Our Kizumonogatari interview translations continue with a conversation with CG director Shinya Takano, who reminisces about the production and its extremely ambitious approach that grew far bigger than studio SHAFT alone could handle, as well as the deliberate contrast between organic and inorganic animation.
CG Director: Shinya Takano
Interview originally published within Kizumonogatari’s Complete Guidebook, which you can buy over here.
Kizumonogatari’s CG backgrounds, made possible by the works that preceded it
— When did you get involved with Kizumonogatari?
Looking at my notes from the time, the first meeting was in August 2014. In terms of what we were doing at SHAFT at the time, we had just wrapped up Hanamonogatari and were about to start planning for Tsukimonogatari, which I was also going to work on at the same time. For Hanamonogatari and Madoka Magica, we had used 3DCG in certain places for the actual backgrounds, and not just as a reference for the layouts. It felt like we were just starting to branch out as a company in our 3D work. It’s possible that Tatsuya Oishi saw this trend when thinking about how to create 3DCG spaces for Kizumonogatari.
— The school gym that serves as the setting for the basketball scene at the end of Hanamonogatari is fully 3DCG if I recall.
That’s right. There had been times before where we’d selectively used 3DCG, but that scene was the first time at SHAFT that we challenged ourselves to make use of it wholesale like that. The initial meetings for Kizumonogatari took place immediately after we’d finished work on Hanamonogatari, and was still at the stage of ordering 3D models. I’d never have guessed the extent to which we’d end up using the CG.
— What kind of work were you doing at the start, then?
The idea at first was to just try making some models for the major locations using Nobuyuki Takeuchi’s background concept art as a base. That includes the cram school, the train station, and the road outside the school entrance. I didn’t think much of it when I started, but it turns out that there’s a lot more that goes into modeling buildings than I’d thought. I had to do some research to get the details down. And then Oishi and Takeuchi requested a lot of adjustments for things like the amount of detail or how spacious it should be. The level of expectations kept going up as we went iterating on the models meeting after meeting, much to my dismay (laughs).
— After all that, when did you decide to go with 3DCG for the backgrounds?
We decided to use 3DCG for the actual backgrounds after enlisting the help of Cyclone Graphics. I had begun to feel that it would be difficult to meet the standards for Kizumonogatari with only my abilities and our in-house manpower, which I discussed with the production assistants. I continued to work on the modeling by myself all through summer, and then as a joint effort with Cyclone starting that winter. We first decided to look through the storyboards and make clear which cuts we would be using CG for. After we understood the total amount of work involved, we figured out how we would split up the work between SHAFT and Cyclone.
— I hear the meetings going through the storyboards were quite lengthy, since you had to go through more than 4000 cuts, one at a time.
They really were. I remember the Cyclone folks coming out of the first one quite flustered (laughs).
— 4000 cuts would be roughly 1 cours’ worth for a TV anime.
On top of that, Kizumonogatari just has very involved camerawork. That means it’s worth it to use 3DCG for the backgrounds, although it’s not a must. Anime backgrounds are usually a single drawing for a single cut. When you introduce actual camerawork, the background has to move, and the number of drawings keeps going up depending on the length of said scene. This means that you have to think about how the camera moves in each of the 4000 cuts, instead of just stopping at that one number. And so we ended up with an even more daunting total number.
— In any case, it was an enormous amount.
We used traditional backgrounds for places where 3D wouldn’t make much of a difference, like short scenes or still shots, but it still ended up being a lot of 3D work. Even for cuts where you would normally use what we call a “book” (foreground layers in this case drawn by the background team) and slide it on the screen to mimic background movement, Oishi would often ask for us to play with the perspective. You can do some simple perspective shifts with just compositing, but when you take everything into account like the number of cuts in the scene and other technical aspects, we were really left with no choice but to do it in 3D.
— And you were the one to create a lot of the models that form the foundation of that 3DCG work.
It’s true that I created the models during that initial step before we entered full production mode, but considering the total amount, I wouldn’t go that far… Well, it still wasn’t a small number (laughs).
— You were the one who handled the big set pieces like the cram school and the train station.
When you put it that way, maybe it was a lot after all (laughs). However, for the buildings you’re talking about, the models I made early on weren’t ready to be used in the final product. For those models, I passed the work to Cyclone, who visually refined and finished everything. So my work that I remember best is in the shorter scenes, like Meme dashing through the industrial site or Koyomi running at breakneck speed.
— I heard you also worked on many of the props.
I did work on making a lot of those. We made the decision to produce as many of the props as possible in-house, since we didn’t want to spend even more time bringing those into the mix in our communications between the director and Cyclone.
— What was the reason for using CG for the props in the first place?
It was an unusual directive straight from Oishi. Apparently, it was because he wanted stiff objects to stay as such; “I want to make sure they actually look rigid.” Hand-drawn animation results in a characteristic fluidity, but that wasn’t what he wanted. So as a result, even the umbrella that Hanekawa uses in Reiketsu is CG. Inanimate objects should stay as inanimate objects, that was his train of thought. It seems like Oishi associates hand-drawn animation with living things. It’s a simple enough rule on the face of it, but as a result, we had to create CG models for things I would never have imagined.
— For example?
In Nekketsu, there are these baseballs that Koyomi throws at Dramaturgy. Halfway through, he switches to a shot put, and although you can barely see it because it’s covered in blood, Oishi still wanted it to be in 3DCG (laughs).
What we got out of working on Kizumonogatari
— Was there anything you tried to be mindful of as you worked alongside Cyclone?
As for the most difficult part of working on this production, it would have to be the pressure of having to keep up with all of the work that Cyclone was sending back. At the time, it was our first time cooperating with another company for the 3D work, and the software was brand new to us as well, so there were a lot of bumps along the road. But all of the work that Cyclone sent back our way was outstanding. I think that the tag team that is Tomohisa Shitara and Michiya Kato is the key to Cyclone’s strength. There’s a very close relationship between the 3D and the final compositing there. We’re too used to the ways of anime here, so I can’t help but approach 3D work as creating materials for a 2D animation context. So it was a fresh experience to see them incorporate it so closely with the compositing process, and I was honestly a bit jealous.
— How many people from SHAFT were involved in CG work for Kizumonogatari?
It had started off with two of us, but for Tekketsu, it ended up mostly being me by myself. For Nekketsu and Reiketsu, we were under a tight situation and schedule, and really did want to have more people on board. Since SHAFT was working on other titles at the same time, though, it ended up being a two-man effort.
— Which CG files were the largest in terms of file size? I hear the National Stadium was quite enormous.
The National Stadium was especially large with regard to the amount of data required. That being said, there were several other large models, such as the West Shinjuku area or the large school stairway. It inevitably ends up taking up a lot of space when you create models that will look good even at close-up sizes. But they were too big for practical usage, so we trimmed down the parts that we don’t use or aren’t visible on-screen in order to save on size.
— What problems do large models cause for practical usage?
It ends up being slower to work with, due to increased times for both data transfers and opening the files. However, we had the most difficulty not with file sizes but with the rendering stage. In order to get photorealistic backgrounds, we calculated the reflection of light with so-called global illumination when we render, which cost us a fair bit of time. The reflection of light causes bleeding of colors and brightness, but the calculation also covers areas not visible on screen. This means that taking away a single chair can change all the colors on screen, and rooms need to be properly walled in or else the brightness will be affected. That’s why depending on the cut, we’d turn off global illumination before rendering in order to speed up the processing. We ended up making various adjustments like that on a case-by-case basis.
— On the topic of adjustments, you mentioned that being able to change the perspective was one of the benefits of using 3DCG. What do you do when a scene calls for an extreme perspective? Do you make changes to the models?
That’s also on a case-by-case basis, and there are times when you’d alter the models themselves. That being said, the standard procedure was to play with the camera position and the lens as much as possible to try and match the storyboards and the animation. When that didn’t work, we’d bring in Oishi or the relevant unit director to help make direct adjustments to the camera. For buildings such as the National Stadium, there were plenty of times when they wouldn’t match the storyboards as a result of being properly modeled. However, it didn’t seem right to tamper with the architecture for Kizumonogatari, so we spent a lot of time searching for the right angles and lenses to match the cut as intended instead.
— Finally, how do you feel now that your work on the Kizumonogatari trilogy is over?
It was a very intense time from Tekketsu to the final cut of Reiketsu, during which I probably pushed myself to the absolute limit. In that respect, there’s a great sense of accomplishment in having seen things through to the end. On the other hand, I also finally feel free after having worked for such a long time, and I can look back with a sense that it’s all behind me now.
From that neutral perspective, I still think these are incredible movies. Before working on Kizumonogatari, I would never have dreamed of being able to do some things that now feel completely natural to me. I’m really glad to have had the chance to work on this series and learn all sorts of skills from everyone. I would like to thank Oishi, as well as Katou and Shitara from Cyclone Graphics, for all of their help. Shitara in particular ended up taking on work that I should have taken on myself, and came back with nothing but the highest quality. I strongly feel that his work helped me reach new heights as well. We would not have come this far with SHAFT alone, and these movies were made possible by the many hands involved.