Diffusion model method improves story image sequences with adaptive scenes
Not Just a Subject: Capturing Adaptive Scene-Specific Variance for Visual Storytelling
Computer Vision and Pattern Recognition
Summary
Generating a series of pictures from a story is tricky because the images need to keep characters looking the same while changing their surroundings naturally. The authors propose EOS-Story, a way to keep both the character’s identity and the changing scene details balanced without extra training. They split the story into parts about who is in the image and where they are, then carefully mix appearance details so scenes fit better. Their approach also fixes scale problems in images that caused weird shapes before. Tests show their method helps make more coherent and flexible visual stories.
What this means in practice
- •For graphic designers: Produce consistent character images that adapt naturally to varying scene layouts for illustrated narratives.
- •For game developers: Build dynamic story-driven visuals where characters keep their look but move through diverse backgrounds.
- •For advertising teams: Create sequential image ads maintaining brand character identity while changing scene settings creatively.$Commercial implications: Enables design of tailored ad sequences with consistent brand characters fitting varied scenarios, improving engagement.
Authors
SeungJu Cha, Ye-Chan Kim, Kwanyoung Lee, Dong-Jin Kim
Abstract
Recent advancements in diffusion models have significantly improved text-to-image generation, enabling their application to visual storytelling, where coherent image sequences are generated from narrative prompts. However, existing methods primarily focus on identity preservation and often fail to adapt to diverse scene variations, resulting in semantic misalignment and structural inflexibility. We propose EOS-Story, a training-free framework that harmonizes identity consistency with scene-specific adaptation across text and vision spaces. In the text space, we decouple the narrative into identity and scene sub-prompts to prevent information dilution, but this leaves the scene without access to appearance cues. To bridge this gap, we introduce Cross-Contextual Bridging, which softly injects a phrase-level appearance residual into the scene's <EOS> embedding. In the vision space, we find that scale mismatch between identity and scene subjects causes a few identity tokens to be repeatedly transferred, producing structural artifacts. We address this with Adaptive Identity Preservation, which anchors identity features at the scene-specific scale (Identity Anchoring) and integrates them via Identity-Aware Self-Attention. To evaluate performance under varied spatial compositions, we extend the Consistory+ benchmark with diverse framing scenarios (Consistory+_+). Extensive experiments show that EOS-Story achieves competitive or superior performance on both benchmarks.