Step-by-step
HOW Weta Digital brought their Cg alita to life, based on the performance of actress rosa salazar
1. Performance capture
Salazar performed scenes in a custom-built motion capture suit, and wore a head-cam with stereo cameras. It pointed back at her face, which was covered with tracking markers.
2. Translation to CG character
Weta Digital’s proprietary facial solver and a host of facial animation artists take the raw motion capture face data and translate it to a CG avatar of Salazar in the form of Alita.
3. Final shot
The computer-generated Alita, carried out with the studio’s proprietary path-tracer Manuka, incorporated new developments in skin shaders and hair simulation for the final result.
tuning and keyframing done by facial animators and applied to a ‘Koru’ facial puppet. This is a Gpu-accelerated, full-resolution version of the character’s face that the animators can then work with in real time.
Throughout the process, Cozens says Weta Digital always strove to “balance the actual character design with the actor. It’s really important, and never has been as important as Alita, to get as much of Rosa into the actual physical model and the design.”
This was aided by reliance on research surrounding the Facial Action Coding System (FACS), which Weta Digital has capitalised on since King Kong. “This was based on research by Paul Ekman who looked at how muscles work on your face, and how your emotions trigger different parts of your face,” notes Cozens. “When we’re building a face puppet, we tend to look at really broad expressive facial changes, like a really big smile, or a big sad face, break those down into the related muscles that are firing, and turn those into muscle face shapes that we can trigger.
“Now, on this film, because of the amount of dialogue, and the amount of subtle dialogue that happens, if we look at the biggest smile that you could put on your face, where your face is lit up and everything is firing and we [rate] that between 1 and 10 – that’s the 10. That means your dialogue is all happening around 1 and 2, a much more subtler range.”
Cozens adds that detail in the low range proved extremely necessary for Alita, because Salazar was able to form a ‘crazy’ amount of shapes with her mouth. “So, when she’s eating an orange, or reacting to frustration because she’s arguing with someone, her face fires in all sorts of unusual ways, and it’s stuff happening in the low range. We had all the big expressive stuff, but we needed to go deeper into what’s happening specifically around the mouth, and specifically for dialogue. We did a bunch of research talking to plastic surgeons about what is going on around the mouth in the speaking range.”
ANALYSING THE EYES
In making Alita’s face, Weta Digital also drew upon, of course, years of experience in crafting the eyes of digital characters. A mix of anatomy, artistry and technological progression has seen the studio’s approach to eyes change over time. “For Gollum in the original Lord Of The Rings,” says Saindon, “the eyes were simply two textural planes with refraction and some fairly simple shader tricks. Now that we’ve switched to our in-house renderer, Manuka, we’ve been able to rebuild the eyes in such a way that they’re fully realised.”
Just as artists tend to build parts of the face from the inside out,
Alita’s eyes were modelled based on the fibres of real eyes – which actually start life as baby eyes – which are then further ‘grown’ via simulation to add more and more details.
“We have a little simulation of it running and it basically shows all the fibres and how they stretch and billow a little bit and break,” explains Saindon. “The end result is you end up with geometry for the iris itself, which gives you the amount of detail, and it’s crazy. It’s something like nine million polygons per eye. We can achieve proper ray-traced refracted shadows through the eyes because it’s fully realised geometry.”
One extra challenge with Alita’s eyes is that they are intentionally larger than a human’s, channelling the film’s manga origins. Weta Digital had to change the amount of sclera – the white outer layer of the eyeball – that was visible to ensure they were still largely readable as human eyes.
“There’s a scene with Hugo (Keean Johnson) and Alita on a bridge, and it’s a really intimate moment,” recounts Saindon. “Rosa was looking up at Keean a little bit and didn’t lift her chin all the way. Her eyes were up, looking up slightly, but you never saw any sclera, the whites of her eyes under her iris. When we translated that to Alita, once we copied that exactly, you saw a little bit of white under her eyes and it made her look really, really creepy, so because of the bigger eyes, she’s got more white areas now around her eyes.” MOVING THE MOUTH Anatomical studies of the face were also crucial when it came to replicating Salazar’s performance in CG, especially, says Cozens, the jawbone. “Your jaw’s your primary motion source on the face. That’s the only joint that’s really moving around, and it’s really critical in getting the mandible working correctly, and also knowing how
it affects the muscles around the mouth.”
Aiding in the process was the fact that Weta Digital started with scanned data of Salazar herself; she would make a number of poses and read lines of dialogue that ran through a range of mouth movements. “So,” says Cozens, “we look at those scans, and look at what the mouth is doing. There’s subtle changes just based on suction and jaw position that really change the shape of the orbicularis oris muscle – the oris is the ring muscle around your lips – that we had to get right.”
Other muscles and mouth parts, including the modiolus and the incisive, were also considered. The idea, essentially, was that Weta Digital not only considered the look of the exterior surface of the face, but what was going on beneath it as Alita is talking, emoting or performing. When the audience had to see inside the character’s mouth, that was also directly modelled and animated thanks to dental scans of teeth and gums.
“The inner structure of the mouth turned into another thing that we had to chase in order to get the interior of the mouth correct as that jaw opens, and the play of the lips and the inside of the lips against the teeth and the gums,” outlines Cozens. “We were always building from the interior towards the exterior, looking at all those layers, building all those layers into the puppet – it would completely change the detail and the read of the detail on the surface of the face.”
“You may never notice the detail that’s gone into making that work, all those micro expressions and all that subtle fleshy detail,” adds Cozens. “But I think if we didn’t have it in there, it’s when people start going, ‘Hmm, I don’t know…’ So I think it’s critical, and it’s that extra level, that critical level of detail, that brings her face to life.”