A core strength of OmniSVG lies in its technical architecture, which transforms SVG commands and coordinates into discrete tokens, separating structural logic from geometric details. This tokenization enables efficient training and expressive SVG generation, supporting long context sequences of up to 30,000 tokens for highly detailed outputs. The system employs a Vision Transformer (ViT)-based image encoder and a dedicated adapter to project image embeddings into the latent space of a large language model, facilitating seamless multimodal input handling. Textual prompts are processed through a specialized tokenizer and embedder, allowing the model to generate SVG code that accurately reflects both visual and textual instructions. The result is a robust and flexible SVG generation pipeline that integrates smoothly into professional design workflows.
OmniSVG also supports advanced features such as character-reference SVG generation, enabling users to create consistent vector illustrations of specific characters in various poses and scenarios. This is particularly valuable for animation, branding, and game design, where maintaining style uniformity across multiple assets is essential. The project is open source, with datasets like MMSVG-Icon and MMSVG-Illustration already available for public use, and a character dataset planned for future release. By providing both a powerful generation model and rich annotated datasets, OmniSVG is positioned as a key enabler for scalable, automated vector graphic creation in diverse creative and technical fields.
Key features include:
- Text-to-SVG generation for converting descriptions into editable vector graphics
- Image-to-SVG conversion for transforming raster images into structured SVGs
- Character-reference SVG generation for style-consistent illustrations
- Supports highly detailed and complex SVG structures with long context handling
- Open-source datasets and evaluation protocols for research and development
- Efficient and expressive SVG tokenization for logical and editable outputs