3 was
made around the end of the year. Full-scale development work began in the second half of 2012, and the
actual production period was about a year and a half.
Pilot cutscene, for internal presentation.

A 2D Fighting
Game with 3D
Graphics
A 2D Fighting Game with 3D Graphics
Arc System Works has a proven track record of producing several 2D

fighting games, and are seen as craftsmen in their area.
With the exception of the aforementioned Guilty Gear 2: Overture,
the Guilty Gear series has been recognized by many gamers for its 2D
pixel art. Because of that, the decision to switch to 3D graphics was a
big one for Arc System Works.
D. Ishiwatari—— We started work on the new Guilty Gear in the
spring of 2011. Before that, we released BLAZBLUE in 2008, which we BLAZBLUE © ARC SYSTEM WORKS
positioned as the pinnacle of pixel 2D fighting games.
So what could the new GG do to impact the fans? The conclusion we came to was real-time 3D graphics
that look like pixels or cel animation.
Ishiwatari had not decided on 3D from the beginning, but had conducted his own research to see if vector
graphics could be applied to 2D fighting games.
They had been experimenting with 3D modeling of characters from the Guilty Gear series since around
2007, but felt that there were limits to what he could do with so-called photo-realistic visuals.
Around the same time, Junya Motomura began researching toon shaders (also known as cel-shaders),
which are used to create cel animation-like visuals in real time. →
A 2D Fighting Game with 3D Graphics
As for the programming side of things, Mr. Ieyumi said that since he
had experience developing the 3D fighting game Battle Fantasia he
wasn’t too worried about adopting this direction.
Dynamic visuals can be achieved by changing the camera angle. It’s
easy for even a layman to see this is the main appeal of 3D, but
Ishiwatari himself, as the general director, saw another great potential
in this method.
D. Ishiwatari—— If you use 3D graphics, you can make the game
more interesting. BATTLE FANTASIA © ARC SYSTEM WORKS
The use of 3D graphics makes it easier to create a wide variety of

facial expressions, which is difficult to achieve with pixels.
Since you can move your face independently of body movements, you can express a variety of emotions
during battle without dialogue.
Back in the days of low screen resolution, the pixelated facial expressions of characters were not as detailed
as they are today. In recent years, however, as displays have increased in resolution, there is no longer much
room for error in showing those details, so when pixel art is used, you have to raise the drawing resolution
to match the display resolution.
Arc System Works had already achieved this with the BLAZBLUE series, but with GGXrd, they wanted to go
beyond that and focus on 3D graphics that would allow for more flexible facial animation.
What are the
Specs of
Guilty Gear Xrd?
What are the Specs of Guilty Gear Xrd?
The game uses Sega's RINGEDGE 2 (*specs not disclosed) as the system board, and since its OS is 32-bit Windows
XP Embedded, the executable binary is also 32-bit.
The arcade rendering resolution is 720p at 60fps, and instead of the standard anti-aliasing technique of MSAA
(Multi-Sampled Anti-Aliasing), it uses the post-effects FXAA (Fast Approximate Anti-Aliasing).
The total number of polygons per scene is about 800,000. The breakdown is 250,000 for the two characters and
550,000 for the background, but these values are for reference only, and represent the total geometry (vertex) load on
the GPU, including visible and invisible polygons, when rendering a single scene.
A character model is about 40,000 polygons.

Wireframe (left) and final shot (right) of the in-game screen.

In GGXrd, the head is modeled separately for the battle scenes and the
cutscenes, but the number of polygons doesn’t change significantly in
either scene.
When you hear that "3D models of the head are prepared for each
character for both battle scenes and cutscenes" people who are
familiar with 3D graphics tend to imagine that there are two models,
one with many polygons and another less polygons, like LoD (Level of
Detail). However, Ishiwatari said that this was not the case, and that the
actual reason for this was to have a good looking model for each phase
of the battle.
D. Ishiwatari—— Characters in battle will have their faces drawn
small. The same goes for costumes and accessories. However, these 3D
models don't fully express the identity given to the characters when
they're in close-up. For this reason, we decided to create separate 3D
models for battle and for use when the camera is close up.
As an example, the series’ Millia is a cool-beauty female character with
long sharp eyes and a mature face. This is how she is shown in
cutscenes and close-ups, like during certain special moves.
However, if the 3D model with the "mature face" is used directly in
battle, the eyes and nose become too thin, making it hard to express
the character's personality and expression.
So Arc System Works decided to change the look of the models for
battle and for close-ups in order to emphasize the details of the
characters' expressions and the iconic parts of their costumes and
accessories. For example, in the case of Milia, the 3D model for the
battle scene was intentionally changed to have larger eyes.
The difference between the model cutscenes (left) and for battle
scenes (right). Especially noticeable in the size of the eyes.
The main character, Sol, has about 460 bones. The character with the most bones has about 600. However, the bone
structure, including the limbs, is basically the same.
The number of bones in the head is about 170, of which about 70 are allocated for facial expressions. This is a lot
more than the average 3D game character, and that is why they are so particular about facial expressions.
Bones visualized. Body (left) and head (right)

The rendering method is the common Forward Rendering, not the recently
popularized Deferred Rendering, which is an effective method when dealing with a
large number of light sources, but was probably not necessary for the anime-like "2D
style".
However, Z-buffer rendering for early culling (*Z-Prepass, i.e. depth value advance) is
used. This means that geometry rendering is done at least twice for all 3D models. In
addition, another geometry rendering is done for the characters in order to add the
anime-style contour lines. This will be explained later.
The total texture capacity per scene is less than 160MB, and the total number of
shader programs per scene is about 60-70 vertex shaders and 70-80 pixel shaders.
All of the shaders were created within the UE3 toolset. Motomura's team was in
charge of the character shaders, and the background and effects shaders were
handled by each of that team’s members. It seems that there was almost no need for
the programming team to develop shaders individually, and this may be the effect of
the adoption of UE3.
Light and
Shadow
A combination of standard toon shaders
and finely tuned custom techniques.
Light and Shadow
Let us have a look at the graphics technology in GGXrd, one by one, starting with the
lighting system.
The main point of the system here is, put simply, the complete lack of physical
accuracy.
This is related to the fact that the “ultimate goal” of GGXrd’s visuals is to emulate
celluloid animation, and the development team aimed for something that was “more
artistically correct, than physically correct”.
Looking at the game in detail, there is only one global light source, a parallel light
source equivalent to sunlight. This light source is used to cast dynamic shadows from
characters onto the ground.
Conversely, other shadows, such as those cast by static background in objects, are
baked into vertex colors and textures.
Light and Shadow
Final render
Light and Shadow
Tiling Texture Additional Details Shadow Texture Manhole Material Manhole shadow
Light and Shadow
Shader editing screen. The five elements shown before are combined for the final rendering result.
Light and Shadow
The shadows on the characters themselves are created by the dedicated light
sources that the characters themselves walk around with. It's as if "invisible to the
player's eyes, each character's own lighting staff" is following the character around.
D. Ishiwatari— The goal was to create a cel-shaded visual, which means that the
emphasis was on visual appearance.
If you light a scene with a set light source, it may look realistic in certain aspects.
But when a character is placed in a certain position, for example, a shadow may fall
on the face and make it look completely dark, or the shadow may disappear and
make the character look flat.
This difference in appearance should be avoided in 2D fighting games, where
players are always fighting on equal terms.
J. Motomura— In terms of implementation, we have light sources in the lighting

shader for each character. It might be more accurate to say that each character has
its own vector parameters as light sources.
Light and Shadow
Texture + Normal Shading (Left) and Cel Shading (Right)

Light and Shadow
Scene with environmental light direction (left) and with individual character lights (right)
Light and Shadow
The position and angle of this individual light source is adjusted per animation.
For cutscenes, the position and direction of the light source is adjusted
frame-by-frame to make them look better.
The shading that results from the lighting is basically a very common two-toned
toon shader mechanism, where there is “light” if it exceeds a certain threshold, and
"dark" if it falls below. For example, if the result of lighting is a value between 0 and
255, a value above 128 is "light", and a value below 127 is "shade".
For the sake of convenience, we use "shadow" to refer to cast shadows, and "shade"
to refer to areas that become darker as a result of lighting effects.
However, if the resulting values are close to the threshold, the light and shade may
easily be reversed by a slight movement of the character or camera, or the areas may
be broken up into small pieces and look messy, which is not a good look. →
Light and Shadow
An example where the shadow shapes are broken up (left) and an example with all the corrections applied. (right)
Light and Shadow
This is why GGXrd tries to adjust the

appearance of the 3D models by incorporating
some innovations to make them look more like
cel animation.
One is to add a shade-bias parameter to the
Red channel of the vertex color, on each
polygon of the 3D model.
This parameter is set to be stronger in hollow
areas and areas that are occluded. In the case
of the main character, Sol, this is used for the
self-shadow-like shadows under the chin and
around the neck, and this weight parameter
contributes greatly to the shadows.
Displaying the “shade-bias” parameter. Ie.:
the Red channel of vertex color.
Light and Shadow
Difference between no shadow editing (left) and with shadows edited via vertex color (right)
Light and Shadow
J. Motomura— The "shading-bias parameter" ends up being something like an ambient occlusion
parameter, but it's set by hand based on the artist's sense.
We have a secondary "shade-bias parameter" also set in the Green channel of our lighting control
texture, which we call the ILM texture. When this parameter is set to its maximum value, it creates a
baked shadow. For example, we use this method to shade areas that are always heavily shaded,
such as the neck area below the chin.
Ky's ilm texture (left) and its Green channel (center). On the right is a zoom of “skirt” area of the Green channel.
Light and Shadow
Ky before shade editing (left) and after (right).

Light and Shadow
Comparison where only light is applied without controlling the "shade-bias” parameter (left) and after shade is controlled. (right)
Light and Shadow
Another way to achieve the cell animation-like shadows is to adjust

vertex normals. In few words words, a vertex normal, is the orientation
of the vertices.
Lighting calculation in 3D graphics requires three parameters: the

direction of the light source (light vector), the direction of the line of
sight (view vector), and the direction of the surface (normal vector). In
other words, the appearance of shadows can be adjusted by
changing the direction of the surface.
In the case of GGXrd, the character models are about 40,000 polygons,
as mentioned above. So if you don't do anything, you'll end up with
complex shading that matches those 40,000 polygons, but what you
want is rough shading that looks like cel animation. This is where the
idea of "maintaining the 3D model shape of 40,000 polygons, and
simplifying only the shadows resulting from lighting” comes from.
Face orientation is represented by a normal vector given to each vertex, which the team adjusted to produce
simplified shading.
By default, UE3 always recalculates the vertex normals when importing rigged meshes (meshes that support bone
deformation for characters). As a result, the normals edited in Softimage, a 3D graphics production tool, couldn’t be
reproduced in-game. →
Light and Shadow
So, programmer Ieyumi modified the engine's mesh import function and so that the edited normals could be used
directly in the game. “Without this, it would have been impossible to control shading by editing normals" recalls
Motomura.
Now that UE3 supported adjusted normals, there are several options for to go about doing it. In the past, Studio Ghibli
used to “flatten” the CG elements of Howl's Moving Castle by finding the average the values of adjacent normal vectors.
What method did this team choose?
J. Motomura— For GGXrd, the appearance of the character was the most important thing, so we paid close attention
to the shading on the face. For the face and head, the artists manually edited and adjusted the normals.
For example, if the normals around the cheeks are aligned closer to the normals around the temples, when the
temples are dark, the shadows around the cheeks will be dark as well. The 3D model itself is modeled as a curved
surface, but by aligning the normals, we can give the high-polygon model a rough shade like a low-polygon model.
The process of editing normals.

Light and Shadow
The upper row is before editing normals, the lower one after editing.
Light and Shadow
J. Motomura— In other cases, for clothing, head hair, etc., I use Gator to transfer the normal distribution
of a simpler shape model so that a simpler shading shows on the complex shape model.
Sol's pants (or, legs) have a complex, uneven shape, but the shading is still simplified. I prepared a
cylindrical model of about the same size, and transferred its normals to the pants model.
As Motomura says, this part is not done by hand, but is generated semi-automatically by using the 3D
model attribute transfer function of the 3DCG software called Gator. The development team used Softimage
2013.
Example before (left) and after (right) using Gator for normal The target mesh used by Gator as a source.
transfer.
Light and Shadow
The hair on the heads of the characters is quite complex, but the shading also looks correct from a cel animation
perspective. If the shading were done normally, it would have been broken up into several uneven chunks, but by
combining normal editing with normal transfer of simple models, and by making special adjustments during the
modeling stage, they were able to achieve a good result.
Before and after editing vertex normals.
Before and After editing vertex normals, and the shape used by Gator.
Light and
Shadow
Control over specular reflections and
pseudo sub-surface scattering techniques.
Light and Shadow
In cel-shaded 2D graphics, specular highlights are not as commonly seen as they could be. But in GGXrd, specular
highlights are added to the lighting as a result of techniques based on the language of highlights in hand-drawn
illustrations.
I-no’s ILM texture and its Blue and Red channels. With specular disabled, enabled, and seen from a different angle.
Light and Shadow
So what is the “language”?
It's hard to describe in a few words, but to give some examples, it's like making sure that highlights don't stick to each
other when they are closeby, but made of different materials, like skin and clothes, or making sure that highlights follow
uneven or material boundaries even though they are physically incorrect.
(1) The highlights on the chest, hair, lower edge of the headband, cheeks, and lips are mostly fixed, but they change in thickness depending on the angle and disappear at extreme angles.
(2) For Potemkin, the texture is set to highlight the muscles even on top of clothes.
(3) The metallic sheen changes with the angle, as it would look strange if it were fixed.
(4) The angle of the camera and the light makes the highlights always appear in the best position, which makes the image more convincing.
Light and Shadow
Specular highlighting is controlled by the "highlightability" parameter, which is stored in the Blue channel
of the ilm texture. This parameter adjusts the intensity of the specular reflections, so the maximum value
always results in baked-in highlights, and conversely, the smaller the value, the more likely the highlights
will fade.
Meanwhile, the Red channel of the texture for lighting control is the "specular highlight intensity"
parameter, and larger values are set for areas of metal and slippery materials.
However, even with all these innovations, the development team was still not convinced. What were they
missing to achieve their goal of a "perfect cel animation-like” flavor? →
An ILM Texture and it’s several channels displayed. Alpha channel, Red Channel, Green Channel and Blue Channel.
Light and Shadow
J. Motomura— Color. It's the color.
Light and dark are created as a result of the toon shader, but I felt that gave a
somewhat boring look. A simple cel shader process that simply multiplies the
shaded areas by a single shade of color would lack the credibility and richness of
the material.
But, in TV animation production, there are artists who specialize in color

design, and they set the colors individually in detail, saying "this part of the
character is this color when the sun shines on it, and this color when the
character is in the shade”. This is where I thought the difference might lie.
Of course, that’s not possible in this case. But as a result of his research for
systematic implementation, he came to the conclusion that the color designers
of TV animation instinctively decide the colors to be set by examining the
ambient light color of the scene and the light transmittance of the material to
be expressed. Motomura said that when he tried to implement the system
based on this theory, the results were quite close to the ideal, and he decided to
include it in the final specs.
The actual system is not too complicated. First you prepare a “Shadow Tint”
texture that corresponds to the base texture applied to the 3D model. The team
calls this texture the "SSS texture" (SSS: SubSurface Scattering) for convenience. Junya C. Motomura (Lead Modeler and
Technical Artist) is in charge of overall
If the lighting results in a shade, then the shader multiplies the value of the character model creation and bone and rig
Base Texture by the value of the SSS Texture to obtain the pixel color. If the design for GGXrd. His past works include
lighting results in a lit area, the SSS texture value is ignored, so only the light the BlazBlue series, the fighting game
source color is affected. version of the Persona 4 series, and Guilty
Gear 2: Overture.
The development team was satisfied with the results of this separation
process, which brought the colors closer to the cell animation style.
Light and Shadow
For example, the shade on the skin tone of the character will have a slight reddish tint, and the shade on
the clothes will retain the saturation of the color of the clothes. In other words, SSS textures are composed of
such "red tones" and "colors with remaining clothing saturation.
Comparison between without (left) and with (right) the SSS texture.
Light and Shadow
J. Motomura— SSS textures do not simulate subsurface scattering, so the name may not be strictly correct (laughs).
If I had to give a supplementary explanation, I would say that SSS textures simply show "how much light is
transmitted through a material. Shadows on a thin sheet of paper are lighter in color, right? That's the kind of image
you get with this texture.
Color adjustments via ambient light and light source.
TN.: This image in particular is curious, as I don’t think Guilty Gear Xrd -SIGN- made use of these color adjustments.
They would later be visibly put into use in the following title, Guilty Gear Xrd -REVELATOR-.
The Secrets
Behind the
Linework
The inverted hull method.
The Secrets Behind the Linework
One of the most important elements of GGXrd's anime-style visuals are the outlines. Two approaches are combined to
create these lines. The inverted hull method is used for the most basic line drawing of the 3D models.
Normally, when a 3D model is drawn on the GPU, polygons on the back side of the model are discarded as "invisible"
and are not drawn. This mechanism is called "backface culling," which is based on the idea that the polygons on the
back of a front-facing character model are not visible from the viewpoint anyway, so they are not drawn.
Here, backface culling is combined with flipping the faces of the polygons.
Without and with outlines.

To explain the process, the first step is to generate the hull, a process which consists of slightly inflating and inverting
the 3D model. This results in a pitch-black silhouette of the 3D model being drawn, which will be saved for now.
The second step is to render the 3D model in its original size using a standard process.
Lastly, the pitch-black silhouette and the normal rendering result are combined. Most of the silhouette is overwritten
by the regular render, but since it’s a slightly inflated 3D model, only the outline remains as a result.
GGXrd's rendering pipeline uses the Z-buffer first, so the contour lines are nearly perfect at the first stage. So the final
compositing phase may seem unnecessary, but the concept is as follows.
The rendered inverted hull, normally rendered character model and the final image. There's another secret to this, but we'll get to that later.
In fact, this is a classic technique that has been used since before the rise of programmable shader technology, but Arc
System Works uses vertex shaders to implement a unique extension of this classic technique.
Contour generation using the inverted hull method was also used in 3D game graphics before programmable shaders. This
slide is from "XIII" (Ubisoft, 2004), which utilizes the technique. The original was for PlayStation 2, and of course did not utilize
programmable shaders.
Their original additions include controls to prevent lines from becoming too thin or too thick regardless of camera
zoom or character perspective, as well as controls to increase or decrease the thickness of lines in curved and
straight areas to reproduce the feel of hand-drawn drawings. The lines in GGXrd look as if they were drawn with a
real pen, and are the result of this kind of vertex shader control.
J. Motomura— We adopted the inverted hull method because we felt it had the advantage of
allowing the artist to control the width of the lines.
In the 3D model, we have a "line thickness control value" in the vertex color, which allows the
artist to freely control the width of line drawing. This makes it possible to create styles such as
those seen in hand-drawn animation, where the cheeks are thicker and the chin becomes thinner.
Result of adjusting the thickness of the outline. Adjusted line thickness (Displaying the ALPHA Channel of Vertex Color)
Without (left) and with (right) thickness adjustment. Note the nose, cheeks and chin area.
Here you can see the changes from the shoulder to arm area. Without (left) and With (right)
According to Motomura, the breakdown of the use of vertex color in GGXrd is as follows.
RED: Offset to the shading threshold. 1 is the standard, and the darker the shadowy area, the more
likely it is to be shaded. 0 means it will always be shaded.
GREEN: Determines how much the contour line expands compared to distance from the camera.
BLUE: Z-offset value of the contour line.
ALPHA: Thickness factor of the contour line. 0.5 is the standard value, 1 is the maximum thickness,
and 0 is no contour line.
Green and Alpha are the parameters that control the thickness of the contour line; Blue determines how much the
expansion is shifted in the depth direction (Z direction) relative to the viewpoint. The larger this value is set, the easier
it is for the inflated model to be buried in the neighboring surface, resulting in the occlusion of contour lines. Motomura
says this parameter was added to prevent unwelcome wrinkle-like lines from appearing in the hair or under the nose.
Geometry (left), without (center) and with (right) Z-offsetting the outline.
D. Ishiwatari— The reason we didn’t go for a post-process approach was because we thought it would be difficult
to control the thickness of the lines as we did.
With this method, we can adjust how the lines will appear on the final platform, from the 3D model creation stage,
on the artist's side, so both the model and outlines are created simultaneously.
The post-processing that Ishiwatari refers to is where lines are drawn on the rendering result, using pixel shaders.
Specifically, the contour pixels are determined by detecting the depth difference in the rendering result, or by detecting
the difference in the inner product value of the pixel-to-pixel line of sight (view vector) and the direction of the surface
(normal vector). This method is often employed when the geometry load is considered to be too high for the inverted
hull method, and was recently used by GRAVITY RUSH exclusively for the backgrounds.
GRAVITY RUSH (C) 2012 Sony Computer Entertainment Inc.

The character models were drawn using the back method, and the background was drawn using post-effects.
The Secrets
Behind the
Linework
The newly developed “Motomura-Line”.
In GGXrd, they added another method to show contour lines, such as muscle ridges,
grooves, seams on clothing and accessories.
J. Motomura— There are some areas, such as grooves in the 3D structure, where it is
impossible to draw outlines using the inverted hull method. And if we use normal
texture mapping, jaggies will appear when the camera is zoomed in, and the
difference from the clean contours of the back method will be noticeable. Therefore,
we went back to basics: "what kind of situation would cause jaggies in texture
mapping?" to understand how to create clean line segments that do not depend
on the texture resolution.
The result was a unique line-drawing technique, called the Motomura-Style Line by
the development team. →
The inverted hull outline alone cannot be used to draw a line in any part of the figure. Instead those lines have to be drawn with texture maps.
Results of drawing a freehand line (left), you can see the pronounced jagged lines. In contrast(right), the “Motomura lines” appear sharp even at the same texture resolution
Jagged lines happen in texture mapping when a particular texel (a pixel that
constitutes a texture) is drawn as a single texel on a polygon surface. On the
other hand, when there are adjacent texels, the jaggies are less apparent
because the shape of the square texel virtually disappears.
However, even if they are adjacent to each other, if they are diagonally above
or below each other, it’s basically the same problem as having a single texel,
and jagged edges will appear. In other words, if the pixels are lined up
horizontally or vertically, you may get some blur, but the jagged lines can be A closer look at thelines (with wireframe)
avoided.
Realizing this, Motomura created a texture consisting of only vertical and

horizontal aligned pixels, with solid lines given as contour lines. He then
designed a UV map (i.e., data that shows where each polygon on the 3D model
corresponds to on the texture map) with distortions and bends so that when
applied to the 3D model, diagonal lines and curves would be mapped where
desired.
”When I tried texture mapping using this method, I was amazed. Even with a
texture of not so high resolution, I was able to obtain beautiful and smooth line
drawings.”
The actual texture used for the Motomura line, is stored solely within the alpha
channel of the ILM texture, since only the black and white information is needed.
Example of the UV Map for the line technique.

Example of a texture for a Motomura-style line. Here, vertical lines are drawn. (Top Left)
UV unwrap of a Motomura-style line. (Top Center)
The result and an example of the mesh structure. Note that the topology has been devised according
to the line to be drawn.(Bottom Left)
An enlarged view of the result of applying the Honmura line. Note the absence of jaggies. (Top Right)
For reference, here is an example of a line texture. You can see that the jaggies are especially
noticeable in the final image. The mesh itself is also divided in a way that’s completely unrelated todo
with the lines to be drawn (bottom left).
In the example of the Motomura line, you can see its strange
texture looks like a city plan, with only orthogonal line
segments, but for example, a texture framed by a square line
would be an outline applied to a muscle ridge. The muscle
ridges have an oval hemispherical shape, but in this method,
that oval is distorted to fill a square shape.
As the elliptical becomes a square, the area inside it will be

stretched and distorted considerably. Therefore, if there are
any characters or patterns inside the square, they will naturally
be distorted as well. However, since this texture is only for
adding contour lines, it does not include such symbols or
patterns.
Naturally, texture mapping is done based on the distorted and

bent UVs, and the texture is affected by scaling and rotation
when the character or viewpoint moves. When the character
and viewpoint move, the texture is also affected by scaling and
rotation, which means that the lines drawn will also be affected
by these changes, resulting in curves and slanted lines. But
since bilinear filtering is applied to the texture mapping, such
curves and slanted lines are slightly blurred, which doubles as Without (left) and with (right) bilinear filtering.
an anti-aliasing effect.
By the way, the line segments drawn in the Motomura method also have a thickness variation. However, the texture
itself does not have such a subtle line strength, and is basically uniform. So how does the final line have variation?
J. Motomura— The thickness of the strokes in the Motomura line is controlled by the design of the UV map. If you
want to create a thicker line, you can design the UV map so that the polygon surface is more widely allocated to the
texels that represent the lines in the texture.
An example of setting the thickness of the stroked line. The left image is without any settings, and the center image shows how the overlap between the
UVs and the line is widened or thinned. As shown on the right, you can also create styles with natural breaks in the lines.
I have a feeling that depending on how you do it, these lines can be applied to a wide range of game graphics other
than anime.
Deformed Geo,
Swapped Geo
Deformed Geo, Swapped Geo
The reason why GGXrd's graphics look like celluloid animation is not only because of the use of toon shaders. The
unique innovations in other areas of the game also play a major role.
For the battle stages, It's easy to see from our point of view that the various buildings and background objects that
line the stage are represented by 3D models rather than "single pictures".
But when the two fighting characters are moved to the left or right, the near view moves at one speed while the
distant view moves much slower, a familiar visual in 2D fighting game graphics, but in fact there is an incredible
secret hidden here.
Ky and Sol fighting each other in Illyria Castle Town. You can see a hotel in the
foreground and a street stretching into the distance on the left.
Yes, the background 3D models of the near scenes and buildings in

the scene are modeled with a very strong perspective, as if they were
squeezing into the depths of the scene.
On the other hand, the 3D models of the distant scenes, although not
in exact perspective, are much smaller than the actual scale. There is
a reason for this modeling and arrangement.
Were they modeled in full scale without any perspective, we would

not be able to showcase the depth of the scene.
The distant objects are scaled down for the same reason, but they are
made smaller and placed not too far away so that they can be moved
left and right as the characters fighting in the battle stage move left
and right.
In real life, if you wanted a building to appear this small in the

composition, you would have to place it at a considerable distance. As
you've probably experienced in the real world, the distant view doesn't
move much when you move a few meters to the left or right. This
would mean that the background would not move as it does in 2D
fighting games, so instead, small 3D models were placed in close
proximity to have them move as well. The Illyria Castle Town stage, a mesh of buildings is displayed by itself,
and the camera is moved. The graphics are quite surprising.
Here’s an example of the movement in a battle stage.

After reading the above explanation, some of you may be wondering what is going on in the game world
behind the camera. Ishiwatari explains:
D. Ishiwatari— In the K.O. scene, the camera goes around and faces the front side of the screen. So we set
up background objects in the foreground as well, as far as the camera can see.
Also, in some cutscenes, we use 2D drawn backgrounds. When we want to move the 2D background
dynamically according to the camera work, we use a TV animation-like theory for the 2D background itself
(laughs).
One additional note on the backgrounds: the mob characters close to the battle stage are 3D models, while the
ones in the distance are animated billboards in a PaRappa the Rapper style.
As mentioned before, the main characters in

the battles are modeled with about 40,000
polygons, but they are animated in a slightly
different way than the usual 3D
graphics-based game characters.
In general 3D graphics, game characters are

posed, deformed, and animated by moving
the bones inside the model and having the
outer 3D model follow suit.
In GGXrd, most actions are implemented in

this way, but this is not the case for attack
actions where hair transforms into the shape
of a blade, for example. So in such special
cases, they deal with it by replacing some Faust's opened shirt is made with additional parts.
parts of the 3D model.
In the case of Milia, whose hair is transformable, additional parts are made, which can be replaced.
H. Sakamura— Each character's movement is controlled in a unique way to create a dynamic,

hand-animated style. Unlike most 3D game graphics, GGXrd characters' eyes, noses, and mouths on their
faces "move" a lot. In addition, parts of the body that would be impossible in a real person will expand,
contract, or shrink.
It's all about looking good on the game screen for the player. This is the motto of GGXrd’s graphics.
An example of the use in Bedman’s Instant Kill cutscene. If you create facial expressions in a conventional way so that they look safe from any direction
(left), you won’t capture same expression as the reference (center). To solve this problem, the position and angle of the eyes, nose, and mouth are
adjusted using a rig (right), and the character is effectively presented by targeting only the image as seen by the camera.
The characters are equipped with bones same as other 3D

games, but are given a high degree of freedom in how they are
moved and controlled.
For example, the eyeballs can move freely within the face, and
the mouth can deform and open and close to the extent that the
volume of the face changes considerable. The arms and legs can
grow and shrink freely. This is implemented to achieve the
"cartoony and aggressive facial visuals" that we are used to.
However this system cannot handle cartoonish expressions
such as comical "round eyeballs" or anger marks. Therefore, when
such expressions are necessary, they are handled by replacing
parts, adding parts, or using textures.
D. Ishiwatari— Some of the poses, such as reaching out arms
and legs, could not be expressed dynamically if we only adjusted
the angle of view.
If we widened the angle of view to emphasize the perspective
to make a reaching hand look bigger, the face in the back would
become too small.
Instead, we would extend the arm of the 3D model and bring
the hand closer to the camera. In the end, we didn't change the
angle of view of the camera, but rather we added the angle of Final shot and revealing the deformations to achieve it
view to the character models (laughs).
H. Sakamura— Actually, this kind of scaling and stretching of bones for each part of the character model was
not possible with the standard UE3.
T. Ieyumi— In a fighting game, it's not uncommon to emphasize parts or postures depending on the action. In a
punching action, the fists are slightly enlarged, and in a kicking action, the legs are sometimes extended. UE3 did
not support the x-, y-, and z-axis individual scaling systems for bones that are necessary for those visuals.
T. Yamanaka— UE3 is a game engine that was originally designed for photorealistic game production. It may be
that the unrealistic "freely scaling bones in three axes" wasn’t needed.
J. Motomura— In the standard state of UE3, it was possible to scale bones by applying a single parameter to all
three axes simultaneously. Using this method, we could create a “super deformed” (chibi) character, though.
“In order to create effective animations, a system that allows the bones to be scaled up and down in each of the
three axes is absolutely necessary" insisted Sakamura, the animation supervisor. The lead programmer, Ieyumi,
half-heartedly agreed and modified UE3 to implement this system. He recalled, "We were able to do this because
UE3 provided the source code.”
Initially, Ieyumi suggested to Sakamura that it might be possible to deal with the problem by creating more bones
and moving the bones themselves, but Mr. Sakamura objected. →
Normal-scale and “super-deformed” characters. Note that the deformed character shown here (right) is not the result of the basic UE3 function of
simultaneously scaling up and down three axes, as described in the text, but of the changes Arc System Works implemented into the engine.
An actual use cast of bone scaling. On the left is the final animation, and on the right, the animation without the scaling of bones.
H. Sakamura— It's true that scaling of monster

parts was implemented in some unmodified UE3
titles. I guess that was acceptable because they
were just some characters. But for us, all the
characters needed to have this level of
expression, and moreover, when we calculated the
number of bones we would need for all the limbs,
we found we’d need 4 times as many.
T. Ieyumi— I had to break when I found out about

that (laughs). Quadrupling the number of bones
would be really hard from a performance
An example of limb scaling in the viewport. The rig is set up so that the user can make
standpoint. changes by simply moving numbers up and down in a small window in SoftImage.
As a result, Ieyumi modified UE3 to support the three-axis individual scaling of bones. Before moving on to the actual
production stage, he also created a custom interface that allows the user to easily scale, expand, and contract limbs.
Thus, they were able to create animations while using this custom interface to adjust the scaling and expansion of limbs.
T. Ieyumi— We went through a lot of trouble to get it working, but it was implemented as a standard feature in UE4,
which came out afterwards. Well, I guess a lot of people other than us were requesting it (laughs).
3D Modeled
Effects
3D Modeled Effects
With a few exceptions, many of the smoke, flames and various flashing effects that accompany the special moves you
perform are also modeled in 3D.
D. Ishiwatari— I felt it would look awkward if the effects were billboarded and flimsy, so we took the liberty of modeling the
effects in 3D (laughs).
This is easier said than done, but it was skillfully implemented.
For example, the smoke effect shown below looks like a bit of fluid animation, but in fact, it was modeled frame by frame by
the artist. The smoke effect, which looks like it is rising, expanding, and bursting, was also created frame-by-frame by the artist
to look like that.
Smoke effect (left). Since it is modeled 3D data, the appearance changes when the camera is moved (right).
3D Modeled Effects
Wireframe and texturing of the model.

3D Modeled Effects
H. Sakamura— Initially, we experimented with moving

multiple sphere models connected by bones, but it didn't
turn out to be satisfactory. ...... In the end, I had to resort to
forceful techniques (laughs).
In the case of Zato, a character who manipulates

shadows, some of his moves involve melting into shadows,
so I transformed Zato's character model up to the middle
of the move and then swap it with a "shadow melting"
effect model that I prepared separately. Zato’s shadow melting model, before animation.
The smoke effects we talked about are not dynamically lit, their shadows are baked in via texture, and
have no outlines. Luminous effects such as flames and flashes only appear to glow, but do not
actually illuminate the surrounding area.
On the other hand, lighting is used for effects that are part of the character, such as the Zato’s
shadow-melting, and the company will carefully consider the use of 3D effects in the future.
3D Modeled Effects
The final model on the left, after animation, and as seen from a different
angle on the right, proving it is 3D
Creating Cel-like
Limited
Animations
Creating Cel-like Limited Animations
“Cel shading" has become quite known among gamers. Despite this, many people who see GGXrd for
the first time can't tell that the graphics are 3D-based. One of the reasons for this is that the game’s
graphics don't move smoothly.
There have been many cel shaded game graphics in the past, but since they were 3D graphics, they
moved smoothly at 30 or 60 frames per second. This was the case with Jet Set Radio (SEGA, 2000) for
the Dreamcast, and XIII (Ubisoft Entertainment, 2003), mentioned in the first part. Naturally, after all,
they are based on real-time 3D graphics, so every frame is generated when the character moves.
However, Arc System Works decided that this smoothness didn't suit the feeling of Guilty Gear.
As some of you may know, cell-based animation, such as TV anime, is made up of 8 or 12 frames per
second. Therefore, when you are watching TV anime and encounter a scene where drawn characters
and CG-based mecha live together, you may have noticed the difference in smoothness of movement
between characters moving at 8 frames per second and mecha moving at 60 frames per second(*), and
it may have felt weird. Perhaps people who are used to watching traditional animation are more
uncomfortable with the smooth movement of anime-style designs.
TN.: The author makes a mistake here. TV animation runs at approximately 24 frames per second, so a mecha in anime would never
appear to be 60 frames per second. Secondly, although he is correct that it is mostly animated at 8 or 12 frames, it is not too rare to
see more intense or important movement animated at full 24 drawings per second.
H. Sakamura— In the end, we went with a 3D graphics system that can animate at 60 frames per second, but we dared to
use a frame rate that was more like drawn animation. 15 frames per second is the basic rate for Guilty Gear’s cutscenes, which
is slightly higher than the 8-12 frames per second of cell animation.
Since the battle animation is directly related to the performance of the attacks, the number of frames to display for each pose is
specified separately for each attack. The amount of frames each pose is displayed is not fixed, but is rather something like "2F, 3F,
3F, 1F, 1F, 2F, 2F, 3F, 4F" (*Author's note: 1F≈16.67ms) for each technique.
In the animation industry, "full animation" is when all 24 frames per second are drawn and animated, as in the case of Disney(*),
while "limited animation" is when the same frames are displayed in succession and moved at 8 to 12 frames per second.
This animation style is quite different from that of normal 3D game graphics.
In general 3D graphics, character animation is based on creating "trajectories" of bones inside a character model. The trajectory is
created by using a high-degree curve function (F-curve) or based on data obtained from motion capture.
On the other hand, GGXrd's limited animation is done by moving the characters little by little and creating poses frame by frame.
In other words, it's more like stop-motion.
H. Sakamura— In the beginning, we tried using F-curves to move the characters at 60 full frames per second, and then just
reducing the frame rate. But that just looked like 3D graphics with dropped frames (laughs).
TN.: Here the author repeats a misconception about Disney and full animation. Disney often animated at 12 and rarely at 8, as well,
and you can see it in any of their feature films, depending on the scene.
Example of the same animation in full frame rate and limited.

TN.: This video was uploaded before youtube supported full 60fps motion, so the example isn’t as strong as with the final game.
In the actual production process, a storyboard is created first. In this example, the specifications for a fighting game
are set, such as for a special move: "The entire move will be composed of 60 frames" and "At the 30th frame, a punch
will be delivered, and the hitbox will appear at this timing”. The storyboard is then given to the animator, who decides
the pose of the character model for each frame.
The artists at Arc System Works are professionals in the creation of 2D fighting game action, so they must be familiar
with such character poses. “We don't rely on interpolation using F-curves and such" Sakamura says.
Storyboard for May’s Special Move.

20 frames of the previous animation, show in 2 second intervals.

What does it mean to say that the game displays at 60 frames per second even though it is a limited
animation? It means "60 frames per second as a video specification".
For example, under 60fps display settings (16.67ms per frame) if a pose is designed to take 2F time, the
same pose will be displayed for 33.33ms (= 16.67ms x 2).
Movements of parabolic trajectories that occur when characters run, jump or fly through the air are still
updated at smooth 60 fps. Also, I don't think it needs mentioning, but the timing of the player's
command input is accepted at intervals of 1/60th of a second.
D. Ishiwatari— When posing a character in a cutscene, we adjust the position of each part of the
character model and the position of the light source for that character, frame by frame, in order to
improve the appearance of the cutscene. In the case of animations where the camera goes around the
character, I also adjust the position of the nose for each frame (laughs).
The unique graphical expression of GGXrd would not have been possible without not only the
advanced technology, but also the accumulation of these skilled animators' artistic senses.
TN.: Here the author repeats a misconception about Disney and full animation. Disney often animated at 12 and rarely at 8, as well,
and you can see it in any of their feature films, depending on the scene.
Limited animation as showcased in May’s Instant Kill. No sound.

TN.: Video at half speed, 30fps. Play at double speed to see how it looks in-game.
Making the
Fighting Feel 2D
Setting up collision and mitigating 3D
Making the Fighting Feel 2D
The various efforts made to make the game look like a 2D fighting
game while using 3D graphics paid off, and they were able to reach
a level where the screenshots could be judged as "2D graphics" by
the untrained eye.
However, the development team was still not satisfied and made
further improvements to make the gameplay feel more like a 2D
fighting game. Yamanaka looks back on this process:
T. Yamanaka— In GGXrd, even though the graphics are 3D-based,

This tool is called “Collision Setting Tool” at Arc System
the gameplay is a 2D fighting game, so we didn't want to use Works. Collisions are created as rectangles, same as
2D games.
spheres, cubes or capsules like in a typical 3D game.
The collision detection area, specifically the hurtbox (the area where damage is taken when an enemy
attacks) and the hitbox (the area that inflicts the damage), were set using familiar in-house tools, as briefly
mentioned in the previous part. Each frame of the character's motion created by the limited animation
system was considered as 2D graphics, and the 2D collision detection area was set for each frame using
in-house tools.
However, there was an issue. 3D graphics are the result of cutting out and displaying the view from a viewpoint on
a rectangular screen. However, in the first place, the field of view that we are looking at is one that is reflected on the
inner wall of a sphere, whether we are aware of it or not. The problem is how to see the "spherical inner wall view" when
it is projected onto a rectangle. In a racing game, you may have noticed that when a car approaches you from behind,
into your field of vision, it may look strange and stretched out.
In fact, in GGXrd, when the characters were rendered without any changes, they tended to look a little wide at the
left and right edges of the screen, and a little thin in the center. If this were left unchecked, the aforementioned "2D
collision data" won’t correspond, depending to a character’s position on-screen..
To solve this problem, you can either adjust the collision detection according to the position of the character in the
screen, or you can mitigate the "fatness" caused by the position of the character in the screen.
The development team chose the latter approach, which was a natural choice, since in 2D fighting games the
appearance of the characters should not change just because they are positioned in the left, right, or center of the
screen. Incidentally, Street Fighter IV series, the predecessors of 2D fighting games based on 3D graphics, took this
same approach.
The actual adjustment was done by changing the method of drawing the 3D model on the 2D screen. There are two
types of 3D to 2D projection: perspective projection, which makes near objects appear larger and far objects appear
smaller, and parallel projection (orthographic projection), which makes no such difference.
This prevents the characters from getting wider or thinner depending on their screen position, resulting in an almost
constant size at all positions.
D. Ishiwatari— Strictly speaking, the same kind of hybrid projection should be applied to the vertical direction as well, but it's not as critical
as the horizontal direction, even if it's not applied, and the test players didn't feel any discomfort, so we decided not to introduce it.
A screenshot with 100% perspective projection (left) and hitbox displayed.

A screenshot with 100% parallel projection and hitbox displayed.

Final version, a combination of 30% perspective and 70% parallel projection (left) and hitbox displayed.
D. Ishiwatari— Another thing that we had to introduce was a processing system for situations where two characters overlap.
With pixel sprites, one of the characters is always above the other, but in 3D, each character has its own three-dimensional size, so
when two characters are close together, one character's protruding arm may be clipped into the other character. We had to find a way to
avoid this.
The final solution was to offset the depth (=Z) value of the attacker's character by about one meter in 3D space, so that the overlap would
not occur. In short, the attacker's character is drawn so that it always passes the depth (=Z) test, ignoring the 3D back and forth.
In the screenshot on the right, the camera is moved while Sol and Ky are facing each other (left). You can see that the characters are on the same
axis.
J. Motomura— The two characters are fighting each other, and in terms of 3D coordinate
processing, they are on the same Z axis. So the only intervention is in the drawing process.
In the case of a move where the character grabs the opponent with both arms and throws him,
the offset is adjusted so that the thrower is inside the thrower's arms. Some parts are covered up by
the flame effects (laughs).
Character rendering before depth adjustment (left) and its Z-buffer (right), where the two characters overlap in the center.
With the implementation of this system, most situations are no longer unnatural, but there are
some cases where characters that are large in the depth direction, such as the giant Potemkin
character, are in close contact with each other. However, it was judged that it would not interfere
with play, so it was left as is.
Character rendering after depth adjustment (left) and its Z-buffer (right)
Rotating the camera during a battle scene in GGXrd.

Making the
Fighting Feel 2D
Rendering the gauges and flipping
characters
The screen is designed to be played as a 2D fighting game, and this is also reflected in the rendering of various gauges such as
health. Textures are rendered onto planar polygons, which are placed in a Z position behind the fighting character, so when a
character jumps high and overlaps the gauge, the gauge will be drawn behind them. There are exceptions to this, such an attack
that launches you into the air, where the gauge is drawn in front.
J Motomura— When the camera angle changes drastically, we foresaw that the gauges would interfere with the background,
so we disabled the special processing that placed the gauges behind the character and placed them in front. This was based on
the idea that the gameplay itself is temporarily suspended during this camera angle effect, so placing the gauges in the front
would not cause any stress to the player.
Gauges displayed behind the character (left) in normal battle scenes, and in front (right) when the camera angle changes.
In the real world, when two fighters face each other in a fighting pose with
a shoulder forward, one fighter will have his chest facing the camera and
the other will have his back facing the camera. But in a 2D fighting game,
the on-screen "appearance" of the character changing depending on
whether the character is facing left or right, would make it difficult to play.
Traditionally, games simply flipped the sprites horizontally. So, can we

do the same thing in 3D for GGXrd?
Actually, if you just want to flip a character model symmetrically, all you
have to do is flip the 3D model with respect to the vertical axis (Y-axis) (i.e.,
swap the positive and negative X-coordinate values of the vertices that
make up the 3D model) and draw it. Of course, in order to flip the lighting
results as well, the aforementioned character's exclusive light source will
need to be flipped to match the 3D model. A battle screen in the 2D prequel
GUILTY GEAR XX #RELOAD © ARC SYSTEM WORKS
A potential problem is the text on some of the character costumes. For example, Sol's belt has the word "FREE" engraved
on it, but if you simply flip the 3D model of the character, the text will be mirrored.
The game avoids this problem by flipping the UV map for the area where the text is drawn so that the text is always
texture mapped in the forward direction when the 3D model is flipped.
A closer look at the characters during battle. Not the “FREE” on the belt and the decals on the gloves.
J. Motomura— For texturing, we have implemented another little trick: we

have doubled the luminance of the texture before combining it with the base
texture. Since the designed texture luminance is stored multiplied by 0.5, when
it is doubled, it becomes 1.0 and the texture is applied as designed. If the
brightness of the texture image exceeds 0.5, it will be brighter than the ground
color. By using this system, you can easily create a "three-dimensional design
with shaded text" kind of look.
The decals are attached to polygons, like a translucent sticker, over the
surface of a 3D model, similarly to a plastic model kit. The system was originally
designed to ensure sufficient resolution by separating the base texture from
text.
When rendering these decals, it is important to know how to blend them
with the base texture.
Assuming that "Dest" is the base texture and "Source" is the character
texture, the formula is as follows.
Dest x Source + Source x Dest = twice the result of the multiplication
In this case, the texture content is multiplied by a factor of two to the base
texture, so the result is "darker when less than 0.5, transparent when 0.5, and
lighter when over 0.5. This blending method was not a standard feature in UE3,
so Ieyumi added it.
If you try to do shaded text in a simpler way, you would have to texture map
the textures for the light and shadow areas of the text. This technique, however,
can be achieved with a single pass of texturing.
Sol’s texture atlas and belt. The base color is a neutral

gray.
Effective Use of
Post-Processing
Effective Use of Post-Processing
In recent years, photo-realistic 3D game

graphics have tended to include a wide variety
of post-effects. In contrast, GGXrd, with its
cel-shaded animation style as its goal, is
relatively simple.
As mentioned before, the game uses FXAA as
post-processing anti-aliasing. It uses a 64-bit
buffer with 16-bit floating-point alpha and 16-bit
floating-point RGB as render targets, and
implements a pipeline that supports HDR (High
Dynamic Range) rendering, so high-luminance
areas overflow. A light bloom post-effect is also
used.
With FXAA Without FXAA

The game with all post-processing active.

The game with post-processing disabled.

D. Ishiwatari— This kind of light bloom effect is commonly used

in recent anime, so I don't think it feels out of place. In addition, we
also use post-effects to create the Diffusion effect.
J. Motomura— Yes. Diffusion is useful for adding a softer touch

to an animated image that is divided into two parts, light and
dark. While light bloom is applied to brightness values of 1.0 and
above, diffusion is applied to lower brightness values.
May with(top) and without (bottom) diffusion.
With(top) and without (bottom) light bloom.

Ramlethal and Potemkin with their luminescent textures.

Final Word
Final Word
Looking at the final game screen, it looks so natural as 2D

graphics that most players wouldn’t even recognize its entirely
3D.
At first the development team was concerned about whether

the 3D graphics system would be able to reproduce 2D fighting
game graphics, but by the time development was complete,
they were more concerned about whether people would notice
that it was even 3D at all. They wanted to incorporate more "3D
aspects" into the game system, but for an arcade game, they
decided to prioritize ease of play and a sense of comfort as a
2D fighting game and settled on the current state.
Hopefully, the PS4 and PS3 versions, will include such extra
elements that are "only possible with 3D graphics".
The accumulated a lot of know-how in "2D visuals using 3D

graphics" through this GGXrd project. At the same time, they
feels that there are still unexplored areas of this method, and
they would like to continue to develop technology in this vein for
use in future titles.
While it is true that there are fields such as "NPR" (Non Photorealistic) visuals and "Stylized Rendering" in 3D graphics, there are very few that
specialize in the Japanese anime style. In order to develop in this field, it is necessary to have the ability and knowledge to theoretically
analyze the style of Japanese animation, and in this sense, the Arc System Works development team is well qualified. I hope they will surprise
us again by not only developing new technology, but also improving their ability to apply it, for example by applying it to games other than
2D fighting games.
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.

4gamer - The Secrets of GGXrd's Graphics

Uploaded by

Copyright:

Available Formats

4gamer - The Secrets of GGXrd's Graphics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

4gamer - The Secrets of GGXrd's Graphics

Uploaded by

Copyright:

Available Formats

The secrets behind

Guilty Gear Xrd -SIGN-’s real-time 3D Anime graphics

This new series, "Examining Game

Original Article 1 / 2 .: Writer_西川善司/Zenji Nishikawa ; Translation and slides by .: @LeoBGKK on Twitter

Eventually, the 2D ﬁghting game turned to 3D graphics in the early 1990s

It may look hand-drawn, but in reality, it's 100% in 3D.

Why UE3? We ﬁrst asked the development staff

Takeshi Yamanaka— Epic Games has announced

Takuro Ieyumi— Of course we considered it. But for

Takuro Ieyumi (Lead Programmer), who

Hidehiko Sakamura— The UE3 UDK (development kit), or the

The UE3 Tools

← The shader adjustment

The shader adjustment

As of the time of this report, the development

Excluding management, there are 4

The idea for the game itself was conceived around

Pilot cutscene, for internal presentation.

Arc System Works has a proven track record of producing several 2D

The use of 3D graphics makes it easier to create a wide variety of

A character model is about 40,000 polygons.

Wireframe (left) and ﬁnal shot (right) of the in-game screen.

Bones visualized. Body (left) and head (right)

J. Motomura— In terms of implementation, we have light sources in the lighting

Texture + Normal Shading (Left) and Cel Shading (Right)

This is why GGXrd tries to adjust the

Ky before shade editing (left) and after (right).

Another way to achieve the cell animation-like shadows is to adjust

Lighting calculation in 3D graphics requires three parameters: the

The process of editing normals.

Before and after editing vertex normals.

So what is the “language”?

J. Motomura— Color. It's the color.

But, in TV animation production, there are artists who specialize in color

Color adjustments via ambient light and light source.

Without and with outlines.

BLUE: Z-offset value of the contour line.

GRAVITY RUSH (C) 2012 Sony Computer Entertainment Inc.

Realizing this, Motomura created a texture consisting of only vertical and

Example of the UV Map for the line technique.

As the elliptical becomes a square, the area inside it will be

Naturally, texture mapping is done based on the distorted and

Yes, the background 3D models of the near scenes and buildings in

Were they modeled in full scale without any perspective, we would

In real life, if you wanted a building to appear this small in the

Here’s an example of the movement in a battle stage.

As mentioned before, the main characters in

In general 3D graphics, game characters are

In GGXrd, most actions are implemented in

H. Sakamura— Each character's movement is controlled in a unique way to create a dynamic,

The characters are equipped with bones same as other 3D

H. Sakamura— It's true that scaling of monster

T. Ieyumi— I had to break when I found out about

Wireframe and texturing of the model.

H. Sakamura— Initially, we experimented with moving

In the case of Zato, a character who manipulates

Example of the same animation in full frame rate and limited.

Storyboard for May’s Special Move.

20 frames of the previous animation, show in 2 second intervals.