How Tokyo Drift Was Filmed in Los Angeles
How Tokyo Drift Was Filmed in Los Angeles
One of the more fascinating stories of the Fast and Furious franchise was how Tokyo Drift was filmed in Los Angeles. Most people thought we really did go Tokyo to capture these scenese but in reality, the difficulty of getting filming permits plus the complicated (and expensive) logistics, coupled with actor’s schedules made this impractical.
The third movie in the series, this movie was a new twist to the story we heard in the first.
What you’re about to read is taken from an article posted in 2009 or so from an interview given by Matt Sweeney, who was a big-wig on the VFX team (visual effects). The full, original article is here.
Released in 2001, The Fast and the Furious featured some 150 visual effects shots. Two years later, 2 Fast 2 Furious more than tripled the shot count. With The Fast and the Furious: Tokyo Drift (opening today, June 16, 2006), Universal Pictures is raising the bar even higher with no less than 768 visual effects shots. Some 10 vendors contributed to the project, including (in credits order): Hammerhead (206 shots), Rhythm & Hues (123), CIS Hollywood (172), CafeFX (58), Pacific Title (164), Illusion Arts (4), Industrial Light & Magic (2), Perpetual Motion Picture (19), Digital Backlot (4) and Hatch FX, who provided cliff extensions and a complete environment for six Hammerhead shots. Sixteen additional shots were created by Proof, the previsualization unit set up by overall visual effects supervisor Mike Wassel, a veteran of the first two movies. Coordinating the project with Wassel were vfx producer Crystal Dowd and vfx plate producer Lynn Gephart.
There’s lots of technical talk in this article, so if you don’t know what we’re referencing, Google the word in question and add the phrase “movie making” behind it to help narrow your search. Let’s jump in.
New Challenges Lead to New Ideas
In the two previous movies, visual effects were often utilized to create impossible camera moves, and to simulate car stunts that were too dangerous to perform live. In the third movie, the requirements were completely different. Director Justin Lin wanted the car races and action sequences to feel real, Wassel recalls. He avoided using the kind of impossible shots that we had in the first two movies. On this project, the main requirements of the visual effects were creating greenscreen composites for car interior shots, and also placing the action in Tokyo, Japan. Some of the car races take place in the Japanese capital, but it was impossible to have the streets closed down for our location shoot. So, we had to create the environment digitally, which became a huge challenge.
With so much of the action featuring the characters driving around, it was of paramount importance to set up a methodology that could convincingly place the principal actors behind the steering wheel. Three different techniques were used. For some shots, the actors were photographed with a tow rig. In other shots, they were filmed manning a mock steering wheel, while a stunt man was driving the car from the other front seat. Finally, a large number of shots required the actors to be shot on a green screen. Most of the time, the cars were photographed without any window or windshield. They were later added in CG, with all the proper reflections. Special effects coordinator Matt Sweeney built an air casters rig that was able to float multiple car bodies on a large dance floor on the greenscreen stage. This gave the director very organic movement between the cars and the camera.
The Wassel Rig
Wassel and his crew then set out to capture the background plates, using a newly designed rig. On the first movie, we used a six-camera Circlevision system that captured the 240° of the environment, using 28mm lenses, Wassel says.
On the second one, the rig comprised two VistaVision cameras equipped with 15mm lenses, which allowed us to shoot with a 200° field of view. For the third movie, the races were of a different kind. It was very important to the director to have the stunt drivers close to the camera. To this purpose, we designed a double array of three Arriflex 435 equipped with 20mm lenses, creating a 150° field of view per camera set. The nodal points using this system were separated by approximately six inches, while with the Vistavision rig, they were 30 inches apart. This enabled us to keep the perspective offset very small, and allowed objects to get very close to the camera, without parallax becoming a problem.
Most of the race sequences were filmed with real drifters driving the cars. When the action proved impossible to capture in camera, CG cars were used. The real vehicles were LIDAR scanned, which allowed Hammerhead to build a very accurate 3D representation of each car. The models were then textured with still photographs of the real cars provided by production. The LIDAR data was shared between Hammerhead and Rhythm & Hues (R&H), both in charge of sequences necessitating CG cars. In particular, Hammerhead handled the suicide run sequence where two characters race down a treacherous mountain road at night. CG animation was used to show one of the cars flying off the cliff edge.
Recreating Tokyo on a Parking Lot
R&H was responsible for the most complex visual effects sequence in the movie. In this sequence, the characters race through Shibuya Square, a landmark location that is to Tokyo what Times Square is to New York: always crowded, always busy and impossible to close down for a movie shoot. So, the art department recreated the whole plaza as a full size set on a parking lot in Los Angeles. They built the entire ground level, including sidewalks, streetlights, etc, but none of the facades, Wassel observes. Everything from the ground up was added digitally. For this sequence, we basically had two types of long shots: we had either real cars shot on the L.A. set that we combined with a CG Tokyo, or CG cars that we combined with the real Tokyo.
The complex assignment was overseen by Raymond Chen, who shared visual effects supervision duties with Michael Meaker for R&H; Lisa Goldberg was the vfx producer. It was a huge undertaking as we needed to build a 3D representation of Shibuya square, but also create crowds of people as well as cars, Chen explains. Production provided us with various elements, including LIDAR scans of the real Shibuya Square and the tarmac set used to stand in for Shibuya, thousands of high-resolution photographs of the actual location, 3D models of the cars, and running footage of the camera car driving through the real square. One of the problems they had on location was that there were always people on the sidewalks. You could never clearly see the storefronts in the plate photography. In order to rebuild the facades, we needed to know what they looked like at street level. So, a small crew went out early in the morning to get a clear shot of every storefront on the plaza. It allowed us to build an exact representation of the buildings.
R&H used the LIDAR scans as the basis for the 3D models that were built in Maya. When we built the city, it was pretty tricky because we wanted to be efficient and not add details that we didnt need to, but we also wanted to be flexible enough for the director to change shots and try new things, Chen continues. We added details in the square itself as we knew this was going to be the centerpiece. On the other hand, for elements that were further away from camera, we simplified the geometries. We didnt really do traditional turntables for the streets and the buildings. Instead, we broke up the intersection into five different pieces of building blocks, and did our texture and material iterations using the camera choreography from a couple of the all-CG shots, to better reflect how they would eventually be used.
The elements were rendered with Wren, our proprietary rendering software, using a bit of global illumination for the city and for the cars, Chen adds. We had high dynamic range images that showed both the Tokyo location and the Shibuya set at 360°, but in the end, we didnt use global illumination that much. Because it was a night scene and there were a lot of shiny reflective surfaces, what contributed more to the lighting were the specular reflections. So, we did a lot of various specific reflection passes. For instance, we used some of the multi-camera footage of the real Shibuya Square to reflect into cars and buildings. We also used a lot of footage of animated signs and advertisings for reflections. The CG team included modeling lead Ann Sidenblad, texture lead Angela Ursilo and lighting lead Brian Bell.
Crafting Massive Agents
“Crafting massive agents” refers to processing files inside the viual animation software. When the Matchmove team, led by Jeff Smith, tried to combine the Shibuya CG model with the live-action plates of the Shibuya set, they found out that the elevations (filming angles) didnt match. A lot of shifting and adjustment was required before both elements could be combined. Other adjustments included adding numerous reflections on the L.A. set plates to better match the lively ambiance of Shibuya square. Once the environment was all set, Rhythm & Hues populated it with crowds and extra vehicles. Dan Smiczek, our fx/Massive supervisor, oversaw the creation of the digital crowds, Chen says. Depending on the shots, we had up to 6,000 CG people in frame. We went to Giant Studios and motion captured about 200 different actions. Then, we built six base models that were used with two rigs, male and female. We also built a series of CG props such as briefcases, shopping bags, etc. This was just for modeling variation. The color, texture and scale variations were much more extensive. In overhead shots, the crowd was mostly made of Massive agents. For ground level shots, we mainly had greenscreen extras in the foreground and Massive agents in the background, but we also had shots with greenscreen people in the center and Massive extras on the sides. One shot even featured a combination of a real crowd and Massive agents, both filling up different sections of the Shibuya intersection. As for the Massive car agents, we had only four base models, but again color and texture variations were used in the renders.
For the hero cars, the production provided R&H with a LIDAR scan of each vehicle. It was a good starting point, but we found out that the scans werent accurate enough for the kind of highly reflective surfaces that we were dealing with, Chen continues. So, we used an arm digitizer to get a better level of detail. The models were then built in Maya. In order to reproduce the shiny surfaces, we rendered out at least two reflection passes for each car: one overall pass and one for the hotter edges. One of the things that we tried to highlight a lot was the Fresnel edging. Our technical directors then spent a lot of time placing and timing the reflections on the cars, so that the CG models would perfectly blend in the environment. With all the illuminated storefronts, the multiple billboards and the animated signs, it was a real challenge to get it right. The last touch was to add the smoke generated by the tires. To this purpose, particle animation was created in Houdini making use of R&Hs in-house fluid simulator Ahab, and rendered using the companys field expression language Felt via Wren. Finally, the elements were combined in Inferno, Shake and proprietary compositing package Icy by 2D supervisor Harry Lam and his team.
After the numerous in-your-face money shots of the first two The Fast and the Furious movies, the invisible visual effects of the third film came as a welcome departure for Mike Wassel. Subtle effects are fun to do as well. The synthetic Tokyo work that we did is extremely effective. You really get a sense of the scope of the work when you see the before and the after of these shots. Seeing the plates that we shot on a parking lot near Los Angeles, and then watching the final results, is really impressive. You would swear that this car race was shot in the middle of Tokyo!
With all of this said, this is not to say that NOTHING was filmed in Japan. Some close-up shots and other shots we could should on the run were done and in some cases, bootleg style.