ragavsachdeva/magiv3

From Panels to Prose

Generating Literary Narratives from Comics

Ragav Sachdeva and Andrew Zisserman

University of Oxford

 

Magiv3 is a unified vision-language model for comic understanding. Given an image and a task prompt, it autoregressively produces text-only outputs for a range of tasks: localising panels, characters, texts and speech-bubble tails, performing OCR, and grounding characters referenced in a caption to their locations in the panel.

Usage

model = AutoModelForCausalLM.from_pretrained("ragavsachdeva/magiv3", torch_dtype=torch.float16, trust_remote_code=True).cuda().eval()
processor = AutoProcessor.from_pretrained("ragavsachdeva/magiv3", trust_remote_code=True)

model.predict_detections_and_associations(images, processor)
model.predict_ocr(images, processor)
model.predict_character_grounding(images, captions, processor)

License and Citation

The provided model and datasets are available for unrestricted use in personal, research, non-commercial, and not-for-profit endeavors. For any other usage scenarios, kindly contact me via email, providing a detailed description of your requirements, to establish a tailored licensing arrangement. My contact information can be found on my website: ragavsachdeva [dot] github [dot] io

@InProceedings{Sachdeva25,
      title={From Panels to Prose: Generating Literary Narratives from Comics},
      author={Ragav Sachdeva and Andrew Zisserman},
      booktitle={IEEE International Conference on Computer Vision (ICCV)},
      year={2025},
      eprint={2503.23344},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.23344}
}