"Hiring the right talent is pretty critical to getting a successful shoot," says Jake. "If you don't have somebody really good at martial arts, or someone who is even just so-so, it won't be enough."

Action Specialists, located in Valencia, California, does mostly movie stunt work. Mark Dirksei, president of Action Specialists, says that, right now, probably 10% of their work is motion capture for games, but the demand is growing as the needs of new games become more and more advanced. They've been doing game motion capture for six years.

Mark is on hand today to help with the complex combat moves. The move they're working on now isn't quite up to Mark's standards.

It involves a slash, a round kick, and a final, solid slash to the back. Mark has been playing the opponent for Hiro Koda. Hiro has worked as a stunt man and a coordinator on many movies, including Blade, Ghosts of Mars, and the forthcoming Time Machine, Blade 2, and Windtalkers.

"It looks too compressed," says Mark, referring to the final slash to his back. "I'll give you the flat of my back."

Everyone pauses a moment as they watch. Though the weapon Hiro is wielding is a wooden sword, it still looks formidable.

"Hit you?" asks Hiro.

"Well, pull the blow, but yeah."

Mark takes the blow, but even with Hiro pulling it, it looks like the hit smarts just a bit.

Throughout the session, Hiro slashes, twists, leaps, falls, and kicks often, but at the beginning and end of every move he has to return to a default stance. This is often a "T-pose," in which he stands straight, looking forward, chin up slightly, with both arms held out straight from his shoulders like the wings of a plane. This is very important to getting a clean capture. Several people edit the animation, and having these start and stop positions offers a helpful framework.

Not all of the actions need to start in the T-pose, however. As the team records a set of similar actions, a series of one-handed slashes, for example, Hiro can actually start in an action-ready state. When all the data is captured and the artists go back to Austin to start implementing them, these common states will allow them to blend between different actions smoothly.

"For example," says Joe, "standing_climbing_rope starts out in the basic stand and ends in a state where the model is hanging onto a rope. Climbing_rope begins and ends with the hanging-on-rope state. Climbing_rope_up_to_ledge begins in the hanging-on-rope state but ends in the basic stand again. This assists in blending between animations."

Brian Rausch, who directs the technical side of the process, heads the SCEA motion capture studio. The facility uses optical capturing technology to record animations, which is extremely accurate. Positioned around the perimeter of the room, hanging from the ceiling, are 16 high-resolution cameras. Another 4 cameras hang in a tighter square nearer the room's center. Brilliant red lights are attached to each camera. All are focused on the marked foot positions where Hiro stands.

What the cameras are watching are the small marble-sized reflectors attached to the Hiro's suit. Each of these is wrapped in a highly reflective white tape. "It's set to reflect at about a 3 degree angle," says Brian, "so it gives almost an exact reflection back to the cameras."

Despite all the fluorescent lighting and the flashes from our photographer's camera, none of these interferes with the data capture. "A filter in the camera is set to gather that particular color light," says Brian, indicating the bright red strobe rings around each camera. "That's how they're able to differentiate between the reflection of those lights and the fluorescent light's reflection."

Brian's monitor reveals a green grid that represents the floor of the room. At the center hangs a whorl of white dots, vaguely in the shape of a human.

"All those markers that are on the guy are reflecting data back at the cameras," explains Brian. "A tvd file is created, which is basically a 2D television data image. What I'll do over on this machine is process that, and actually turn it into three dimensional data. Then I go through and label the points. Right knee, right wrist and forearm, so on and so forth."

As the points are labeled, a framework of lines begins to form, until suddenly the whorl is now a definite human figure. Brian sets the image in motion, and suddenly the inanimate drops of reflected data spring into fluid life, leaping and kicking about the grid plain, striking at an invisible opponent with a long polearm.

"Okay, next is the five-move combo," announces Joe. I thank Brian quickly for the technical info and dash down from the control box to stand beside Jake Rodgers.

I'm not about to miss this.

1-2-3-4