- Home
- AI Vector Generators
- OmniSVG
OmniSVG
What it is
OmniSVG creates editable, scalable SVG graphics from natural-language descriptions and converts PNG, JPG, BMP, or WebP files into SVG. Image uploads can be up to 5 MB. Results can be previewed, downloaded, or copied as code for editing in an SVG-compatible tool. The free demo requires no account and shows remaining daily generations in the browser, with five free generations per day. The research project also supports character-reference SVG generation. Its system combines a pretrained Qwen-VL vision-language model with an SVG tokenizer and introduces MMSVG-2M, a dataset of two million richly annotated SVG assets. The Apache License 2.0 codebase includes local Gradio deployment instructions using Python 3.10, Cairo, and CUDA 12.1-tested dependencies. Listed model downloads range from 7.69 GB to 17.2 GB, while reported inference memory use is 17 GB or 26 GB for two larger variants. A Hugging Face Spaces demo and community ComfyUI plugin are listed.
For and against
For
- Generates SVG from text descriptions
- Converts four listed image formats
- Outputs editable vector graphics
- No account required for free demo
- Supports local Gradio deployment
Against
- Free demo allows five generations per day
- Image uploads are limited to 5 MB
- Listed models require large downloads
- Inference reports up to 26 GB GPU memory
The verdict
OmniSVG offers both text-to-vector generation and image conversion with editable SVG output. The browser demo is free without an account, while local use involves substantial model downloads and hardware requirements.
Who it suits
OmniSVG suits web developers, UI/UX designers, content creators, and print or merchandise users who need editable vector output. It also fits users wanting local deployment or image-to-vector conversion.
Plans and prices
Free
5 free generations per day · image uploads up to 5 MB
Read 30 Sep 2026 · omnisvg.net
On the record
- Community integration
- The repository lists a community-contributed ComfyUI OmniSVG plugin.github.com · read 30 Sep 2026
- Dataset
- The project introduces MMSVG-2M, a multimodal dataset containing two million richly annotated SVG assets.omnisvg.github.io · read 30 Sep 2026
- Editing and export
- Generated SVGs can be previewed, downloaded, or copied as code for editing in an SVG-compatible tool.omnisvg.net · read 30 Sep 2026
- Generation modalities
- The research project supports Text-to-SVG, Image-to-SVG, and Character-Reference SVG generation.omnisvg.github.io · read 30 Sep 2026
- Image conversion
- The image converter accepts PNG, JPG, BMP, and WebP files up to 5 MB and outputs editable SVG.omnisvg.net · read 30 Sep 2026
- License
- The OmniSVG code is licensed under Apache License 2.0.github.com · read 30 Sep 2026
- Local deployment
- The repository provides local Gradio deployment and installation instructions using Python 3.10, Cairo, and CUDA 12.1-tested dependencies.github.com · read 30 Sep 2026
- Model architecture
- OmniSVG uses a pretrained Qwen-VL vision-language model together with an SVG tokenizer.omnisvg.github.io · read 30 Sep 2026
- Model downloads
- The repository lists OmniSVG1.1_8B at 17.2 GB, OmniSVG1.1_4B at 7.69 GB, and OmniSVG-3B at 8.49 GB.github.com · read 30 Sep 2026
- No account
- The free demo does not require an account and displays the remaining daily generations in the browser.omnisvg.net · read 30 Sep 2026
- Online demo
- The repository links to an online Hugging Face Spaces demo.github.com · read 30 Sep 2026
- Performance limit
- The repository reports 26 GB GPU memory usage for OmniSVG1.1_8B inference and 17 GB for OmniSVG1.1_4B.github.com · read 30 Sep 2026
- Project affiliations
- The project lists Fudan University and StepFun as the authors’ affiliations.omnisvg.github.io · read 30 Sep 2026
- Text generation
- OmniSVG converts natural-language descriptions into clean, scalable SVG graphics.omnisvg.net · read 30 Sep 2026
- Use cases
- The site lists web developers, UI/UX designers, content creators, and print or merchandise users as target audiences.omnisvg.net · read 30 Sep 2026
