A review of Cel-look style basis until 2017

(Polygon Pictures Inc. / Studio Phones)
This material is written based on content discussed in previously held sessions. We plan on brushing up the content of this material and making revisions as we hold future sessions.
translated by PPI Translation Team


While our studio was gradually starting to implement methods to replicate the Japanese 2D cel-look anime visual style using CG tools from around 2000, initially this was limited to using CG only partially, with characters or other important objects typically being drawn by hand. From around 2010, as the number of fully CG animated titles using cel shading increased, the field known as “cel-look CG” was born. With regards to this cel-look CG, various studios have been developing their own unique methods and workflow, while still respecting the traditional Japanese anime style, and continue to do so until today. In terms of methodology, while each studio has its own idiosyncrasies, I would like to look back on the period up to 2017 with regards to the fundamental elements of a common workflow which could be considered characteristic to this style. Screenshots taken from productions at Polygon Pictures have been prepared for this document; as such we accept there may be some bias in our writing but we ask for your understanding.
Also, regarding pipeline and infrastructure aspects of cel-look CG animation please see the separate related document, "Cel-look CG pipeline and issues".

Cel-look CG pipeline and issues

The possibility of cel-look pipeline and challenges to take advantage of the hand-drawn animation

■Color design

Color design or color specification was a workflow phase in the cel animation era when animated characters, vehicles, props, etc., would be drawn on 2D cels. The colors to be used in order to match the style and direction of a scene would be decided, and then a color palette created which would be used during the coloring phase of the workflow.
This role and work phase has been carried on by many studios in their transition to CG, where the color palettes created during the color design/color specification phase are used as a base during the rendering process, so that the colors specified in the color palette are those seen in the rendered image.
Also, in most cases the color palette will need to be adjusted to the location of the story or content of the scene, and in the course of a single TV series this may necessitate the creation of several hundred color palettes.
Standard CG production utilizes color correction processes, however as cels would be colored one at a time, all using the same colors, retaining the color palette system could be seen as one unique characteristic of the cel-look CG animation workflow.
The issues with color design from an infrastructure standpoint are in the design of the communication model, such as tools to help share the created color palettes among the staff and other communication tools, and these

A color palette created during the color design phase.

© Gamon Sakurai, KODANSHA/AJIN Production Committee. All Rights Reserved.

A color palette such as the example above is loaded as parameters into the shader in 3D CG, and then the image is rendered.
The palette numbers and cel areas are mapped and cel colors decided using color values based on the palette numbers. The pipeline is set up so that, depending on the scene and lighting, cel colors can be automatically changed by switching the palette. While there may be some slight differences in specifications, I believe that many cases of cel shading using color palettes will utilize a similar system, including the use of the same commercial software.

Parameter conversion from the color palette to the shader (in the case of Polygon Pictures’ shader)

© Gamon Sakurai, KODANSHA/AJIN Production Committee. All Rights Reserved.

■Matte painting

In the same manner as traditional cel animation, in cel-look CG animation the standard method to create background is 2D.
Even in cases where the background are created in 3D, the textures for these frequently use texture mapping based on a painted image.
Before the background are created, the area which the background must cover needs to be conveyed to the background artist, and the data which forms the “blueprint” for this is known as a layout diagram (or genzu, in Japanese).
This background layout stage is extremely important to create dynamic camerawork using 3D camera movements, and in many cases the image created from this diagram can be used with projection mapping to create a “2.5D” composite.

Reasons often cited for using these painted background are the costs of 3D asset creation and the necessity of 2D background for imitating some of the characteristic visual patterns of cel animation. In recent years technology has been being developed to create background images or even 3D assets automatically from photographs, but there are still areas, including stylistic ones, which cannot be fully automated, and while efficiency is always sought after, painting these background by hand continues to be the standard.
While I believe the demand for this background-style of matte painting will continue in future, it can also be envisaged that this area will be subject to some interest in the implementation of processes to automate background creation and painting. Also, while management tools for highly communicative art direction such as brush management and color palette integration are not yet at a point where we can say they are being actively implemented, going forward I believe it will be necessary to look into development of tools to seamlessly link art direction and the actual art tasks while feeling out the possibilities relating to workflow on the server side.

A painted background

© Gamon Sakurai, KODANSHA/AJIN Production Committee. All Rights Reserved.

Image showing the area which needs to be painted for the background, generated from the DCC tool’s camera

The final layout diagram (genzu) including notes to the background artist.

The finished background.

© Gamon Sakurai, KODANSHA/AJIN Production Committee. All Rights Reserved.

■Limited animation

Perhaps the most important, and difficult to achieve in CG, aspect of the cel-look style is the animation. Japanese cel animation is not animated realistically, with movement on every frame, but in a simplified manner where the movement occurs on every second or third frame. This method was originally employed to reduce the number of cels to be drawn and thereby reduce costs; however this characteristic animation style has been developed over a long history, and known now as “limited animation” it is well-established as the style of Japanese animation.
This style of animation is ill-suited to CG animation in which keyframes are automatically interpolated, however the solution is also not as straightforward as simply dropping (or “stepping”) every other frame. Each studio has been refining its own methods to replicate this hand-drawn animation style through a process of trial-and-error.
One area consistent across all studios is likely to be the use of tools to support the process, however in most cases there are still areas that depend heavily on the animator’s skill, and will have a large impact on the quality of the finished title.
Also, in cases where animation curves are stepped in order to drop frames, this can cause issues in the rendering process for phases which require uninterrupted data for each frame, such as simulation and motion blur. Therefore, a pipeline which is able to handle animation data in both stepped and unstepped forms is required.

A support tool for limited animation.


In order to replicate the look of cel animation, in which certain areas are filled with a certain color, specialized cel-look shaders are utilized.
Another necessary component of the cel shaded-look is the outlines, which are typically generated at the same time as the rendering of the image. The rendering of these outlines is extremely important to the cel shaded CG look in order to maintain the feeling of hand-drawn animation, and in order to keep the numerous variations these can have, a renderer which is both flexible and feature-rich is a necessity. This is an area in which all studios struggle, and continually attempt to refine as it is also an area in which the studio’s individualized look can be showcased.
Also, when using ray tracing-based renderers to generate outlines, there are issues with outline accuracy and outlines flickering between frames. With commercial renderers, as the inner workings of the software are not usually available to interact with from the user side; i.e., full access to the source code and not simply via an API, this can be a limitation on their use. Among the issues to be tackled going forward are infrastructure setup transitions and approaches to long-term operation, including the long-term use of the data for the various kinds of materials.

The basic setup for outlines and cel shading

■Lighting and Compositing

In the case of standard CG production, the light’s energy is calculated physically from the positions of the light sources, etc., and how this is affected by an object’s color or surface material; however in cel-look CG, as the output color is determined by the color palette, the main function of the rendering process is to define the area of the cel and area of shadow, and assign the colors from the color palette. At Polygon Pictures when we first began working with cel-look CG, we used shaders that had been developed to integrate color palettes, however the workflow needed to support this became too complex and drove up costs, and therefore we currently do not use the color palette system when rendering, and instead have implemented a solution giving similar results in the compositing phase instead, making for a much simpler workflow.

For this reason, the information handled at the lighting stage is mostly simple, generally using a light called a “directional light” which maintains the light in a certain direction. Also, as there are many cases where this simple CG render is not sufficient to create the lit/shadow area shapes desired, in these cases various processes are added to match the cel anime style, such as editing the object normals or even editing the geometry of the object itself.
Anime-style post treatments are also added to the rendered, composited images. In the cel anime era this was known as satsuei (“shooting” or “cinematography”, “compositing”) in Japanese, but many of the traditional techniques still exist. Making use of the nature of CG, layers with many kinds of information can be output when rendering, which can then be used during compositing to create a wider range of visuals, as well as make use of 3D cameras with 2.5D-style compositing to create shots with dynamic camerawork. Also, the painted background mentioned previously are composited with the 3D renders at this stage.

In the cel anime era, this satsuei compositing stage was limited to working with the finished cels, but in CG animation, as the lighting and compositing phases are closely linked, in many cases these are treated as the same phase. This means that thought is generally given to the balance of the processes to be done in rendering, and work to be done in compositing, and adjustments made accordingly. However, when a renderer needs to output numerous buffers this leads to issues with increased file sizes and slower responses from the software being used for compositing. As this approach places increasing performance demands on the storage, I believe we will need to also consider approaches which do not rely so much on buffer renders, or look into options further on the storage side.

Buffer images output during rendering, for use in compositing

© Gamon Sakurai, KODANSHA/AJIN Production Committee. All Rights Reserved.