Due to the abstract nature of artworks such as anime paintings, it is difficult to generate high-quality animation portraits that match the user's intentions even by simply using generation AI.In particular, rough sketches by general users lack the overall balance and detailed information of each part during production, making it even more challenging to incorporate generative AI methods.

　The research team has developed a new generative artificial intelligence (generative AI) technology that simulates the way lines are drawn in the production process and generates high-quality animation images from rough sketches.Based on a deep learning framework, this technology searches the latent space (low-dimensional space in which the features of the learned image are distributed) in the creative process and generative AI.

　In addition, based on the proposed framework, we developed a user interface that has a function to display guidance (a rendering of the finished animation) on the screen during the production process.This makes it possible to consistently generate high-quality animation images throughout the production process without depending on the drawing order of the sketches.

　We asked 15 graduate students to evaluate our proposed generative AI technology through an animated portrait production interface.As a result, it was highly evaluated in terms of satisfaction with the generated results and the matching rate between the user's line drawing and the generated guidance.

　In the long term, the results of this research will contribute to the popularization of generative AI technology, supporting users' creative activities, eliminating skill barriers tied to physicality, and maximizing human creative capacity. It is hoped that this will be extended to the limit.

Paper information:[ACM SIGGRAPH 2023] AniFaceDrawing: Anime Portrait Exploration during Your Sketching