Visual Writing Prompts: Character-Grounded Story Generation with Curated Image Sequences

Abstract Current work on image-based story generation suffers from the fact that existing image sequence collections do not have coherent plots behind them. We improve visual by producing a new image-grounded dataset, Visual Writing Prompts (VWP). VWP contains almost 2K selected sequences of movie shots, each including 5-10 images. The are aligned with total 12K stories which were collected via crowdsourcing given and set grounded characters corresponding sequence. Our collection filtering process has allowed us to obtain more coherent, diverse, visually compared previous work. also propose character-based model driven coherence as strong baseline. Evaluations show our generated grounded, diverse than current state-of-the-art model. code, features, annotations available at https://vwprompt.github.io/.