ToonOut Model Weights

Please check out:

• our repository: https://github.com/MatteoKartoon/BiRefNet
• our paper: ToonOut: Fine-tuned Background Removal for Anime Characters
• the dataset we collected to fine-tune this model
• Replicate endpoint to try out the model (courtesy of J. H. Cho @sprited)

Model Summary

ToonOut is a fine-tuned variant of BiRefNet specialized for background removal in anime-style images.
BiRefNet performs strongly on realistic imagery but struggles with stylized content (e.g., hair wisps, line art, transparency).
Fine-tuned on the ToonOut Dataset (1,228 images), ToonOut delivers a notable boost for anime segmentation:

• Pixel Accuracy: 95.3% → 99.5% (on our test set)

---

Model Details

• Architecture: BiRefNet (fine-tuned)
• License: MIT
• Training data: ToonOut Dataset (CC-BY 4.0)

---

Example usage

Please refer to the demo notebook from our GitHub repo.

---

Citation

If you use ToonOut, please cite:

@misc{muratori2025toonout,
  title={ToonOut: Fine-tuned Background Removal for Anime Characters},
  author={Muratori, Matteo and Seytre, Joël},
  year={2025},
  eprint={2509.06839},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2509.06839},
  doi={10.48550/arXiv.2509.06839}
}

---

Authors & Contact

• Authors: Matteo Muratori (University of Bologna, Kartoon AI), Joël Seytre (Kartoon AI)
• Contact: joel@kartoon.ai, matteo.muratori9@studio.unibo.it

⸻

Project by Kartoon AI, powering toongether, check us out at kartoon.ai & toongether.ai