Massive Crowd Control in Thor: Ragnarok

The third act of Thor: Ragnarok is primarily a major battle sequence on Asgard, as Thor and his superhero friends take on Hela, who is intent on claiming the throne. The sequence includes a confrontation between scores of Asgardians and members of Hela’s D-Guard army, many of which were realized – often in close-up – using crowd simulation software Massive.

Behind the work was Framestore. VFX Voice caught up with Crowd Supervisor Reza Ghobadinic to find out how some of these battle scenes were created.

Going with crowd sims

There was a lot of character animation required in the third act – Hulk, Hela’s giant wolf Fenris, the fire god Surtur, and many other CG creations. So why was crowd simulation used for the Asgardians and D-Guards? The answer was their sheer number. Otherwise, they would have been handling the 2,000 fleeing Asgardians on the Rainbow Bridge and the attacking D-Guards by keyframe, which could have been a lengthy process. One shot even has the D-Guards climbing up and over each other, which also would have been incredibly complex to achieve any other way.

Watch the trailer for Thor: Ragnarok.

“Some of these shots needed to go through a process of iterations to develop the look and feel that was satisfying,” says Ghobadinic, “and with the complexity and the big number of the characters in some shots, hand animating them was definitely out of the question. So in the end, we decided to go with crowds as the base, with a handful of animated or promoted characters at the front rows in front of the camera.”

“We needed a crowd software that gave us custom control in certain aspects of animation and character behavior with the ability to simulate lots of hair and cloth for more of a realistic look and feel. It became evident that Massive was the software that would give us the flexibility we needed.”

—Reza Ghobadinic, Crowd Supervisor, Framestore

Framestore hadn’t used Massive for a number of years, but the crew became familiar with the software again on King Arthur: Legend of the Sword. That film served as a suitable test bed for what would be needed in Thor: Ragnarok, which required significant close interactions between Massive agents and the agents being relatively close to camera.

“We needed a crowd software that gave us custom control in certain aspects of animation and character behavior, with the ability to simulate lots of hair and cloth for more of a realistic look and feel,” describes Ghobadinic. “After looking into a few options, it became evident that Massive was the software that would give us the flexibility we needed at the time and provide us with the confidence to get the job done successfully and with less effort.”

In certain shots, D-Guards would be brought to life by both hand animation and motion capture, and Massive crowd simulations.

Getting started in Massive

To get started on the Massive crowds, Framestore began by developing ‘agents’ to represent the Asgardians and D-Guards. The Asgardian behavior needed to be panicked and packed together as the enemy presses them from both sides on the bridge.

“We started to develop an agent for those moments,” outlines Ghobadinic, “using raw motion clips at first, then starting to improve the actions with the help of our animation department and more motion-capture clips at the end.”

“Manually setting up agents with that many geometries and variations from those rigs would be quite tedious and error prone, which is why we made a tool to automatically read those rigs and find out all the variations from the geometry tags, create related geometry and option nodes and their agent variables, sort out the skeleton shape and properties, and save it as an agent, all pretty much by a single click of a button, in less than few minutes.”

—Reza Ghobadinic, Crowd Supervisor, Framestore

“For the D-Guards,” he adds, “the main challenge was getting their style of motion right, which was mostly the animation department’s responsibility. But on a more technical side, for us, was getting them to climb up on each other or some other geometry at all kinds of different angles, sometimes even upside down. So that is what we started to do first. We had a very few actions to start with and when we got the maths right, we could concentrate on the quality of our actions to make them better.”

The Asgardians and D-Guards face off on the Rainbow Bridge.

The Massive agents were highly detailed, with the male and female Asgardians developed as separate agents with five variations for each gender. The D-Guards had three different major variations and a few different helmets and capes. “That is,” says Ghobadinic, “three rigs, two for the Asgardians and one for the D-Guards that hold all the variations, too. The crowd version of D-Guard had the same number of geometries as the Hero, plus a couple of lower resolutions of each geometry as level of detail. That amounts to a few hundred geometries for the D-Guards, for example, in a single rig.

“Manually setting up agents with that many geometries and variations from those rigs would be quite tedious and error prone,” notes Ghobadinic, “which is why we made a tool to automatically read those rigs and find out all the variations from the geometry tags, create related geometry and option nodes and their agent variables, sort out the skeleton shape and properties, and save it as an agent, all pretty much by a single click of a button, in less than few minutes. Later we would add the motion tree, actions and the brain [from Massive] and start testing them in the shots.”

The signature shot

Perhaps the stand-out crowd shot in the battle sequence is a view of the D-Guards ‘massing’ onto Hela’s spikes towards a now lightning-enriched Thor as he descends towards them. This shot, added late to the film, proved tricky for several reasons, as Ghobadinic explains.

Thor descends upon the D-Guards.

“First, they had to run towards a point and start making a shallow pile. Then the rest would jump on their back and start climbing up on others, not on a straight path upwards, but on a curve towards the sky and the spaceship. The problems we had were ranging from getting a lot of penetrations, having a difficult time keeping them on the curved track, changes in the Z to Y forward vectors, trying to keep them facing the right direction while climbing and keeping the crowd density uniform, and keeping the shape of the whole thing right as well, just to name a few.

“For the D-Guards, the main challenge was getting their style of motion right, which was mostly the animation department’s responsibility. But on a more technical side, for us, was getting them to climb up on each other or some other geometry at all kinds of different angles, sometimes even upside down. So that is what we started to do first. We had a very few actions to start with and when we got the maths right, we could concentrate on the quality of our actions to make them better.”

—Reza Ghobadinic, Crowd Supervisor, Framestore

“Eventually,” continues Ghobadinic, “using a mixture of 3D lanes, sound, agent fields, manipulating the rates and blending actions smoothly, we managed to create what we wanted. The base crowd sims for these simulations ended up being quite clean for its purpose with a few visible collisions and penetrations between the agents. So we used a smaller version of that as the base for the shot, and the animation department jumped in after our crowd was approved to fix some of the penetrations and to add a couple of their own characters in the mix.”

That kind of attention to detail certainly makes the Thor: Ragnarok battle scenes stand out from the crowd.

Read about the future of Massive developments here