Hailed as the greatest CG sequence ever made for a video game, the CG for Onimusha 3 took 260 people one and a half years to create. Robot Japan gave Digit an exclusive look at how it was done.
As opening sequences to video games go, the Robot Japan- created CG movie differs from the pack in two ways. The first is the sheer amount of energy, time, and effort that was poured into its genesis – the audio alone required the work of some 160 people, including the Warsaw Philharmonic Orchestra.
The second is its sit-up-and-stare, drop-dead beauty. Working as the scene-setting device for the Capcom game Onimusha 3 – which was recently released for the Sony PlayStation 2 in the UK – it’s being received by the industry as quite simply the best CG sequence ever created for a home console. Period.
The sequence is an action-packed fusion of sword-fighting, magic, fantasy creatures, epic battlescapes, and personal quests – and its sprawling five-minute length (long, for a video game) was produced by Mr Kurasawa of Robot Japan.
The five-minute piece opens with a view of a mighty, insect-like army marching on a palace, complete with flying monsters and mammoth tanks. The hero battles his way through an army of evil creatures before facing the head bad guy on the open plain. It’s gripping stuff.
Kurasawa-san had previously worked on the CG for the previous two instalments; but this time, the Capcom producer of the Onimusha game, Mr Inafune, wanted Robot Japan to create a CG movie that would appeal to a broader, worldwide audience.
It proved a good choice. Robot was established in 1986, and creates content for TV spots, films, broadcast pieces, interactive media, and CG character and animation content. The staff includes a mix of producers, directors, artists, and programmers, says Robot – and it describes itself as very productive and free spirited.
And that free-spirited nature meant the CG sequence was no simple, virtual construct from start to finish; Robot’s Kurasawa-san proved to be more ambitious than that. The finished movie pulled in famous Hong Kong action-flick director Donnie Yen, FX and filmmaking expert Mr Yamazaki, full-on motion-captured fight scenes, a miniature model of the entire set, and even a trip to Mount Fuji. And that’s even before it even saw one polygon placed in 3DS Max.
With the storyboard roughed out, Robot says the production took around six months from developing the plot to initial motion-capture, then a further year was needed to create the CG, building miniatures, and perform additional motion capture. Over 100 people were involved in the CG creation – with the aim, according to Robot, to create a CG sequence that was as real as possible.
The trick Robot used to create its look was down to rendering and photography. Instead of creating the movie from scratch in a 3D package, and painting in some suitable textures, Robot built the entire set in miniature, complete with highly-detailed miniature painting work and a slavish attention to lighting. The scenes were later shot using motion-controlled cameras that were then mapped onto 3D scenes, and the textures and lighting data from the real-world, miniature scenes stripped out and baked onto the 3D models.
The potential problem, according to Robot, was that many of the team had never used the likes of mo-cap and miniatures before in the creation of CG animation. So, based on Yamazaki-san’s set design, part of the team split from the main group to start working up the main character models in Discreet 3DS Max, while an art team worked on blueprinting and creating the set ready for motion capture.
Real, not digital, 3D
Unusually for a video game CG shot, a huge amount of work went into the creation of real-world elements that would form the blueprint for the digital CG work. Miniature sets were created of all the key locations on a 50:1 scale, then painted in fine detail, and digitally photographed. This captured the textures and – thanks to the use of a HDRI set-up involving a chrome ball, tripod, and digital camera – the skydome lighting detail, such as would reflect in shiny objects in the final CG sequence, could be captured.
Alongside the model work, a mo-cap stage was created using the Vicon8 camera system. The stage measured 12-x-8-x-5.5 metres, and deployed 24 cameras to capture the data. Over 150 action shots were directed by famous Hong Kong director Donnie Yen.
Once done, the team turned to motion capturing, and started to create rough 3D characters in Discreet 3DS Max (for characters with more than two legs) and Character Studio (for bipeds). The lead character – Samanousuke – is played by Japanese star Takeshi Kaneshiro, and his face was digitally captured and used as a texture for the 3D model.
Additionally, the team used Shag:hair to create Samanousuke’s 3D hair, one strand at a time. Here, though, the team ran into difficulties. The strands of hair were making the data for the model too heavy, so the team split it into body, hair, face, and shadow objects for separate rendering. Everything was later composited together in Adobe After Effects.
A digital animatic – an animated storyboard featuring plain, block characters with minimal texture and lighting data – of the sequence was created using the rough 3D models, and the camera paths were decided upon. Using the animatic as a guide, the team set to work motion-capturing the main action and filming the miniature set with motion-control cameras so they could capture the textures and lighting of the real-world for extraction and placement in the CG scenes.
The action scenes, all featuring frenetic sword battles and bodies flying through the air, were directed on a mo-cap stage by Donnie Yen. The team shot over 150 different action shots for the sequence – and Robot Japan says the camera work was the most complicated process of all, with the team having to match camera data from the action mo-cap sequences, the shots from the miniature sets, and any CG-only camera shots that were needed when it proved impossible to capture a shot in the real-world.
Character animation additionally proved a challenge. Two-legged characters, such as the hero and most enemy characters, were based on the mo-cap data, with the team using as much mo-cap data as possible to ensure a more realistic result and to keep the action dramatic.
However, the data had to be tweaked extensively, says Robot, for example to avoid putting actors into dangerous positions – we are talking ferocious sword fights here, so characters had to be digitally moved closer together, and the impact of weapons altered slightly. Non-biped characters, such as the giant, insect-like tanks seen at the opening of the sequence, were animated by hand.
Finally, as part of the epic battle sequence is outside, Robot scouted out deserted areas in Japan, and hit on doing some film shoots around Mount Fuji. With the need for raw texture and lighting material for the sky, landscape, and clouds, again the team resorted to HDRI capture.
The team says it shot generic mountain scenes to supply visual cues for creating the CG landscape, as well as detailed filming of the ground so as to capture plenty of raw, texture material. All filming was carried out with HD cameras.
Once all the prep-work – motion-captured data, miniature building, photography, and motion-control filming – had taken place, the CG began to take shape. Here, the team deployed a mixture of 3DS Max for general 3D grunt work, and Chaos Software’s V-Ray for rendering. 3DS Max was used due to the teams’ knowledge of it, and due to its high-level of photoreal output, according to Robot.
V-Ray was chosen as it works well with Max and offers the best HDRI writing, according to the team. Software used for motion-tracking included 3D Equalizer, while all the various renders and elements were composited in Adobe After Effects.
With so much attention to detail, and high-resolution textures and rendering, the actual CG was always going to be very heavy. It was a worry that Robot came to realize, and some lateral thinking was called for.
Where possible, models were broken down into manageable parts – for example, the lead character was broken down into body, head, and hair parts. Other elements that made up a scene, such as characters, shadows, and backplates, were all created as separate rendering passes. This speeded up the overall rendering time.
The downside was that everything had to be stitched together in a realistic manner, and here the team used not some high-end FX package, but a bundle of the more humble Adobe After Effects.
The result is astounding – and easily beats any five-minute segment of the most famous CG movie to date, Final Fantasy. The team reckon that was also down to the international nature of the project, with an army of Hong Kong stuntmen directed by Donnie Yen, plus people from the US and Europe, being praised.
Looking back, the team says that photorealism was the key to the success of the movie, and they are happy they achieved their goal of expanding the Onimusha universe and adding a high-level of drama to the piece. The feedback from Capcom’s Inafune-san was very positive, the team says.
So, what’s next? For Robot, the dividing line between movies and games is becoming increasingly unclear, especially with the upcoming advent of the next-generation of games consoles. However, while they reckon that this next-generation of systems will hold the biggest challenge for a CG creator to date, they caution that games with high-quality graphics are not as necessarily fun to play, in the same way as movies with massive budgets are not always fun to watch. Luckily, Onimusha 3 also has gameplay by the bucketload.