VectorFusion: Text-to-SVG by Abstracting Pixel-Based Diffusion Models

Ajay Jain; Amber Xie; Pieter Abbeel

doi:10.1109/cvpr52729.2023.00190

CVPR2023顶会

VectorFusion: Text-to-SVG by Abstracting Pixel-Based Diffusion Models

Ajay Jain, Amber Xie, Pieter Abbeel

DOI 出版方

2023年份

62顶会引用

摘要

Figure 1. Text-to-SVG with VectorFusion. When (a) raster graphics sampled from Stable Diffusion are (b) auto-traced, they lose details that are hard to represent within the constraints of the abstraction. (c-d) VectorFusion improves fidelity and consistency with the caption by directly optimizing paths with a distillation-based diffusion loss. Find videos and more results at https://ajayj.com/vectorfusion .