Reproduce “Dynamic Typography: Bringing Text to Life via Video Diffusion Prior” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zichen Liu, Yihao Meng, Hao Ouyang, Yue Yu, Bolin Zhao, Daniel Cohen-Or, Huamin Qu
Researchers from Hong Kong University of Science and Technology, Ant Group, and Tel-Aviv University developed "Dynamic Typography," a method for automating the creation of semantically-aware, dynamic, and legible vectorized text animations. The approach transforms static letter representations into motion-infused videos that align with user prompts, achieving superior legibility and temporal consistency compared to existing pixel-based and vector-based animation techniques.
Key insights
- Directly optimizing vector paths with a video diffusion prior, rather than relying on multi-stage pixel-based approaches, effectively addresses text readability challenges during animation.
- Decomposing text deformation into a static 'base shape' and a dynamic 'motion field' allows for separate yet coordinated control over semantic meaning and movement.
- Explicit regularizations, including perceptual legibility and a novel mesh-based structure preservation, are crucial to enforce letter legibility and temporal consistency in dynamic vector typography, compensating for the limitations of general video generation models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.