Also, how the stunning CG dog Cosmo was created, and the insane 120 fps shoot behind that corridor oner fight scene.
The scenes of a young Rocket meeting and befriending a group of anthropomorphized test animal subjects–the otter Lylla, the walrus Teefs and the rabbit Floor–in James Gunn’s Guardians of the Galaxy Vol. 3 are perhaps some of the most touching in the Marvel Cinematic Universe.
While final animation of these characters would be deftly handled by Framestore–the same visual effects studio that had first produced a CG Rocket (voiced by Bradley Cooper and played on set by Sean Gunn) for the original Guardians of the Galaxy–scenes of the animals locked in High Evolutionary cages in Vol. 3 actually began life as an early ‘virtual production’ shoot with actors in motion capture suits. This enabled the director to get a very quick take on the performance, and giving VFX artists plenty of time to nail the final effects shots.
This is just one aspect of the film’s making shared here in befores & afters’ interview with production visual effects supervisor Stephane Ceretti, who re-teamed with James Gunn after having worked on the first film, and on the Guardians of the Galaxy Holiday Special (filmed at the same time as Vol. 3).
Ceretti also shares details of the impressive work, also by Framestore, carried out for the CG dog Cosmo (voiced by Maria Bakalova), and how the almost 2-minute long oner fight scene in the Arete ship corridor was pulled off by Wētā FX. Finally, Ceretti reveals where you can see his own cameo in the film…
This is only the start of befores & afters’ in-depth coverage of Vol. 3, stay tuned for detailed pieces on many of the different VFX studio contributions.
‘I want to shoot the scenes in the cages the same way I shoot everything in the film.’
b&a: Stef, where did you start in terms of the flashbacks to the caged animals, in terms of conversations with James about how they would be brought to life?
Stephane Ceretti: The thing that James wanted to convey mostly is, they’re kind of monsters, but they’re super cute at the same time. For Rocket, we talked a lot about the evolution of his body throughout the steps of the film, because you find him as a little kid in the cage where he is just a little raccoon. But then he goes through all those steps of transformation and being worked on. The first image James showed me was an image that we actually looked at even a long time ago on the first film of a little baby raccoon with a big head and these big eyes, and it was all about the super cuteness that he wanted to get out of it.
When we did the first film, we actually did one day with a raccoon called Oreo. Oreo the raccoon was our main reference for Rocket. And on that day we also had little baby raccoons that were just born. I had a picture of James holding a baby raccoon from 2012 back in London. I showed that to him again on this film. He said, ‘Yeah, that’s exactly it. He needs to be that tiny and that cute and that frail at the beginning.’ Then we also had to develop the different metal plates and things he has on his chest as he grows bigger.
b&a: I love that James has been sharing Framestore’s test animations and turntables. What kinds of things do you say to Framestore at the beginning of the process to help them explore the animation of the characters?
Stephane Ceretti: We go through the script and we tell them the different beats they’re going to have to do. And actually, one thing we did was start by rebuilding Rocket. There were a few things that had changed across the films that I wanted to bring back a little from the first film, especially around the eyes and how black the eyes are, because he’s an animal. That’s something we worked on a lot in the first film, was keeping his animalistic features as much as possible. We also knew we would be much closer to him, so we had a lot of detail to add to everything that we had. We would have some super, super extreme close-ups.
Then they started building all the different phases of Rocket. But at the same time, in terms of animation, we really hit the road running on this because we decided that we would shoot all the scenes in the cages and the white corridor scene, when he finally meets Lylla in the afterworld, we shot all those scenes the first two days of production. We did that because we knew we would need as much time as possible to make those characters as good as possible.
Initially, it was later in the schedule, and everybody was a little bit unsure about how we would shoot this, and that’s when we decided to somewhat of a virtual production to shoot it. We didn’t use LED walls or anything like that. It was more like shooting a reference shoot.
Filmmaking has really evolved quite a bit between all the different films and between these films and The Suicide Squad, James’ DOP Henry Braham had been using very compact RED cameras that are essentially on his chest, meaning he could shoot almost hand-held style, but still very stabilized. James said, ‘I want to shoot the scenes in the cages the same way I shoot everything in the film.’ So, we ended up doing motion capture of the cameras, not of the actors, but of the cameras. We shot the entire scenes the first two days of production with three cameras shooting at the same time, all motion captured with our friends from Imaginarium.
Now, we did have motion capture from the actors, but it was more for triangulation of the 3D space, more than anything else. It was not for capturing animation. We do all animation with hand animation.
We shot the entire sequences very quickly. All the scenes were shot in two days, and the editor, Fred Raskin, could edit that directly with what we shot, because they had the shots that James wanted to have. Then we turned that over to Framestore for postvis. We could show Framestore that first cut and they saw very quickly what it needed to be.
Then, as we were building the models, we were already giving Framestore all the animation reference they needed to have in terms of what actions they would have to cover. That meant they could start to do animation on their side, they could start to study animation of little runts and kids, how real baby raccoons move, and how they play with each other.
b&a: Just to be super clear, do you mean you were shooting this virtual production reference with the actual voice actors, or stand-ins?
Stephane Ceretti: They were the voice actors. However, Sean Gunn was still doing Rocket, as he does normally on set. And then later the voice is replaced with Bradley Cooper. Actually, a few scenes are still with Sean’s voice.
b&a: Oh really?
Stephane Ceretti: Yeah, when Rocket is very, very tiny.
b&a: Did you shoot these scenes on a ‘volume’ set, ie. that might normally be used for mocap, or was it on built sets with mocap cameras positioned around?
Stephane Ceretti: We had a set. Obviously our actors are different sizes than the raccoon and other animals. So, we had a set which was the size of the cages, but multiplied by a certain factor. Lylla was always the same size, and so were Floor and Teefs, but Rocket needed to grow over the different scenes. I took Linda Cardellini’s (Lylla’s) size, and that was my multiplying factor between the actual cages that we had on set and the volume that we created for capturing those scenes. It was about a 2.3 factor. It was a little complicated because once Sean would reach his highest size on the scene, when they escape, they needed to be pretty much roughly the same size. But Linda is smaller than Sean, but they had to be the same size. So, for the otter, we had a little pedestal or platform. One side of the cage was on a platform, the other side was just on the floor. Then we built some oversized tubes for the cage bars.
James loved shooting it this way because it was so liberating. He didn’t have to wait for lighting or anything. We were going really fast, and it was just three cameras shooting at the same time. It obviously didn’t matter if the cameras caught the other cameras in the frame. It was just all about performance, as if we were in the theater.
Actually, Asim Chaudhry who played Teefs performed in a wheelchair, because Teef’s in a wheelchair. To give him the right kind of physicality he had a huge ball behind him and a huge cardboard cutout of the head of the walrus. Floor was played by Mikaela Hoover, and she was just on the floor. She was constantly on the floor, on all fours, running around, jumping. They all had a lot of fun doing it.
The creation of Cosmo
b&a: If that wasn’t tricky enough to craft Rocket and then these other characters in CG, I mean, you had Cosmo, and I really feel like you’ve taken photoreal CG dogs to a new level here. What was the toughest part of making Cosmo?
Stephane Ceretti: The toughest part was that we actually had to do it really, really quickly because she had to be in the Christmas special. Framestore did an amazing job, and Weta actually picked up a few shots towards the end. And then I had James on my back! He was like, ‘When are we going to see Cosmo? When are we going to see Cosmo?’ We were getting into the woods on the Christmas special, which was something like 570 shots to take care of, while we were also doing the 3,000 shots for the film!
For Cosmo, I was really adamant that we needed to have a real dog on set for lighting reference, and that we could scan the dog, and that we could just base everything on this. That’s what we did. It was great reference for us, but the dog was a little crazy! It was called Slate, and it was just moving so much, we couldn’t control that dog. There was no way we could do anything or shoot anything useful with that.
But, that wasn’t the idea, anyway. It was just about getting the dog there in-camera for good lighting and size reference. We did a separate reference shoot with the dog doing different things, very close up, a little bit wider shots, walking, running–which actually Cosmo didn’t end up doing that much of.
Also, it was good to have a dog with us on set because James loves dogs. So if every time we had to do a Slate reference, James would just be super excited. That added extra fun to the shoot. We ended up scanning Slate as much as we could. Obviously it’s still pretty hard to scan an animal. We replicated her as close as possible, and the first thing that we did was, we shot a test shot and we had a side by side between Slate and Cosmo to match it completely together.
We also tried not to ever be pushing the animation of the dog towards something more human, because she’s not talking. She’s talking through the speakers. It’s not the same thing as Rocket. We wanted to really separate and differentiate what happened to Rocket and Lylla and Teefs and Floor, which are cybernetic-enhanced animals. They had gone through all these kinds of terrible things that happened to them. Cosmo is a totally different story–our backstory is that she cannot speak as a human or cannot emote or do facial expression as a human, but she can speak through the speakers.
That actually created something for us that was an opportunity to really make her look like a real dog, not over-animate her. Even when she’s reacting a lot like when she’s being called a bad dog at the beginning of the film and you see her doing that face, that actually came from a dog video reference that we saw on the Internet. The way she breathes, the way she looks, she’s not super expressive with her eyes. The eyes are always wide open and happy.
The oner, at 120fps
b&a: The oner on the Arete is something that I think will go down in history as one of the best oners ever. Tell me about planning that shot and what went into it.
Stephane Ceretti: Well, we had actually had that idea of doing a big oner group fight moment in a corridor on the first film. But we never could really do it correctly there. We couldn’t shoot it correctly, there were issues with the way we wanted to do it. On that film, we ended up re-shooting a scene later in post, and that ended up being the scene where Groot is holding all these people and smashes them against the wall.
But the oner was a long-germinating idea in James’ head. James had been working with a fight coordinator, Wayne Dalglish, including on Peacemaker. He came on this and based on the beats that James wanted to hit, he started to design the sequence. Wayne did the stuntvis which he just films himself and then does a postvis over the top as well. He worked with our stunt coordinator Heidi Moneymaker to design the scene. It was definitely rough, but they were so fast. Even our postvis team, lead by James Baker, who is really great, could not keep up with them! They were constantly shooting something, then adding something, then doing a bit of postvis on it, then rethinking something. It was very chaotic, but it had to be because it was such a complicated scene.
They had to design it and then figure out how to film the characters. We would talk constantly about where we were going to connect all the camera moves. I was a little nervous, I’m not going to lie, because when you’re in that long corridor, it’s not forgiving in terms of perspective. So, if you connect between two cameras and the cameras are not in the right spot, and you are dealing with people dancing and bouncing and jumping and things going around, and the perspective is not right, and people are not in the same place, it’s Hell to connect the shots. And guess what? It was Hell to connect the shots, because there’s no way you can shoot this very cleanly anyway.
b&a: I mean, how do you plan stitches?
Stephane Ceretti: They were planned to some degree. But we had to do that with a camera that is held by one guy who’s in the middle of the scene and you surround that with people jumping around and beating each other up and you’re trying to go from one stitch to another. We shot that over three days. It took us three days to shoot this sequence, which is not a lot, but we did it in order. Actually, there were some suggestions about not shooting it in order.
b&a: Oh wow.
Stephane Ceretti: We did shoot it in order, in the end. It was still complicated. Also, we shot everything at 120 frames per second, all the action, because we knew we wanted to go slow in places and fast in other moments. Once we shot it, Fred edited it the best he could, trying to find the right moments with the music. There’s a song playing, so the beats had to hit some specific moments.
And then, we got this in front of Wētā FX, and everybody was like, ‘Oh my God.’ But they loved it, they absolutely loved it, and they jumped on it. We decided to actually matchmove the entire takes at 120 frames per second. That a two-minute sequence, which suddenly makes it a 12-15 minute sequence.
We ended up with a mixture of CG actor digi-doubles, connecting to real actors, CG backgrounds, fixing the perspective between the different cameras, re-projecting–everything is in there to seamlessly connect the different shots, and add in animation of the Crustaceans and Blurp.
b&a: How did you film the scene to factor in the characters and creatures that would be CG later on, or, where you knew the actors would be doing major stunts so they might be digi-doubles?
Stephane Ceretti: Often it was the real actors because we needed their faces. The actors and their stunt performers were there for all three days. When we filmed a shot, we would always try with the actors first. Then when we knew it was just getting too complicated, we’d bring in the stunties. And then the CG characters, we had performers wearing gray suits standing in for them.
It was very complicated, but Wētā was all over it. It was like preparing a gigantic meal at the table and they were eating it with their faces smiling and crying at the same time.
b&a: Just finally, Stef, are you in the film?
Stephane Ceretti: I am, but it’s a difficult one because it’s at the end of the film and you have to see it in IMAX to see me correctly. It is at the end of the film when Quill goes back to his granddad, and there’s a wall with pictures that we see very quickly. In one of these pictures, where it’s a family picture, it’s actually of me and my siblings, all four of us when I was like one year old, and they were all a little older than me. The camera flies by and you see us. That’s my cameo.
Become a befores & afters Patreon for bonus VFX content