Issue link: https://digital.copcomm.com/i/1540269
really great job building all of that. In the shots where the lookout guy sees the horde off in the dis- tance through the binoculars, it's important to refer that back to some sort of reality. So they went out and shot these plates in Ka- nanaskis in somewhat representative conditions. They were lucky on those days. Then we had to transpose that into our depth space out- side of Jackson, line that all up. Ultimately it served as a really good reference as to what our horde should look like running through a blizzard looking out from Jackson. I really appreciated the grounded approach they took to photog- raphy. Even if a shot's going to end up being full CG, they go out and shoot a plate. In 99 percent of cases, there would be a plate for us to refer to. That's one of the reasons The Last of Us looks so good, because they set this very high reality bar that you have to hit. Even though some of the work obviously is fantastical by its nature, it's all grounded within some level of reality. Tell us the new compositing-based weather control system you built for this season. Nick Epstein: That came about because of my experience on set. The weather and the shooting conditions were all over the place. We obviously had to conform it to the storm as it blows through Epi- sode 2. It's almost like a character in itself. It's metaphorical, and it's basically as much a part of the story as any of the other characters involved. So we knew these wildly varying weather conditions had to be conformed to this very specific time of day, this very specific bliz- zard feel. That meant we needed a very robust way of doing that. We knew we'd have to execute hundreds of shots, so it needed to be something that wasn't going to be explored on a shot-by-shot basis. It was the brainchild of one of our comp supes, Toby [Wiesner]. We'd been looking at some machine learning techniques to extract depth mats from monocular photography. Also, it gave us segmen- tation. We had LiDAR [light detection and ranging], which was cap- tured on set, and we would convert that to deep data. We would then remap these depth passes based on that deep data, which gave us a very cohesive and controllable depth space. So within that, we could insert our CG horde. We could insert any sort of weather we wanted. That would give the compositing artists a very good basis to start with. All we'd ask them to do is really identify the wind direction in the plate, which again could be varying. There could be no wind. You'd look at surrounding shots if that was the case. The system would basically populate a number of particle systems—all based in Nuke—which would be the blowing snow, a number of volumet- ric elements which had been pre-generated by effects. This system would lay them out based on depth from camera and conform all of that to the wind direction and the selected density. Basically, you'd have a fairly convincing weather system inserted into any plate. Then where it was a little bit flickery or the segmentation was a bit rough, that's where we would employ roto to then go in and actually fine-tune all of the edges and things like that. What that meant for animation is also that we were able to give them a pretty good pre- sentation. So even at an early stage, animation was able to insert their work in and show it to the client with a very good representation j u ly • a u g u s t • s e p t e m b e r 2 0 2 5 c g w 1 7