Shawn on animation AI

The MLP fandom has historically struggled with creating show-accurate 3D pony models. After seeing how great your Second Life pony models looked, I wanted to get your thoughts on a few things. If you have time, I'd love to hear what you think.

 

1. It's common practice in the fandom to change the 3D model depending on the camera position relative to the pony. This is supposedly because the 2D perspectives shown in Friendship Is Magic are physically impossible. Your Second Life models don't seem to use this trick and they look good from all perspectives. Your models somehow look both accurate and well-adapted to 3D. If you have any thoughts on why this is, I'd love to hear them. I'm especially interested in which aspects of Cadence you chose to keep and which ones you chose to change when making her model.

 

2. One thing that would guide our early AI efforts is a better understanding of what difficulties and tedium animators face. For 2D, we learned from a fandom animator that one of the biggest barriers to creating high-quality animations is in creating good poses for keyframes. As a result, we decided that our first task should be to create a library of character poses that animators could use as a reference or starting point for their own animations. From your interview, it seems like this is less of a problem for you, and that you find it more difficult to create backgrounds. If you have any other thoughts on what prevents animators from producing high-quality content at a mile a minute, I'd love to hear them. This goes for both 2D and 3D animation. (By the way, we accidentally collected a lot of what's necessary to create AI-generated 2D backgrounds, and after seeing your interview, I intend to help bootstrap this effort too.)

 

=== Response from Shawn ===

Yeah, I know what you mean about the style of the show where every view of the head is almost completely different from the next. When I did Athena, I just used my old Disney/Warner Brothers design sense that every view has to be of the same model. That was my biggest problem when trying to draw a MLP character. I eventually had to not look at the MLP models, and just draw what was in my head.

 

Pre-made poses would help a lot with people if you used the same leg parts. MLP would use, say, a more curved leg for one position, and the other pose might have a slight bend in the other position, which would make it difficult for a new animator to work with, not knowing when and how to switch between the different leg shapes. Athena, being a rush job to make, only spending a day and a half on her, she has very few different leg shapes, so I can knock the animation out as fast as possible. So I tend to rely more on timing and performance than a nice clean image. My animation, you'll see the legs will disjoint at times, or parts of the rig will go wonky. But I'm blasting through poses fast enough that the average person won't take notice. There's no Flash rig I can't break :D

 

But say if I was going to build a Cadence model in Flash...Which....I might:D....I'd start with one view of the head I liked the best, then make the other views look like that view. Like what you were saying, about it looking more like a 3D model, or a traditional Disney character. Then you also get better and smoother head turns :)

 

When I built the Cadence model, I was looking at the 3/4 view. But then I took some model stuff from Fleur to complete the head. I used a body closer to Fluer, but made the waist even thinner. So hodged-podged my Cadence together from different shapes. My version of the pony characters have a slightly different look, only due to my inability to draw anything on model, and I resorted to some of my old Disney shapes.

 

=== Next ask ===

That's a good idea, starting with one good view and trying to generate the rest from that. That's also a good point that more consistent models enable smoother turns.

 

I like what you did here:

It may be worthwhile for us to ask experienced animators to critique animations from novices to get a better understanding of what's difficult. If you have more to say on that, I'd be happy to hear it.

 

Same here:

The shortcuts that experienced animators take would give us good insights into where the unnecessary tedium lies. If you have more to say on the topic, I'd be happy to hear about this too.

 

During the interview, you also mentioned that 60fps animation was a pain in the ass because it's a lot more work. There's some AI already for turning 24fps animations into 60fps.

 

If you do make a flash Cadence, please do send some animations with it. I'm curious to see how it differs from the show-accurate Cadence. I know you probably won't want to send her asset files, but if you do, I can compare it with the show-accurate Cadence assets we have.