This repository provides a comprehensive toolkit for generating and editing Scalable Vector Graphics (SVG) using neural networks and differentiable rendering techniques. It targets researchers and developers interested in text-to-SVG synthesis, image vectorization, and SVG manipulation, offering a unified interface to various state-of-the-art methods.

How It Works

The library leverages differentiable rasterization (e.g., DiffVG) and Score Distillation Sampling (SDS) from diffusion models. SDS allows optimization of SVG parameters (like paths, colors, and shapes) to match text prompts without direct image supervision, enabling text-guided SVG generation and editing.

Quick Start & Requirements

• Installation:
  • Standard: chmod +x script/install.sh && bash script/install.sh
  • Docker: chmod +x script/run_docker.sh && sudo bash script/run_docker.sh
• Prerequisites: Python, PyTorch. Some models require Hugging Face diffusers and may need xformers installed via pip install --pre -U xformers. U2Net model download is required for CLIPasso and CLIPascene.
• Documentation: pytorch-svgrender.readthedocs.io

Highlighted Details

• Supports multiple SVG generation methods: DiffVG, LIVE, CLIPasso, CLIPascene, CLIPDraw, StyleCLIPDraw, CLIPFont, VectorFusion, DiffSketcher, SVGDreamer, Word-As-Image.
• Integrates various styles for text-to-SVG: Iconography, Sketch, Pixel Art, Low-Poly, Painting, Ink & Wash.
• Employs Score Distillation Sampling (SDS) for text-to-SVG without explicit image supervision.
• Includes methods for image vectorization and semantic sketching.

Maintenance & Community

The project is actively maintained, with recent updates in December 2023. Contributions are welcomed via pull requests.

Licensing & Compatibility

• License: Mozilla Public License Version 2.0.
• Compatibility: Generally compatible with commercial use, but users should review the MPL 2.0 terms.

Limitations & Caveats

The project relies heavily on external models and libraries, which may introduce complex dependency management. Some advanced features might require significant computational resources (GPU, VRAM). A TODO list indicates planned features like an SVG Layout Pipeline.