The planes were modeled from source material comprised of photos, survey data and in the case of the main camera plane, Lidar scanning of a 1/4 scale miniature. The models were built in perfect smooth form, and then the modelers added imperfections, like slightly bent wings or dents in the fuselage to give the planes a hand built feel. Photographs of miniatures were used for texture maps.
The choreography of the sequence was inspired heavily by the shots in the original movie, Hell's Angels, capturing the high energy and danger of the 1931 film. The planes were animated by standard key framing to maintain control of the choreography. The plane's flaps and cables were also animated based on the motion of the planes. The intention for the sequence was to match the high-flying spirit of the original film, as Howard moves his camera from one plane to the next as the aerial dogfight unfolds before him.
The planes were then rendered in multiple passes. We rendered two occlusion passes. One blocking out ambient light and the other gave us reflection occlusion, where a matte showed us where the object was self-reflecting. We used this in combination with the environment pass to give us a less computationally expensive look for reflections. Finally, warm key lighting was added to give the sense of late afternoon sunlight.
The compositing of the sequence was done using a proprietary node based compositor called Bonsai. All of these elements combined to create the high-energy shot of Howard Hughes trying to capture the action of a massive dogfight for his film Hell's Angels.