Emilio Ferrara’s 9 October PAGE preprint studies presidential portrayals in American comics, mainly from 1938–1956. Among references drawing sitting presidents, 57% withhold the face (n=116), against 3% for historical presidents (n=143).[1]
Claude Opus 5.5 coded silhouettes, shadows and views from behind as withheld faces. These codes have no second coder yet. Historical references include stamps, statues and portraits, whose faces appear by convention.[1]
The material comes from COMICS, introduced by Mohit Iyyer and colleagues in 2017. Its creators downloaded scanned comics, trained detectors using manually marked panels and text boxes, then transcribed the lettering. The release keeps page images alongside those transcriptions.[2][3]
The original research tested whether a model could follow a narrative across panels. One task blanked a text box and asked it to choose the missing text from candidates using the surrounding pictures and text. That design treated the drawing as information a transcript alone would lose.[2]
Searching transcribed names and titles selected candidates; the model examined 1,082 pages. An unnamed likeness could still be missed unless another presidential term brought it into that set.[1]
Printed names alone cannot express this difference.[1]
Sources & further reading
- Studying Comics at Scale with Artificial Intelligence: The PAGE Framework and Three Studies of American Comics and Japanese Manga arXiv / Emilio Ferrara · 9 Oct 2026 · Accessed 20 Oct 2026
- The Amazing Mysteries of the Gutter: Drawing Inferences Between Panels in Comic Book Narratives Iyyer and colleagues / CVPR · Accessed 20 Oct 2026
- COMICS data downloads University of Maryland · Accessed 20 Oct 2026