SVGDreamer is a CVPR 2024 paper implementing a diffusion-based approach for text-guided SVG generation. It targets researchers and artists seeking to synthesize high-quality vector graphics from textual descriptions, offering control over style and editing capabilities.

How It Works

SVGDreamer utilizes a diffusion model to generate SVG paths. It employs a two-stage process: first, a Sketch-Inference-and-Vector-Editing (SIVE) stage for initial shape generation and refinement, followed by a Vector-SVG-Path-Diffusion (VPSD) stage to produce the final SVG output. This approach aims to balance synthesis quality with vector graphic editing potential.

Quick Start & Requirements

• Installation: Run bash script/install.sh or use the provided Docker script bash script/run_svgdreamer_docker.sh.
• Prerequisites: Requires a pretrained Stable Diffusion model (e.g., Stable Diffusion 2.1 Base). The model can be auto-downloaded by setting diffuser.download=True in conf/config.yaml.
• Resources: enable_xformers=True is recommended for faster optimization. state.mprec='fp16' can reduce GPU memory usage.
• Documentation: Examples.md

Highlighted Details

• Supports multiple generation styles including iconography, painting, pixel art, low-poly, sketch, and ink/wash.
• Offers control over generation through various parameters like skip_sive, token_ind, result_path, and x.vpsd.t_schedule.
• Includes a newer version, SVGDreamer++, with enhanced visual representation and editing capabilities.

Maintenance & Community

The project is associated with the CVPR 2024 paper. Links to community channels or roadmaps are not explicitly provided in the README.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive license suitable for commercial use and integration with closed-source projects.

Limitations & Caveats

The README mentions a "TODO" list, indicating ongoing development. Specific limitations or known bugs are not detailed.