Application of Comics Generation Techniques in the Visualization of Vietnamese Folk Tales

In recent years, the integration of Artificial Intelligence into tasks requiring creative cognition has become increasingly prevalent. Although the nature of machine creativity remains a subject of debate, it is undeniable that these models have exerted a positive influence, ranging from the reduction of lead times, significant cost optimizations for enterprises to versatile support in day-to-day activities. This study focuses on the generation of Vietnamese folk comics utilizing the Story-Iter architecture, which possesses an inherent advantage in maintaining contextual consistency via historical image references. Furthermore, we propose an enhancement involving the selective curation of reference image sets rather than incorporating the entire dataset or a fixed sequence of recent frames, thereby mitigating the integration of noise during the generation of novel imagery. This strategy effectively minimizes the inclusion of noisy data during the synthesis of new images. Experimental results performed on our self-built small-sized story dataset demonstrate that the approach yields efficacy when synthesized with the original Story-Iter framework, while simultaneously highlighting persistent challenges.