Sam Berry
b1436719
Innovations Project Report
The development of a 2D animation system within
Maya
ABSTRACT
The report focuses on the study of the techniques used to combine 2D and 3D
elements in animated film. The report then goes onto describe how these methods
can be applied in the research and development of a 2D animation system that can
be implemented in the 3D environment of Maya.
INTRODUCTION
Many modern
animated films combine 2D drawn elements and 3D computer generated (CG) elements
at some point in the piece.
These films demonstrate that animators today have access to a wide selection of 2D
and 3D tools, and they use the appropriate tool for the task to create the
film quickly and efficiently, at the lowest possible cost while maintaining a high standard of
quality. The aim of this project is to examine the relationship between 2D and 3D
animation, and to research the methods that are used to successfully combine the
two in a single shot.
As a result of this research I hope to produce an innovative technique or system
that allows the user to combine both 2D and 3D animated elements together easily. I will then produce a short piece of animation that demonstrates
how the system can be used.
AREAS OF RESEARCH
The films included in my study are
Disney's "Tarzan" (1999),
"Belleville Rendez Vous" (2003),
"The Iron Giant" (1999) and
"Blood: The Last Vampire" (2000). In these films, I found that
3D computer animation was generally used to create the environments in the
piece. The use of CG
environments in animated
features can greatly improve the flexibility of the shot as far as
cinematography is concerned. In a CG environment, the camera can be placed
anywhere in the scene and complex camera moves, such as those seen in Disney’s
“Tarzan” (1999) can be employed. These camera moves were created with “Deep
Canvas”, a piece of software developed specially for the film (Figure 1). The
"Deep Canvas Process" allows the jungle environment to be created as 3D
geometry, then a digital artist is able to paint onto the geometry using a
graphics tablet, as if it were a canvas, thus creating
a sort of 3D painting. The characters can then be drawn into the environment a
frame at a time using traditional 2D animation methods. An illustrated
description of this process can be found in
Appendix 1. Shots created with this process are designed to “draw you into”1 the film
and they allow the viewer to follow Tarzan as he swings through the jungle canopy in
some impressive animated sequences. If a traditional animator had to create these
sweeping camera moves by hand, one frame at a time, it would not
only be very difficult, but it would also take a large amount of time and money
to produce.
When producing
traditionally animated films, it is
occasionally necessary to produce some characters and props using CG. Characters and props
are usually produced in this way if they are made from hard geometric
surfaces, such surfaces are suited to CG as they are easy to produce and animate. The director of “Belleville Rendez
Vous” (2003), Sylvain Chomet, explains why he chose to create vehicles in the
film using 3D computer graphics:
“When you want to
draw cars and bicycles...if you give that to...a 2D animator he is going to
become mad...because there is no life to bring to a bicycle.”2
An example of when
CG is used to create a character in a traditionally animated film can be seen in
“The Iron Giant” (1999). Although the Giant
(Figure 2) is one of the main characters in the piece, he is
still a
mechanical hard surface object. Therefore he can be produced and
animated more easily
using CG. As a
main character, it is essential that the Giant is brought to life, unlike the
cars and bicycles in “Belleville”.
This was achieved
mainly through the character’s eyes. The Giant’s emotions are
communicated
through the colour
of the glow in his eyes and the way he blinks. In the film, all the backgrounds
and scenery was painted in the traditional method by layout artists, the only CG
element is the
Giant. The
combination of CG and traditional elements in the same film in this way can
cause problems, as observed by Scott Johnston, Artistic Coordinator on "The Iron
Giant":
"Ideally, you don't want the audience to
see a big difference between a computer generated character and your traditional
characters"3
Sylvain Chomet also identifies this
problem, describing how CG generated images look "too clean"
and "synthetic", and he describes
the need to "destroy the look" of
images created in CG to make them look more like hand drawn images. This problem
is often solved by using "toon shaders", a shader designed to mimic the quality
of line and shade of a hand drawn image. All the films that I have mentioned so
far are good examples of the use of toon shaders to successfully combine CG
elements and hand drawn elements.
Although CG
techniques are advancing rapidly, they have only recently reached a standard
good enough to represent organic characters. Elements such as the movement of
skin and muscle, hair and fir and the cloth of clothing are still hard to
simulate. Pixar animation
studios, a forerunner in computer animated films,
produced five animated feature films over almost ten years before they developed
the techniques to represent human characters, as can be seen in their latest
film "The
Incredibles" (2004)
(Figure 3). This limitation of CG technology means that generally, even if
the environments are CG, the characters in the film are often hand drawn.
In addition to "Tarzan",
a good example of the combining hand drawn characters with CG environments can
be seen in the film
"Blood: The Last Vampire" (2000).
Unlike "Tarzan", this film is set in an urban environment, rather than a natural
environment. Director
Hiroyuki Kitakubo's uses the geometric shapes
of buildings, interiors and man made objects to create strong perspective in his
shot composition, and many of these shots are very striking (Figure 4).
My research into these films has led me to conclude that when a film attempts to
combine 2D and 3D animation, it is essential that this is done seamlessly. The
2D element and the 3D element must blend together so the shot is viewed as a
whole, rather than a combination of two different parts. I have identified two
methods that are used to achieve this. Firstly, a 3D computer generated image is
generated using mathematics, so the quality of the line and the shapes and
colours of the piece are technically perfect, therefore a toon shader must be used to
lose the CG feel of the image, as described above. Secondly, the
2D hand drawn element must be of a high quality to match the standard of the 3D
element. This can be achieved by using techniques such as foreshortening to
mimic the perspective in the 3D element. To produce a drawing of this quality
requires a skilled and experienced artist, with a good knowledge of the laws of
perspective. These are
the areas that I decided to research further to get a better understanding of
the two methods.
I researched a number of methods
to produce outlines on rendered objects in Maya, and produced standard toon shaders that could be rendered
using the Maya Software Renderer, and contour shaders rendered with the
Mental
Ray (Figure 5). I found that these shaders could be created very easily, and
when
using the Vector Renderer included in Maya 6.0, the same results could be
achieved using standard Lambert shaders and therefore no shader writing was
necessary. It was clear that this area did not warrant extensive research.
The creation of more complex toon shader that
produced a more painterly
result, such as the Tomcat Shader (Figure 6), required the writing of Maya plugins.
This area of heavy
programming did not appeal to me, so I decided to begin researching
possibilities other than toon shader development.
After my brief study of toon shaders, I began to focus my research towards 2D
media, namely the methods used to
create perspective drawings. I found that this area was much more interesting
and offered many more possibilities for experimentation. It was my study of
perspective drawing that caused me to re-evaluate my approach to the project.
Initially, when I thought of the combination of 2D and 3D animation, I thought
of the 2D element as hand drawn and the 3D element as computer generated. But
what if the 2D element was also created in the computer? Could the 2D methods of
perspective drawing be applied in the 3D context of Maya, and could these
drawings then be animated? After performing a number of tests in Maya I found
that this theory would be possible to realise, and I started to develop a 2D
animation system in Maya that was based on the rules of perspective drawing.
Before starting development of the animation system, I did
some research into existing 2D animations systems and how they are implemented.
I found that most computer aided 2D systems attempted to produce inbetween
frames from a number of key frame drawings supplied by the animator. Many of
these systems were not very successful, in his paper
"The Problems of Computer Assisted Animation", Edwin Catmull explains the
fundamental reason why most 2D animation systems failed:
"The principal difficulty is that the animators' drawings
are really two dimensional projections of three dimensional characters as
visualised by the animator, hence information is lost. For example, one leg
obscures another. It is this loss of information which severely limits automatic
in-betweening."4
Catmull goes on to state that to produce a series of key
frame drawings of an object, the animator must visualise the object in 3D in his
head and then convert that to a number of 2D projections on paper. The assistant
animator then creates the in-betweens based on the 3D visualisation in his own
head, and an additional wealth of information in the form of character sheets,
concept designs and 3D maquettes of the object. Catmull argues that a computer
cannot be used to create these in betweens without access to the same
information. It is for this reason that the use of such 2D systems was halted to make way for 3D animation systems. These 3D systems were only
slightly more difficult to implement, but they could hold much more information
about the animated objects.
Not all 2D animations were unsuccessful however,
Nestor Burtnyk and Marceli Wein were
recognised as
"Fathers of Computer Animation Technology"5
for their work in computer graphics and animation systems in the 1960s
and 70s. Burtnyk and Wein developed a 2D animation system which was used to
produce a number of successful short films, notably "La Faim" (1974) and "Visage" (1977). "La Faim"
(Figure 7) received great critical acclaim, and "became the first
computer-animated movie to be nominated for an Academy Award as best short"6
in 1974. The film also won the Prix du
Jury at the Cannes Film Festival and a number of other international awards.
I
believe that the reason that many other 2D computer animation systems were
unsuccessful is because they were inspired by the traditional 2D animation
process. Traditionally, a head animator would produce the key frames, then pass
them onto a junior animator to create the inbetweens. The work involved in
creating these inbetweens lent itself to an automated, computer process:
"The work of the artist's assistant seemed like the ideal demonstration vehicle
for computer animation."7
But as
Catmull observed, the key frame drawings on their own do not provide the
computer with enough information to produce accurate in-betweens. I believe that
the system that I intend to develop will be successful because it is not based
on traditional animation methods. My system will not attempt to create animation
by generating in-betweens based on given key frames, but it will generate
animation by manipulating the points that define a perspective drawing.
PRODUCTION AND
IMPLEMENTATION
To produce a 2D
animation system based on the rules of perspective drawing, these rules must be
clearly understood. To help the understanding these
rules, the method used to draw a cuboid using two-point perspective was examined. These steps were broken down into a
set of rules, for creating true two-point perspective drawings.
These rules,
hereafter referred to as the "Construction Rules", and the method used
to produce the perspective drawing can be found in Appendix 2.
These Construction Rules
apply to a static drawing, but would the same rules apply to an animated
perspective drawing? To answer this question, the
cuboid was constructed in Maya
using the the same rules (Figure 8). The cuboid was created in the top view,
converting the 3D space to a 2D plane. Each vertex of the cube and the vanishing
points were defined with locators, and as a visual reference, the lines defined
by pairs of locators were drawn with straight line curves. The ends of the
curves were constrained to the locators using clusters to ensure that the curve
always remained between the two locators that defined it. Therefore, the user
could animate the shape by moving the locators that define the vertices of the
shape around in space and the edges would follow.
To ensure that
the Construction Rules were
obeyed, I found that the user should only be able to animate some of the locators that defined the
vertices of the shape. The other vertices had to be driven procedurally to
maintain the form of the shape.
The two vanishing
points and the two points that define the front edge of the cube are the first
points that are plotted when drawing the cuboid on paper, and all other points
are defined by the position these points. Therefore, logic suggests that the
locators that define these points in Maya should be the ones that the user has
full control over (the "Control Locators"), and all
the other locators should be driven procedurally. The Construction Rules were
used as a basis to develop a number of mathematical
expressions which controlled the position of the driven locators (see
Appendix 3).
Once the behaviour of the locators that
defined the vertices of the shape was defined, it was relatively easy to create
polygonal planes to define the visible faces of the shape. The shape of each
polygonal face is defined by four locators, so the face could be represented with a
four sided polygon with each vertex constrained to one of the locators. This was
achieved in Maya by creating a cluster at each vertex of the polygon, and then
parenting each cluster to the correct locator. When the Control Locators were
animated, each face is deformed based on the movement of the locators that
defined the vertices of the face , as demonstrated in this
video.
With the system now capable of drawing
and animating simple shapes, the next task was to find a method of rendering the shapes. I
decided quite early on that I wanted the shapes to be rendered in the toon
style, with flat colours and outlines. Initially, a geometry based method for
creating the outlines was tested, ie each edge of the shape was defined by a
piece of geometry, usually a long narrow plane. This was not satisfactory,
however, as the consistency of the line was not very good. I found that a better
solution was the Maya Vector Renderer, which produced the look I intended. This also
proved a quick and easy solution to the problem, so I was able to move onto the next task.
Having
proved that the system that I had
developed could be used to construct, animate and render simple shapes such as
cuboids, I began to experiment to find out how flexible the system was.
Because the system was programmed in Maya, a 3D package, the shapes that are
created can be animated in both 2D and 3D, this produces some
interesting results. I then went on to
test the system with slightly more complex shapes. I found that these shapes
were more difficult to create,
and in
some
cases, a shape that was created in a static position looked fine, but when the
shape was animated it began to distort in unexpected ways (Figure 9). Through
experimentation I eventually found the correct method to construct complex
shapes. Firstly, the shape is drawn by hand and brought into Maya as an image
plane. This image is then used as a reference to create the locators that define
the edges of the shape. Finally, the faces could be added and the image could be
rendered (Figure 10). When constructing more complicated shapes, I learnt that it
was important to decide how
the shape would need to be animated before construction starts. Then the shape
has to be constructed bearing this in mind.
Having established that the system was
capable of producing more complex shapes, I had to create a method to
produce a large number of these shapes in a scene. The current system was quite basic,
and to construct a shape, the expressions that controlled the position of each
locator had to be
written out by hand
in each case. This would not be a feasible method for
creating a large number of more
complex shapes, so an automated method had to be developed. A MEL script was written which contained
all the procedures needed to create a complete shape (see
Appendix 5). The script included
procedures to create all the required primitives, points (locators),
lines (curves) and faces (polygons). It also included procedures to apply the
required constraints to Type One and Type Two locators. To make the script as
user friendly as possible, constraints could be applied to locators by simply
selecting the driver locators and then the target locator and running the
constraint procedure. The procedures were also made quite robust, and included
error handling code to prevent unexpected evaluations. Finally, a
Graphical User Interface (GUI) was created to
allow each procedure to be run at the push of a button (Figure 11).
Appendix 4 contains a tutorial on how to use
this interface.
With the first version of the system
fully complete, I had to prove that it would work in practice by using it to
create a piece of animation. Through discussion with my tutors,
I
decided to produce piece of abstract animation, with shapes
animated in both 2D and 3D. Following the production method
I had developed previously, I produced an image on paper showing the
group of geometric shapes I intended to create and animate (Figure 12). I planned
to create an animation that showed these shapes appearing as if they were being
drawn. Then once all the shapes had appeared, they will all fall flat and it
will be
revealed
that the shapes are 2D. The system performed quite well in the construction of
the shapes (Figure 13). I had designed the system so each shape only had a few
elements that could be animated, and the other elements were driven
procedurally, so the actual animation itself was easy to produce and it was
completed quite quickly. For the colour palette in the piece, I chose a
selection of colours that helped add more depth to the image. Colour theory
states that warm colours appear closer to the observer, while cooler colours appear to
be further away. So the shapes were shaded appropriately
depending on their depth. This colour scheme proved effective when
compared with a set of shapes with
colours assigned randomly.
FURTHER DEVELOPMENT
I believe that there is room for a lot
of further development to this system. As it stands, the system is quite basic,
I can only support two-point perspective and it only works predictably and
reliably
with shapes
with angles of 90°. However, I believe that it would be relatively simple to
expand the
system to support and any type of geometric shapes, including curved
surfaces. Such shapes can be drawn on paper using the perspective method
(Figure 14), so to create these shapes using
this system would only involve identifying and expanding the set of rules that
have been defined for simple shapes. I also believe that the rules could be
expanded to support three-point perspective drawing. I have briefly studied the
possibility of this and I have discovered that simple three-point perspective
shapes (Figure 15) can be constructed, but I have not yet experimented with
animated three-point perspective shapes.
Another possibility for further
development would be to apply the perspective method to true 3D shapes in Maya.
This would allow the user to define his/her own vanishing points for objects,
and they would not have to conform to the conventional rules of perspective. I
believe that this would produce some quite interesting results.
Another interesting concept,
which I only discovered right at the end of the project, is the possibility of
simulating light sources that affect the shapes. The use of lighting will allow
the user to render the shapes in a more realistic manner, in the same way that
real 3D shapes are rendered. It can be seen from the
test renders that it is possible to
simulate lighting in the 2D system, and with further development the 2D system
could potentially be a valid alternative to 3D, or a method of creating "2˝D"
backdrops for 3D action in a similar way to matte painting.
CONCLUSION
In conclusion, I
believe that the outcome of this project and my areas of research were very
constructive and informative. The final product demonstrates an
innovative approach to the production of 2D animation in a 3D environment, and
this combination of 2D and 3D working methods meets the requirements of my
original aim. I am pleased with the final product, which
demonstrates a good balance between the technical and the artistic, and also
demonstrates an alternative method to creating an automated 2D animation system
which avoids the problems of the 2D animation systems discussed in my research. There is also a lot of scope for further research and development of the system,
that unfortunately, I did not have the time to achieve in this project.
APPENDICES
BIBLIOGRAPHY AND CREDITS
