COMICS Dataset

The COMICS dataset comprises over 1.2 million comic‑book panels (≈120 GB) extracted from a wide range of mainstream titles, each paired with automatically generated textbox transcriptions and metadata about panel order. It is designed for multimodal narrative understanding, providing three cloze‑style tasks that require models to predict narrative or character information for a target panel given the preceding n panels as context, thereby testing the integration of visual and textual cues.

Please cite the following if you use the data:

Iyyer, M., Manjunatha, V., Guha, A., Vyas, Y., Boyd‑Graber, J., Daumé III, H., & Davis, L. (2017). The Amazing Mysteries of the Gutter: Drawing Inferences Between Panels in Comic Book Narratives. In IEEE Conference on Computer Vision and Pattern Recognition.