AIGC and Graph Neural Network Fusion for Intelligent Generation and Dynamic Processing of Animated Scenes

Traditional animation scene generation methods rely on manual design, which is inefficient and difficult to meet the rapid demand for high-quality content in the modern animation industry.The rise of artificial intelligence content generation technology brings new opportunities for animation production, and graph neural networks show strong advantages in processing complex relational data.In this paper, we propose an animation complex scene generation and dynamic processing method that combines AIGC and graph neural networks.The method constructs a scene graph generation model based on two-layer graph convolutional network, divides the scene objects into primary and secondary objects through the object layering module, and fuses the visual, spatial and semantic features using the two-layer feature extraction module to realize the intelligent generation and optimized processing of complex animation scenes.The study is validated on a self-built ASG dataset, which contains 52k labeled animated scene images, covering 154 object classes and 70 relation classes.The experimental results show that the proposed method achieves an accuracy rate of 95.94% in the scene graph recognition task, which is 7.08 percentage points higher than that of the recurrent consistent generation adversarial network method; in terms of the animated scene generation rate, the rate of generating 100 images is 13.78 min per image, which is significantly better than that of the comparison method; in the evaluation of the effectiveness of the deblurring process, the PSNR value on the validation set reaches 35.44 dB, and the SSIM value is 0.923.The study shows that the method effectively improves the generation quality and processing efficiency of complex scenes in animation production, and provides technical support for the intelligent development of animation industry.