Technological Development Project at Graphinica

Production Utilization of Style Transfer Technology

We have developed technologies to effectively utilize style transfer, a technique that has existed in academic research for many years, in real-world animation production. This effort led to the creation of an experimental animated film, which, despite being a 3D animation, incorporates 2D touches such as the nuances of watercolor and oil painting.

Summary

Graphinica pursues the fusion of art and technology, driving the development of techniques that enable diverse and creative anime production. In this R&D project, we focused on developing technologies to make a technique known as style transfer practical for use in real-world animation production. This enabled us to apply diverse 2D touches, such as watercolor and oil painting styles, to 3D animations, making it possible to create various styles of visuals that were previously difficult to achieve. Utilizing the outcomes of this technical development, we produced the experimental animated film 'Forest Tale' and authored an academic paper, which was presented at SIGGRAPH Asia. Moving forward, we will continue to support even more creative anime production by developing technologies that maximize artists' creativity.

© ARCH · Graphinica · Salamander Pictures

Experimental Animated Film

We produced the experimental animated film 'Forest Tale' using the outcomes of this technical development.

Technology and Academic Publication

The technology we employed is based on classical texture synthesis techniques. This approach does not rely on machine learning, providing an environment that is both artist-friendly and easy to control. For further details, please refer to our paper (accepted at SIGGRAPH Asia 2024 Technical Communications).

Paper Title

A Practical Style Transfer Pipeline for 3D Animation: Insights from Production R&D

Paper Authors

Paper Abstract

Our animation studio has developed a practical style transfer pipeline for creating stylized 3D animation, which is suitable for complex real-world production. This paper presents the insights from our development process, where we explored various options to balance quality, artist control, and workload, leading to several key decisions. For example, we chose patch-based texture synthesis over machine learning for better control and to avoid training data issues. We also addressed specifying style exemplars, managing multiple colors within a scene, controlling outlines and shadows, and reducing temporal noise. These insights were used to further refine our pipeline, ultimately enabling us to produce an experimental short film showcasing various styles.

Download

Supplemental Video

Talks

Publications

Media

Awards