ECCV2026 Accepted: Semantic Line Diffusion

We are pleased to announce that our paper, “Semantic Line Diffusion: Character-Consistent Line Art from Text-Annotated Storyboards,” has been accepted to the European Conference on Computer Vision 2026 (ECCV 2026).

This work presents PanelDiff, a diffusion transformer framework for generating clean, character-consistent line art from rough, text-annotated storyboard sketches. The framework combines multi-reference character conditioning, a memory-augmented representation bank, and a panel-aware diffusion transformer to improve line-art fidelity, identity preservation, and cross-panel consistency.

Authors:

• Seo-Yeon Choi
• Kyungsu Lee (corresponding author)

This paper was accepted to a Top BK/CS conference. Congratulations to the authors on this meaningful contribution to character-consistent generative modeling for webtoon and storyboard production.

Related link:

• Semantic Line Diffusion: Character-Consistent Line Art from Text-Annotated Storyboards