Animation generation of traditional ethnic elements based on memory-enhanced self-supervised networks

Aiming at the problems of cultural distortion and action homogenisation in traditional ethnic animation generation, this paper proposes a memory-enhanced self-supervised network and AIGC fusion framework.First, a dual-channel memory module is constructed to decouple ethnic visual patterns and semantic features.Second, a culturally constrained mixed density network (MDN) is designed to generate probabilistic, compliant and diverse action sequences based on the learned features to effectively overcome the singularisation problem.Finally, dynamic symbol implantation pipeline is developed to realise high-fidelity and controllable animation synthesis of ethnic elements.Leveraging self-supervised learning on unpaired ethnic images, the framework achieves 89.7% cultural compliance along with enhanced action diversity.Generation efficiency reaches 25-FPS real-time rendering at 0.48-kWh energy consumption, and spatiotemporal synchronisation attains 0.12 s latency.Experiments confirm significant improvements in cultural fidelity, action diversity, and efficiency, establishing a new paradigm for digital preservation of intangible cultural heritage.