VAnim: Rendering-Aware Sparse State Modeling for Structure-Preserving Vector Animation

Scalable Vector Graphics (SVG) animation generation is pivotal for professional design due to their structural editability and resolution independence. However, this task remains challenging as it requires bridging discrete code representations with continuous visual dynamics. Existing optimization-based methods often destroy topological consistency, while general-purpose LLMs rely on rigid CSS/SMIL transformations, failing to model geometry-level non-rigid deformations. To address these limitations, we present VAnim, the first LLM-based framework for open-domain text-to-SVG animation. We reconceptualize animation not as sequence generation, but as Sparse State Updates (SSU) on a persistent SVG DOM tree. This paradigm compresses sequence length by over 9.8x while mathematically guaranteeing topological isomorphism and identity persistence. To enable precise control, we propose an Identification-First Motion Planning mechanism that grounds textual instructions in explicit visual entities. Furthermore, to overcome the non-differentiable nature of SVG rendering, we employ Rendering-Aware Reinforcement Learning via Group Relative Policy Optimization (GRPO). By leveraging a hybrid reward from a state-of-the-art video perception encoder, we align discrete code updates with high-fidelity visual feedback. We also introduce SVGAnim-134k, the first benchmark for vector animation. Extensive experiments demonstrate that VAnim significantly outperforms state-of-the-art baselines in semantic alignment and structural validity, demonstrating a robust capacity for synthesizing high-fidelity non-rigid deformations without requiring explicit temporal consistency constraints.

Vector graphics are widely used for icons, websites, and user interfaces because they stay sharp at any size and can be easily edited. However, creating smooth animations for these graphics is still mostly a manual process, especially when objects need to bend, morph, or move in coordinated ways. Existing AI tools either generate ordinary videos that are hard to edit, or change SVG code in ways that can break shapes, colors, and object identities. We introduce VAnim, an AI system that turns a static SVG image and a text instruction into an editable vector animation. Instead of redrawing the whole image frame by frame, VAnim keeps the original drawing structure and updates only the small set of parts that need to move. It first identifies which visual objects correspond to the user’s instruction, then plans and generates their motion. We also build SVGAnim-134k, a large dataset of professional vector animations, and train VAnim using feedback from rendered videos. Our results show that VAnim better follows user instructions while preserving the original structure of the drawing. This could help designers and developers create high-quality, editable animations more quickly.