By Yoshiyuki Watanabe



The original Appleseed was a big hit around the globe and opened up a new world of full CG animation. Three years later, the top-class creators behind the project reformed to give us the sequel, Appleseed Ex Machina. It is a dramatic and powerful film. In addition to harder action than the original, it also has a love story and humor that extends its appeal to women and other groups, going beyond the traditional anime audience.


As well as the action and story, the key to creating the Ex Machina world was the graphics. Running at 105 minutes, this high quality, full CG animation inspires wonder in all who see it. It was voted a prize-winner at the 2008 Siggraph, the annual international festival for the CG industry. The prize is proof that the quality of the film is recognized the world over. If you still have not seen Ex Machina, we highly recommend that you get your hands on the DVD and experience for yourself this revolutionary film.


In charge of CG creation for the Ex Machina project was Digital Frontier, Inc. We asked the production team how they used XSI in the film.

Many different CG applications were used in creating Ex Machina. XSI was mainly used for shots involving character creation, such as crowds or when depicting the main characters. (The team used different software for creating backgrounds such as buildings, for editing motion capture data, and for pre-visualization work (animatics).)


The workflow for character creation was as follows. First, they used the original rig generation tool in XSI to generate a rig. To minimize the data load, the team's policy was to construct this rig out of bones that were as simple as possible and add any complicated gimmicks later.

The most useful function in the envelope setup process was Gator. First, the team set the weight for models with low polygon shapes. They then used Gator to transfer the weight information to more complex high polygon data. Gator can cleanly transfer not only envelopes but also UV information from low models to high models. Using this method greatly reduced the data load when performing weight transfer to complex parts and when creating clothes patterns. The team was also able to effectively perform rig setup by systematically applying envelope information with integrated template characters to many of the main characters. The effective reuse of data with Gator played an important role in the pipeline.


After completing the data setup in this way, the team executed polygon reduction again. XSI's polygon reduction performs clean reduction while preserving the shape details. If the polygon count in the face and joint areas is maintained with the weight map, the data size can be minimized to enable easy movements, and data can be written without losing much detail. The team imported the written data into other tools, where they performed the capture motion flow work and pre-visualization.

When creating minor characters that required a large amount of data, they prepared a tool to efficiently create a large number of characters, using the NetView function to change the characters' hair or clothes patterns. By creating a graphical interface with NetView, it was easy for the team to generate a large number of minor characters.

After finishing the motion editing work, the team returned the animation information to the actual characters on the XSI side for rendering. For some shots, they also performed a cross-simulation calculation for clothes. In the simulation, they did not perform the calculation for the animation in the shot as is. Rather, they used Animation Mixer to calculate from a condition linking two clips: the character's default pose and the animation in the cut. With this method, wrinkles in clothes were represented more naturally in the simulation results.

The team created modeling data with a screen density that far surpassed the first Appleseed, not only in terms of the texture but also for the characters and background. In particular, significantly more detail than in the original film was added to the modeling data for the main characters. Because XSI's base architecture is so good, changes to the models can be performed even after setting the envelope or UV, enabling modifications to the quality map right up to the end of production.

After the final animation is applied to models with completed envelopes, no matter how hard the creator tries, deformation occurs in some areas. This kind of problem cannot be left unresolved in a film, so correction work is required for each cut. With XSI's construction modes, intuitive corrections can be made to animated models, while the animation is still moving , and without returning to the T pose.


CG Director Yasuhiro Otsuka explained, "To create characters that could express emotions in the story, it was essential that we chose XSI for our workflow. This is because it has a non-destructive architecture that lets us make modifications right up to the last minute, helping us achieve the best quality possible. Another important factor is that it enables fine toon control and a clean finish to the ink lines."

The facial animation for the main character, Deunan, was controlled with about 80 shapes. By making toon shading compatible with gradations, the team created finely-textured and realistic emotions with XSI. They said that the toon style was unnatural if areas other than the eyes or nose moved too much. For this reason, they had to change their way of thinking about the shapes used for controlling realistic CG characters. The team managed the data with custom parameters that targeted fine changes in shape, with elements broadly divided into eyebrows, eyes and mouths. They also incorporated realistic muscle movements for some of the male characters.



The Director, Yasushi Kawamura, said: "The XSI shape manager was useful for this kind of work. The shape manager keeps data as clusters (group information for points) instead of target models that duplicate data. This makes the data light and easy to operate. The XSI non-destructive environment is also effective for models with these kinds of shapes. All different types of modifications can be made right up to the last minute without destroying the data, which helps us a lot to improve quality."



The shape manager was also used in many other areas in addition to facial animation, such as showing the destruction of machines under attack. Further, in the scene where the stomach of the "Guardian" enemy is cut open, many vein-like tubes are severed at the same time as the body of the machine. They created the animation that divides and compresses these scenes through the combined use of the push operator and shape manager.

In the special effects for scenes where guns are fired, the flying debris was expressed using the XSI particle system. The team prepared a tool that automatically generated clouds that included the force necessary to control the emitter object. Thanks to this, they were able to smoothly create shots where guns are fired in rapid succession.

XSI's particle system and goal instances were also used i n scenes where the machines line up in battle formation and rush towards the characters. Here, the LOD system was also used, where rendering is performed by substituting low, medium and high model resolutions using clusters. The low resolution model used in LOD is data where a transparent map and two normal maps are applied to a board polygon . In the medium resolution, a normal map is applied to a model for which polygon reduction has been performed. The team frequently used a script that switches low, medium or high LOD for the selected target. This enabled the efficient switching of LOD that depend on a large amount of data. The team effectively used LOD to skillfully control the memory efficiency of rendering, which can often put a heavy load on the system.

The team often used the normal map even at the video production site, to increase rendering efficiency and improve performance. In this project, the team replaced with normal maps all the representations that had previously used bump maps, such as for the machines and threaded holes in weapons. With XSI's Ultimapper, the team was able to quickly and easily apply normal maps and depth maps from the high-resolution models to the low-resolution models. This made a significant contribution to speeding up the rendering.

The team said that the render pass function was excellent at reducing data size. In film production, a final work is created from a composite of many different elements such as shadows, colors, matte and ambient occlusion. To create one of these elements, it is undesirable from the perspective of efficiency and data size to have a workflow in which a scene is created and then saved every time. With XSI, the render pass efficiently manages multiple rendering passes in one scene file, which are then divided up for rendering. This is a very rational and reliable structure, and is one of XSI's major advantages.


Hiroyuki Goto, the Technical Director, explained: "Rendering performance, memory efficiency and data reduction are key issues for us. We are always looking for new ways of using real-time technology in the film production field. To achieve the quality that we desire, we also need to consider solutions such as customizing the graphic sequencer. XSI's open environment allows the rapid introduction of new technology and customization, and this is very attractive to us."

Digital Frontier is involved not only in full CG films such as Ex Machina, but also in a wide variety of projects, including video games and photo-realistic works using live-action film.
They occasionally have openings for people who want to use XSI and be a part of the film revolution taking place in the rapidly-evolving CG industry. If you want to take on this challenge, please check out the Digital Frontier website.

The team at the CG Production Department, Digital Frontier

From top left: Hiroyuki Goto, Atsushi Tsugaruya, Kiyoshi Ishimaru, Kazuaki Kano

From bottom left: Toyokazu Hashimoto, Yasuhiro Otsuka, Yasushi Kawamura, Hiroyuki Okada