When she comes to, Alita is confused by this futuristic world and has no memory of her past, yet soon discovers that she possesses unique fighting skills. As she struggles to adapt, those closest to her have a difference of opinion on how to help her: Ido tries to shield her from her past, while her new streetwise friend Hugo tries to help her uncover it. And, there are those in this unsettled world who would like to control her and use her skills for their own gain.
Alita: Battle Angel marks the first collaboration by the legendary filmmakers James Cameron and Robert Rodriguez and is based on the manga graphic novel series from Yukito Kishiro. Due to his commitments for the new Avatar films, Cameron served as producer and co-screenwriter, while Rodriquez directed.
Finding the person to play the title character in the 20th Century Fox film
Alita: Battle Angel was no easy task. The person had to be small in stature with a big persona. And, she had to perform complex martial artistry. The directors found their Alita in Rosa Salazar, who would perform the lead on screen. Yet, it was the team at Weta Digital that acted as cyber physicians themselves, fusing Salazar’s movements onto a CG character, and then bringing her to digital life as she performs amazing stunts and exudes a range of emotion required of this central character.
For Alita, Weta orchestrated a new level of CG facial sophistication, working at the level of facial musculature. So, it’s no longer about just moving the surface skin, but moving the underlying muscles, says Eric Saindon, VFX supervisor at Weta. This is apparent in how the movements of Alita’s face look so much like Rosa’s. “We spent hundreds and hundreds of hours just working with Alita’s mouth because what makes even a big action scene work is getting even the most human expression, and Rosa has a very expressive face,” he says.
In the film, Alita is 100 percent CG throughout. “It’s actually funny how some articles have reported that she is 90 or 95 percent CG and partly Rosa in the movie. The only time Rosa is on camera as herself is as an extra in front of Kansas City Bar when Hugo is being chased by Zapan,” says Saindon.
Here, Saindon provides an in-depth look at Weta’s work on Alita in an exclusive interview with Karen Moltenbrey,chief editor of
CGW.
How long were you working on the project?
Officially I started onAlita three years ago onset in Austin, Texas, where we filmed for four months on the Trouble Maker Studios backlot. Mentally I started on the film in 2007 when I talked to LEI for the first time about two projects James Cameron was developing –
Battle Angel and A
vatar. I left the meeting and went out and bought the first few volumes on the Manga. From that point forward, I was hooked and wanted to see
Alita in a live-action format.
Why so many supervisors assigned to it?
With the scale of the film, we split the work into five teams headed by an overall visual effects supervisor and an animation supervisor. This way, we were able to focus groups on certain types of effects and performance and specialize the artist types.
In what ways did the work push the state of the art in CGI?
The main thing I think Alita pushes is the ability to bring Rosa’s performance through to a CG character at a level never seen before. It’s the first time Weta has done a humanoid character to this level who needed to perform the lead role in a film. We added extra detail like compression to the skin, so when wrinkles form in Alita’s face, the amount of blood in the sub-surface changes. We also used deep learning to detail the pores on Alita’s face. Instead of hand painting the pores, we grew pores on the skin with displacement and lean mapping that took into account each individual pore. With this same information, we were able to add peach fuzz at the individual pore level.
Describe the modelling process for Alita.
When we cast Rosa, she was never intended to be the look of Alita, but because Rosa gave such a strong performance, it was very difficult to separate the two. Once we started doing facial motion tests for Alita, we found that using 3D scans from Rosa’s mouth and eyes were the best starting point for Alita’s model, although the base for the design was still Yukito Kishiro’s artwork. Gathering lots of still frames from the manga, the art directors were able to come up with a comprehensive design for Alita’s model. The cybergirl model itself was constructed of thousands of individual gears, like an old clock. Each part was rigged to move with the actions of Alita.
How large was the model?
Alita’s cybergirl model was made up of around 5,000 individual parts. Her irises were built using a simulation and solved as individual fibres. Each eye consists of 9,500,000 polygons.
What was the biggest challenge when it came to modeling her?
Her cybergirl body was extremely intricate, so that was a complexity challenge, but perfecting the face shapes, particularly the mouth, was probably the most challenging thing - ultimately solved by using Rosa's mouth and phoneme shapes wholesale instead of re-targeting them to Alita.
Describe the texturing/skinning technique and the use of deep learning for laying out the pores. How did this elevate the model?
We used a new process to lay out pores on Alita’s skin, which followed the flow of her skin in a natural manner. By using this method, we could line up all the additional layers like displacement, lean and skin depth perfectly without cloning.
Describe the approach to her eyes and how it was achieved.
Alita’s eyes were built using a simulation of fibers to represent the dilator muscle responsible for the dilation of the pupil. By simulating the action of this, we were able to get very high-resolution geometry for the iris. The end result was 8.5 million polygons per eye compared to Gollum, who had 50,000 polygons total for his entire character.
Describe the approach to her hair – the tools you used, how it was grown and groomed.
In terms of simulation, we developed a solver that could simulate every single hair individually, rather than a more traditional approach that would simulate ribbons for example, and this could be driven by other participating media such as water. This gave us complete physical accuracy in situations such as Alita entering and emerging from the lake.
How many hairs does she have on her head?
We grew over a million hairs on Alita’s head. Then using proprietary software that mimics a hair stylist’s tools, like brushes, blow dryers and scissors, we groomed her hair to multiple styles for the film.
How many hairstyles does she sport?
Alita only really sports two hairstyles, one for each body, but then has multiple variants of each. Since we were doing full simulations for all her hair, we quickly found that her eyes got blocked in certain shots or we couldn’t get enough light in her eyes. To help this, we built multiple starting positions for her hair to help us through these situations, from hair behind her ear to fully dishevelled when she is cut in half. In all, we had over 40 starting points for hair simulation.
What were the challenges regarding her clothing?
Like any actress, Alita had lots of wardrobe changes. Before we shot anything in Texas, we had all her costumes physically built so we could have a lighting stand-in walk into every shot and roughly act out Rosa’s performance. This way, we had very good lighting and motion reference for what her wardrobe would do in any given shot. In all Alita, had 28 different outfits for the film.
How much of her body performance is motion captured?
Rosa performs Alita throughout the film and carries the character through all the drama. She had a stunt team that performed some of the more difficult pieces of action choreography but she also was involved in performing stunts herself. There are certain sections of the film (like the Motorball sequence) where animators keyframed the performance in order heighten Alita’s abilities or hit stunts that were beyond what was physically possible. Overall, Rosa’s performance is what drives the Alita character. We call it performance capture because we have moved beyond capturing just the physical motion of an actor, and we aim to capture the essence of Rosa’s performance in any single moment. An actor brings conscious choices to how a character behaves through the arc of the story, how they perform with other actors, and how they act and reacts with conscious and sub-conscious intuition. This performance is what draws the audience to a character.
Was there anything new in terms of body performance?
We built a digital actor puppet of Rosa. We used this to verify the performance of Rosa for body and facial (primarily facial). This proved to be invaluable in getting the rig working properly.
Describe the animation rig and the underlying musculature
Alita has multiple bodies in the film. When Dr. Ido, played by Christoph Waltz, puts Alita in her cybergirl, the internal structure is a complex clockwork, and each piece is rigged to move with Alita’s motion. The internals are covered by hard external shell pieces that are semi-transparent. All of these pieces are rigid, so they had to be designed and rigged in a way to avoid inter-penetration while achieving the posing. This is different from the Berserker body she finds later in the film, which is a muscle-based rig. Muscle movement allows you to accommodate the volume changes as flesh expands and contracts in different posing. The cybergirl body was a challenge as we needed to move back and forth between models, creatures, and animation to get the design working properly.
How did you elevate the performance capture?
Our goal was to make our on-set presence disappear so that Robert and the actors were not thinking about the visual effects and could focus solely on the performances and the story. A big part of this was giving Robert the freedom to shoot in live sets and not limit how actors interacted with one another. Rosa performed directly with other actors in live environments, which gave real reactions and vibrant performances that are more difficult to achieve on greenscreen shoots.
Why were two facial cameras used? What was the result of this?
Multiple cameras allow us to triangulate the dot positions and find their position in-depth from the cameras. With one camera, we only understand marker movement in two axes. With the two cameras, we understand the motion in three axes and are able to track and solve the facial muscle movement in all three dimensions.
What challenges did you have in integrating Alita into the live action?
This was actually not too problematic as Rosa was always captured live on set, so contacts were generally accurate, and we always had accurate lighting reference. We also always had a lighting double in shots dressed in the same wardrobe that Alita would later wear digitally, so we knew exactly how her clothing should look. There were some extensive paint-back requirements (the only real downside to capturing Rosa on set), and roto requirements for some of the crowd shots and the scrap yard scene at the start, which had some very atmospheric plates but nothing too out of the ordinary.
What kind of compositing challenge did she present?
The first thing we said to Robert Rodriguez when we started the shoot was not to have the actors hold back in any way because of the visual effects. For this reason, we got great performances from the actors, but we also got a huge amount of interaction between Alita and other characters. Getting the interaction working is always a challenge and add onto this shooting in native 3D, the problem becomes a huge challenge.
What was the biggest technical challenge in creating her?
Probably the most challenging scene was when Alita was eating the orange with Ido. The amount of detail in Alita’s face and body was higher fidelity then anything we have ever done before.
Did you take a similar tact with the other main CG characters in terms of modelling and animation? How did the approach differ?
There are a variety of digital characters in the film. Many are fully digital like Alita, for example Gelda and Grewishka, and others are a blend of live action and digital, like Zapan, Nyssiana and the Motorballers. Blending live action and CG is a challenge on a 2D film, but because this was a native stereo film, we had to accurately connect the CG bodies to live action heads in three-dimensional space, which was another level of difficulty. We also needed a way to adjust these performances in action scenes, so we needed a robust pipeline that let us move these characters around without breaking the stereo on the plate elements.
Describe some of the bigger environments you built, and highlight the technical challenges.
We built set extensions to Iron City itself, the Scrapyard, extensions to Ambush Alley, the factory exterior and interior, the lake environment -–which included the underwater scene and the crashed warship interior, the Underworld, and the Motorball track.
Which was the most difficult and why?
Underwater was probably the most difficult as we opted for a fully path-traced approach. Everything in the frame was rendered together, with the water surface, volumetric interior and underwater camera housing all have an accurate/physically correct effect.
What were some of the bigger effects in the film?
The shots of Alita's Berserker body morph were really intricate. There were some fairly tricky rigid-body/destruction type effects in Underworld and the factory fight, but the biggest effects were probably those in the supply tube flashback scene.
What kind of simulations did you have?
As above, we used a new proprietary solver for water and hair interaction. We studied a lot of heavy artillery reference for the factory fight and supply tube battle sequences, and developed a library of Houdini based set-ups for things like impacts, muzzle flashes, tracers, ricochets
Is there anything else you would like to point out that our readers would like to know about in terms of the work you did that's noteworthy?
The most important bit of work we do at Weta is how we create memorable digital characters. A character’s performance is sold through their face, so we invest a lot of time and energy into creating photo real digital faces. Because Alita’s character would carry the film and the audience would follow her as she experienced and reacted to the world, we needed to research and develop a deeper level of face construction. This involved many departments: R&D, Models, Creatures, Animation, and Look Dev, to name a few. Alita has a young face with very smooth topology, so it was very important to build the structures underneath the skin to have the surface behave correctly in order for the emotion and performance to read correctly. Understanding the shapes that different muscles take, their behaviour in motion, and how all this affects the face was critical to deliver this performance. All of this work needed to be directable, so we constructed the face in a way that we could add missing detail as we developed each shot. The devil is in the detail, and we needed to convey every detail.
In what way does the work, especially on Alita, break new ground (or push forward) in terms of realistic CG characters?
The biggest thing Alita does is allow a director to bring any vision to life, whether its Robert Rodriguez, James Cameron, or any other visionary director. By using the latest technology and artistry ,we were able to capture an outstanding performance and put it into a character that could not be created without visual effects.