What was that problem [that left the facial customization system stalled for about a year]?

The attachment problem. So the attachment problem, it's super interesting to talk about. And it's not the first thing you'd think about within this. And maybe we'll talk some body customization questions later because that's what the big challenge is within there, is the attachments. So morphing a face and doing all these things is complex enough. But then when you tack on attachments to it, so now you've got hair, now you've got scalps, now you've got glasses, now you've got helmets, now you've got monocles, now you've got top hats, now you've got all these kind of things, you have to make sure that those still fit every single face that you can have. And like we showed in ATV, there's a huge range of human anatomy. And generally, that's why they have multiple sizes of hats. That's why they have multiple sizes of helmets. But guess what? We can't make that. I can't go to Josh and say, cool, you've got this awesome idea for a helmet. Great. Give me 30 of those with just slight adjustments to every single one. So we needed a much more dynamic system.

We could put those in our shops. We could have a size one helmet and a size two helmet. But you could only get the one that's for you. You would have a hat shop.

You would have a helmet shop. You would have all these things. So we needed to make sure that we could implement a system that could adapt the shape of the attachments to fit, basically. So that's what we did. When Andreas Rohr came on last year, who did part of the demonstration there at CitizenCon, he was the one that kind of cracked the case on this, which was using some concepts, to be fair, from the film side that he had done most of the work within, but then applying them to a runtime implementation. So these were all these new deformers that we've created that are runtime deformers. We call one elastic, one is non-uniform scaling. And deformers are a method of moving verts around or deforming them, obviously, at runtime, based on some sort of input, and the input is the face. Now, what gets weird about it is there's different deformer types for different types of attachments, and Ivo and Andreas went through this really well in their presentation. The elastic type, basically this is one that follows per vert the offsets for every single vert against every other vert in the attachment. It wouldn't work on a helmet because the whole thing would contort into such a weird shape that you wouldn't be able to recognize that helmet anymore. So it was really important to bring those deformers online so that we could actually roll the system out, and that's why there was a big, big delay within it. We were still using the system behind the scenes; we couldn't apply it or give it to the players without solving the attachment problem.