StarVector: Basic model for generating SVG vector graphics from images and text

General Introduction

StarVector is an open source project created by developers such as Juan A. Rodriguez to convert images and text into Scalable Vector Graphics (SVG). This tool uses a visual language model that understands image content and textual instructions to generate high-quality SVG code. Its core feature is that it turns the process of vectorization into a code generation task, making it suitable for icons, technical diagrams, logos, etc. StarVector is trained on the SVG-Stack dataset, which contains more than 2 million real SVG samples, ensuring accurate and varied results. The project is currently supported by ServiceNow Research and Mila, among others, and was last updated in March 2025.

Function List

• Generate SVG from image: Input pixel image and generate corresponding vector graphics code.
• Generate SVG from text: Create brand new SVG graphics based on text descriptions.
• Supports a wide range of graphic elements: SVG basics such as paths, ellipses, polygons and text.
• High-precision vectorization: The generated SVG maintains details and is suitable for zoom-in use.
• Open source and free of charge: the code is open to the public, users can freely download and modify.

Using Help

The use of StarVector requires some technical foundation as it is a code-based program. Below are detailed installation and operation steps to help users get started quickly.

Installation process

1. Preparing the environment
   • Make sure you have Python 3.11.3 installed on your computer. you can check the version with the command:

     python --version
     

   • If not, download and install it from https://www.python.org.
   • Install Git for downloading code from GitHub. Download it at https://git-scm.com.
2. Creating a Virtual Environment
   • Open a terminal and enter the following command to create a new Python environment:

     conda create -n starvector python=3.11.3 -y
     

   • Activate the environment:

     conda activate starvector
     

3. Download StarVector Code
   • Enter it in the terminal:

     git clone https://github.com/joanrod/star-vector.git
     

   • Go to the project folder:

     cd star-vector
     

4. Installation of dependencies
   • Update pip and install the required libraries:

     pip install --upgrade pip
     pip install -e .
     

   • This will install all the Python packages needed for StarVector to run.
5. Setting environment variables
   • The following variables need to be configured if you want to train the model or use advanced features:

     export HF_HOME=<模型存储路径>
     export HF_TOKEN=<你的 Hugging Face 令牌>
     export WANDB_API_KEY=<你的 Weights & Biases 令牌>
     export OUTPUT_DIR=<输出路径>
     

   • These can be entered in the terminal, or written to the environment configuration file.

Usage

Generate SVG from images

1. Prepare the image
   • Select an icon, logo, or technical diagram image to save in a common format (such as PNG or JPG).
   • Note: StarVector is not suitable for working with natural landscapes or complex illustrations, it is better to use simple graphics.
2. running code
   • Place the image in the project folder and type it in the terminal:

     python scripts/image_to_svg.py --input <图像路径> --output <输出SVG路径>
     

   • For example:

     python scripts/image_to_svg.py --input icon.png --output icon.svg
     

3. View Results
   • The generated SVG file is saved to the specified path. It can be opened for inspection with a browser or vector graphics software such as Inkscape.

Generate SVG from text

1. Enter text commands
   • Open a terminal and run the text generation script:

     python scripts/text_to_svg.py --text "一个红色圆形" --output circle.svg
     

   • Be as specific as possible, e.g. "a black five-pointed star" or "a blue rectangular box".
2. Check Output
   • The generated SVG file will be saved in the specified path, open it to see if it meets expectations.

Advanced Features: Training Models

• If you want to train the model yourself, you can use the SVG-Stack dataset.
• Run the training command (StarVector-1B for example):

deepspeed scripts/train.py --data_path <SVG-Stack路径> --output_dir <输出路径>

• Note: Training requires a higher configuration computer and may require GPU support.

operating skill

• adjust components during testing: If the generated results are not satisfactory, check if the input image is too complex or if the text description is clear.
• customizable: Parameters can be adjusted in the code, such as increasing the number of samples to improve accuracy.
• View Document: The README.md file in the project root directory has more instructions, so we recommend reading it carefully.

With these steps, you can easily generate SVG files from images or text. It takes about 10-20 minutes to install and is easy to use, so it is suitable for designers or developers with basic programming skills.

application scenario

1. Icon Design
   Designers can use StarVector to quickly convert sketches to vector icons, saving time on manual tracing.
2. technical documentation
   Engineers can use it to generate diagrams or flowcharts that can be embedded directly into documents or web pages.
3. Educational Demonstrations
   Teachers can generate instructional graphics with text descriptions to visualize concepts.
4. web development
   Developers can use it to quickly create scalable web elements and increase loading speed.

QA

1. Can StarVector process photos?
   Cannot. It is designed for icons, logos and technical diagrams and is not suitable for natural images or complex illustrations.
2. Do I have to pay for it?
   No. StarVector is an open source project and the code and basic functionality is free to use.
3. Can the generated SVG be edited?
   Yes. The generated SVGs are in a standard format and can be edited in Inkscape or Illustrator.
4. What if the installation fails?
   Check that the Python version is correct, make sure the network connection is working, or look at the terminal error message.