Synthesizing Coherent Story with Auto-Regressive Latent Diffusion Models
This addresses the need for coherent image series in story-telling applications, representing a novel extension of diffusion models beyond single-image synthesis.
The paper tackles the problem of generating coherent visual stories by proposing AR-LDM, an auto-regressive latent diffusion model conditioned on history captions and images, achieving state-of-the-art FID scores on datasets like PororoSV, FlintstonesSV, and VIST, and superior performance in human evaluations for quality, relevance, and consistency.
Conditioned diffusion models have demonstrated state-of-the-art text-to-image synthesis capacity. Recently, most works focus on synthesizing independent images; While for real-world applications, it is common and necessary to generate a series of coherent images for story-stelling. In this work, we mainly focus on story visualization and continuation tasks and propose AR-LDM, a latent diffusion model auto-regressively conditioned on history captions and generated images. Moreover, AR-LDM can generalize to new characters through adaptation. To our best knowledge, this is the first work successfully leveraging diffusion models for coherent visual story synthesizing. Quantitative results show that AR-LDM achieves SoTA FID scores on PororoSV, FlintstonesSV, and the newly introduced challenging dataset VIST containing natural images. Large-scale human evaluations show that AR-LDM has superior performance in terms of quality, relevance, and consistency.