Anda di halaman 1dari 10

CS798 Project: Emulating the Style of H R Giger

Abram Hindle
20196899
Department of Computer Science
University of Waterloo
ahindle@cs.uwaterloo.ca
April 21, 2006

Introduction

forms.
We used Wang tiles to build tilings of toruses and
planes. Wang tiles [CSHD03] have been used to create periodic and aperiodic patterns for textures and
for laying out groups of objects on a surface. Wang
tiles are shapes which tile a surface and meet at
their edges. Each edge has a colour which determines
what other edges it can connect to. Generally Wang
tiles are not rotatable and usually each edge has one
colour. We use Wang tiles for our mechanical tiles
and for our bump-maps. Some of our Wang tiles are
hexagonal instead of square.

In this project we study and emulate the artwork of


H. R. Giger with software. We hope to extend and
emulate his style by creating programs which generate textures, bump-maps and objects which are similar to those in his artwork.
The artwork of H. R. Giger consists of biomechanical landscapes filled industrial panels, circuit boards, fleshy tubes and organic monstrosities.
Gigers work can be summarized as depictions of biomechanical nightmares. Cold machinery is juxtaposed against the softer forms of the body, flesh and
bone. Gigers aesthetic maps well to video games and
industrial music and could be used to decorate video
games or music videos (we hope to use the results for
the latter).
Thus we propose to make programs that given certain parameters generate different bump-maps and
scenes which are similar to Gigers work. Our motivation is derived from our appreciation of his work
combined with our hope that we can use these results
for producing visualizations.

Bump-maps could be generated procedurally and


used for real-time rendering [RTB+ 92] in a shader
routine. Procedural textures are usually mathematically modelled using real numbers and thus can enable near infinite detail or good interpolation. This
is beneficial because one doesnt need to waste memory storing a detailed texture. Our circuit board and
organic bump-maps could be generated using procedural techniques.

Related Work
3

H. R. Giger was born in 1940 in Switzerland. He


started his art career in 1964 producing various artworks. Giger consulted and designed for the movies
Alien, Aliens and Alien 3[Gig03, AB02].
Gigers artwork consists of airbrush, oils and various forms of sculpture although his primary focus
seems to be airbrush. His images are often quite
surreal and have objects existing in space in unnatural ways. To create his nightmarish landscapes he
mixes mechanical devices with smooth blobby organic

Implementation

Our bump-maps were implemented in the OCaml


programming language with the vector graphics library Cairo. Cairo was used to render the bumpmaps to bitmap files. The mechanical tiling and
the New York Spines were implemented in Perl
and POVRAY. The Perl programs would generate
a POVRAY scene file and POVRAY ray-traced the
scene.
1

Figure 2: Ellipse Pattern

3.1

Organic Bump-map

We want to make a bump-map similar to the bumpmap used on the Harkonen base depicted in Gigers
Dune IV (see figure 1). The organic bump-map derived from this image is related to the effect of cabling across a round surface like an ellipse, sphere or
blob. These cables are supposedly meant to serve as
a skeleton or improve the structural integrity of the
objects that they cover, much like how straps secure
large objects.
In order to implement this bump-map we had to
make a repeating texture or pattern which could be
bump-mapped. The texture had to map to a surface
like a torus where the top and bottom edges meet
and the 2 side edges meet. The pattern should make
a smooth transition across each edge.
Thus our initial idea was to draw ellipses on a torus.
We felt the most interesting part of the bump-map
was when and where the ellipses intersected.
To simplify the torus constraint we simply drew
every ellipse 9 times on a grid much like a tic tac
toe board. We translated the ellipses to 9 respective
locations and then rendered the center surface. This
simulated the effect of drawing an ellipse which exited
the boundaries of the middle surface.
Ellipses size and location were chosen uniformly
randomly within a interval of acceptable values.
There was no rotation added to the ellipses. Given
a seed value we would choose how many ellipses to
place (usually less than 10) and then where to place
the ellipses. We had to allow the places of ellipses
slightly outside of the surface in order to properly

Figure 1: Giger: Dune IV

Figure 3: Application of Ellipse Pattern

Figure 4: Giger: Landscape I (Circuit Pattern)


simulate a torus. If we did not limit the size of the
ellipses we would have to increase the number ellipses
we drew, if ellipses are too big our simple method falls
apart. Example bump-maps are depicted in figure 2.
We have 2 methods of rendering the bump-map,
one is an edge rendering where the anti-aliasing of an
edge is used to produce a 2 line outline of an ellipse.
The second rendering method is a filled outline of the
ellipse.
These bump-maps are bitmap images and thus can
be used in any rendering program which supports
them. We demonstrated their use in POVRAY on
randomly placed ellipses (see figure 3).
Future work could include translating this bumpmap into a procedural bump-map shader routine for
a GPU. The input to the function could either be a
list of ellipses to draw or it could be the seed used
to choose the ellipses. The edges of the ellipses could
then be calculated per pixel by a distance metric of
the pixel to an ellipse edge. Other work could be
done in decorating the bands further with small lines
as well as decorating intersection points with cabling.

Figure 5: Giger: Landscape II (Circuit Pattern)

Figure 7: Circuit Bump-map Demonstration

3.2

Circuit

Much of Gigers work include circuit patterns. Giger


received a circuit pattern stencil in 1980[Gig03] and
3

Figure 6: Circuit Bumpmap


Examples of the circuit bump-maps are in figure 6.
An example application of the circuit bump-map is
shown in figure 7.
Extensions to this method could include:

started using it extensively in his own work. Figures


4 and 5 contain examples of these circuit patterns.
We looked at many circuit patterns and it seemed
that the common denominator of circuit boards was
solder points and planar paths between solder points.
Many circuits have straight paths and curved paths;
90 degree angled paths did not seem very common.
We thought a novel way of handling these properties with respect to Wang tiles was to use Hexagonal
Wang tiles. Hexagonal Wang tiles allow us to have
easily connectable straight lines in 6 directions and
still allow us to curve paths in increments of 60 degrees.
Our Wang tiles consist of a blank tile, a solder point
tile, 6 solder points with 1 connecting line, 6 straight
line tiles, 6 lines bent at 60 degrees and 6 tiles of lines
bent at 120 degrees, and one IC. The IC tile is meant
to be a catchall for tiles that didnt fit. These tiles
have their edges colored in 1 of 2 colors: connected
or unconnected. IC tiles are used as a fix and accept
all colors along their edges. Connected edges denote
if the edge is used in a path while unconnected edges
are for everything else.
The tiles are chosen uniformly randomly from the
set of tiles that currently fit. A tile fits if it doesnt
conflict with any other tile. If no tile fits an IC tile is
chosen. This usually happens when a tile is connected
to 3 or more paths. One problem with using Wang
tiles for circuit paths is that there is no guarantee
that paths terminate. Some paths loop into themselves and form shapes with solder points. Ideally
path should be terminated by solder points.

More tile types: Tiles which join 3 or more connections together in one tile and merge the paths
into one path instead of using the IC
Utilize various methods of randomization and
different distributions as described in [CSHD03].
Possibly emphasize certain features more like
blank space or path creation.
Sometimes the paths form their own regions; it
would be neat to fill these in with a solid color.
Decoration of the tiles. Bubbles or small details could be put inside of the solder points and
paths. This more busy noisy look might better
match the stencils used.

3.3

Industrial Tiles

We wanted to generate industrial style walls or the


mechanical walls similar to those depicted by Giger
in his paintings (see figure 8). Thus we created a
few model wall ornament tiles consisting of grooves,
pipes, vents and panels. The tiles were square and
had strict rules of which tiles were allowed to be adjacent to each other. These tiles were to be laid out
much like Wang tiles.
The tiles were defined by their object representation and their names. A configuration file contained
4

Figure 9: Example Tiles

Figure 10: Example Semi-Random Tiling

3.4

New York Spines

Figure 8: Giger: Unknown

a list of the possible tiles and the edge restrictions


each tile had. Each edge had a list of allowed symbols which could be adjacent to that edge. Because of
the complexity of the configuration file we used Perl
to parse and lay out the tiles.
Our implementation was in Perl, this eventually
caused problems with recursive tiling search. Perl
simply wasnt fast enough to properly search for a
legitimate tiling of a large toroidal surface (if this
restriction was relaxed a tiling would likely be possible) and recursive tiling search is potentially NP
(thus not appropriate for Perl). We tried many layout algorithms and found that semi-random placement algorithm worked well. The semi-random placement filled random positions in the grid with allowable tiles. This worked better than the left to right,
up to down iteration of choosing allowable tiles. It
avoided the columns previously seen by the the more
iterative approach.
One problem with tiles is they can produce boring
tilings. We didnt make a lot of tiles which could
change the direction of flow. Thus a common pattern
for the tiles was to create or choose columns of one
type of compatible tile. This lead to redundant and
boring striped patterns.
The tiles we used are shown in figure 9. The results
were unexciting, the tiles looked more interesting if
they were randomly placed than if they followed the
tile placement constraints (see figure 10).
In retrospect the tiles focussed too much on the
mechanical and thus did not mix well. The tiles were
missing the noise that persists through Gigers style.
Given more time there is a definite need for more
tiles. The tiles almost seem too minimal and uniform.
Future work would include making tiles of different
sizes, layering tiles, integrating organic elements into
the tiles and animating the tiles during a fly over.

Figure 11: New York Spines: Transparent Texture


New York Spines refer to the series of pieces by
Giger called New York which feature long repeated
patterns. Based on the figure 12, we can see repeated patterns in the shape of spines, belts, tubes
and blades, some are encased in glass capsules, most
are straight and some are curved. There is an ethereal feeling as the pipes and spines merge in and out
of each other. The image uses many edges and shapes
but has an overall organic feeling to it. Our New York
Spines are meant to emulate these properties of the
New York series.
To approach the organic feeling of the image we
decided to work on spines made of blobs. Blobs are
spheres which are attracted to each other and form
a skin between each other based on the strength of
attraction between themselves. Multiple spine shapes
were made. These spines would then be copied along
a spline path.
The implementation was a spline path generator in
Perl that made splines which either crossed both left
and right sides of the cameras frame or just one. A
random number of spline points were chosen. These
points were given an x axis interval in which they
could exist, the y and z coordinates were then chosen.
For the most part the spline points would not deviate much from the start and end points. The start
and end points were linearly interpolated to generate
the beginning and ending control points of the spline.
The resolution of the spline points and the frequency
of the repeated spines were uniformly randomly chosen per spline.
Initially the spines were rendered with a plastic texture which didnt seem to fit well but had nice Phong
highlights. The results looked structurally similar
6

Figure 12: Giger: NYC XXIII


Gigers, we have learned quite a lot.

with the New York image but lacked an ethereal feel


that the original image had. By using a transparent
texture on the spines and surrounding some spines
with a bubble this ethereal feeling was gained. Although to render an image with so much transparency
takes many CPU hours (40 hrs). Figures 11 and 14
shows a transparent New York Spine, whereas normal
and transparent spines are shown in figure 13.
Clear bubbles around spines were implemented by
dragging a blob along the spines. The blob was
thicker than the spines themselves.
Essentially the results of this program looked similar to the NYC XXIII image by Giger but did not
share the same detail (circuit board patterns and
spines) or color scheme (green, blue light) as the
NYC XXIII image. Thus future work on New York
Spines would be in increasing the number of mechanical spines, improving the textures and colors of the
spines, and adding the mixing of mechanical and organic structures.

4.1

Bio Mechanics

Gigers work is not strictly mechanical or organic; it


is usually a mix of the two. So it is important to mix
panels, vents and pipes with blobby drippy smooth
organic shapes. By merging layers of mechanical and
organic structures one can attempt to create the transitions that Giger does so well with airbrush.
Splines, blobs, spheres, ellipses and other rounded
3D shapes work well as organic objects. When repeated they take on a worm like or spine like structure
which looks analogous to organs and bones. These
blobby shapes seem to work well when they are combined together and repeated.

4.2

Textures and Bump-maps

It turns out that plain plastic textures, when used in


an organic context, give a feeling of smoothness or
perceived sliminess one encounters with worms and
4 Observations
on
Gigers snakes. Transparent textures seem to convey an ethereal quality which is inherent with Gigers airbrushed
Work
works.
Through trying to implement programs which genGiger made heavy use of stencils, such as circuit
erate images, textures and bump-maps similar to stencils, in his work. The circuit and organic bump7

Figure 14: New York Spine

maps generated here although interesting dont seem


to integrate that well. Perhaps it is an issue with the
texture itself.

4.3

Merging

Figure 15: Mixture of New York Splines and Mechanical Tiles


Many of the objects Giger paints merge or morph
into other forms. Merging is much harder to do with
mechanical objects than organic objects. The difficulty with merging is to get the transition right.
What is easy to do in 2D with an airbrush can be
very difficult in 3D (see figure 15 as a coarse example
of merging) .

Future Work

There is much future as this is project that is essentially a search for techniques and methods to emulate
Gigers work.

5.1

Re-implement Gigers Paintings

Much like the New York pipes it would be interesting


to make programs which generates images similar to
his other works. These programs could focus on one
piece and generates scenes which are copies of the
work or mutations of the work.
Figure 13: New York Spines: Plastic Texture

5.2

Humanoid Generator

In order to re-implement Gigers paintings one will


have to generate semi-humanoid character to inhabit
the generated scenes. A humanoid generator would
have general rules about building faces, arms, legs,
9

posture etc as well as to allow for mutation and corruptions of the models. The corruption or distortion
is necessary to maintain the nightmare like imagery
that Giger paints.
We attempted to implement a mutator but simply
couldnt get an acceptable model. Ideas for mutation
were to parameterize the size of features such as the
nose, cheeks, eyes and mouth. Parameters would be
the scales and locations of the features. If the head
and cheeks were blobs dimples could be randomly
created with sub blobs of negative strength. Negative strength blobs will remove material to produce a
dimpled surface. Conversely more little blobs could
be added to produce warts and pimples on the face.

5.3

Animated Fly Over and Animated


Tiles

The mechanical tiles could be animated either with


frames or functions. This would require identifying
moving parts in the tiles and then determining their
animations. This would allow for dynamic animated
fly overs and pans of tiled walls. A challenge would
be to generate the animations in real time with GL.

5.4

Textures and Occult Decorations

Circuit Bump-map: A generator which produces


a circuit board like pattern on a torus to be used
as a bump-map.
Mechanical Wang Tiles: A set of mechanical tiles
which can decorate a plane along with a tile
placer.
New York Spines: A spline generator which emulates the New York City series that Giger painted
in 1980.
Some insights into Gigers style and how to better
replicate them digitally and algorithmically.
We were the most successful with the New York
Spines, which ended up looking very similar to the
work which inspired them. If we were to do this
project all over again we would focus on re-creating
a few select Giger paintings and then elaborate automated methods from those experiences. It feels
like we went in the opposite direction, we built tools
and tried to make them Giger-like whereas we should
have started with Giger-like images and figured out
ways to generate and parameterize them. Overall we
gained a greater respect the labor put into Gigers
work.

We didnt focus on actual textures, we focused on


structure. Perhaps textures would lend themselves to References
a better emulation of the style (much like the sten- [AB02]
Sonja Altmeppen and Leslie Barany, edicils). Texture synthesis techniques might be useful
tors. H R Giger. TASCHEN GmbH, 2002.
to try. Texture synthesis might have helped extract
a generalized Giger texture. Both tiles and spines [CSHD03] Michael F. Cohen, Jonathan Shade, Stecould use more decoration. Decorations would infan Hiller, and Oliver Deussen. Wang tiles
clude objects such as skulls, religious symbols, penfor image and texture generation. ACM
tagrams, pentacles, blades and genitals. Decorations
Trans. Graph., 22(3):287294, 2003.
could also include object modifiers such as runes and
[Gig03]
H R Giger. H R Gigers Retrospective
scratches.
1964 1984. Morpheus International, 3rd
edition, 2003.

Conclusions

What was difficult about this project was it was a


search for a Giger like style which could be algorithmically generated. There were many ideas which
were explored broadly but often too shallow. It was
difficult to pin point what were the key elements of
Gigers style.
Our contributions include:

[RTB+ 92] John Rhoades, Greg Turk, Andrew Bell,


Andrei State, Ulrich Neumann, and
Amitabh Varshney. Real-time procedural
textures. In SI3D 92: Proceedings of the
1992 symposium on Interactive 3D graphics, pages 95100, New York, NY, USA,
1992. ACM Press.

Organic Cabling Bump-map: A generator which


produces intersecting ellipse patterns on a torus
which is to be used as a bump-map
10

Anda mungkin juga menyukai