Article: Stylised 2D animation generation method based on generative adversarial networks Journal: International Journal of Arts and Technology (IJART) 2026 Vol.16 No.5 pp.1 - 17 Abstract: Conventional 2D animation stylisation relies heavily on manual production, resulting in long cycles, high costs, and inconsistent styles across different artists. To address these limitations, this paper proposes a generative adversarial network (GAN)-based method for automatic stylised 2D animation generation. Style-conditional encoding is introduced to guide frame-wise generation toward target styles, while optical flow constraints and an inter-frame discriminator are applied to maintain motion continuity and style consistency across frames. In addition, multi-scale convolutional modules are integrated into the generator to enhance the representation of fine-grained details. Experimental results demonstrate that the proposed method achieves superior performance in style similarity (0.93), optical flow error (0.124), and structural similarity (0.912), effectively improving visual quality, temporal coherence, and detail representation in stylised 2D animation generation. Inderscience Publishers - linking academia, business and industry through research

Title: Stylised 2D animation generation method based on generative adversarial networks

Authors: Zhu Zeng

Addresses: Department of Digital Media Art, School of Art, Beijing Union University, Beijing, 102200, China

Abstract: Conventional 2D animation stylisation relies heavily on manual production, resulting in long cycles, high costs, and inconsistent styles across different artists. To address these limitations, this paper proposes a generative adversarial network (GAN)-based method for automatic stylised 2D animation generation. Style-conditional encoding is introduced to guide frame-wise generation toward target styles, while optical flow constraints and an inter-frame discriminator are applied to maintain motion continuity and style consistency across frames. In addition, multi-scale convolutional modules are integrated into the generator to enhance the representation of fine-grained details. Experimental results demonstrate that the proposed method achieves superior performance in style similarity (0.93), optical flow error (0.124), and structural similarity (0.912), effectively improving visual quality, temporal coherence, and detail representation in stylised 2D animation generation.

Keywords: generative adversarial networks; GAN; stylised animation; frame consistency; style fidelity; multi-scale detail.

DOI: 10.1504/IJART.2026.153269

International Journal of Arts and Technology, 2026 Vol.16 No.5, pp.1 - 17

Received: 09 Dec 2025
Accepted: 16 Feb 2026
Published online: 29 Apr 2026 *

Title: Stylised 2D animation generation method based on generative adversarial networks

Keep up-to-date