Research on Anime-Style Image Generation Based on Stable Diffusion

Hubei University of Education, Wuhan, Hubei Province, 430000, China

^* Corresponding author: Zhangailin@asu.edu.pl

Abstract

Animation style generation technology based on artificial intelligence is gaining increasing attention in the current society, and it has shown a wide range of application prospects in creative design, game development, and advertising. Based on the Animefull-follow pre-training dataset, the effect of using Stable Diffusion technology combined with LoRA model to generate a single anime character image was discussed. The experimental results show that the generated image is highly consistent with the original image in terms of style and detail, and successfully captures the unique characteristics of animation art. Although the Fréchet Inception Distance (FID) value of the generated image is 77. 29 when calculating the FID value, indicating that there are visual differences to a certain extent, these differences usually do not significantly affect the user's perception in the animation field, and the generated images still show excellent visual effects. Combined with the advantages of the LoRA model, the resources and time required for training are significantly reduced, enabling high- quality image generation even in resource-constrained environments.