Houdini was used to handle our pipeline of combining all the crowd animation generated in Massive with our camera, lighting, terrain and dust, finally rendered through AIR. We completed the show at 2k, finally down-sized to PAL.

The average agent ranged between 3k-10k polygons that we switched between as a pre-render process, depending on how close they were in shot.

For shot 2, we had approximately 7000 agents, where originally we shot 500 and found that it was easier to remove the real extras and generate photorealistic extras where we could shape the crowd in any way.

Shot 3 had 2500 agents and Shot 7 had 3500 agents that needed to seamlessly blend in with the real extras.

We completed the show with 15 air licenses, over 6 weeks from start to finish. Each shot took approximately 6-8 hours to render, except for shot 2, which took 15 hours, purely because we were rendering double the number of agents when compared to the other shots. We rendered the following passes for all shots:  self and ground ambient occlusion, color, key light color, self-shadow, ground key shadow, z depth, and dust.