BizGen - An AI infographic generation tool jointly developed by Tsinghua University and Microsoft
BizGen is an AI infographic generation tool jointly developed by Tsinghua University and Microsoft Research, focusing on article-level visual text rendering. It can transform long articles into professional-grade infographics and slideshows with a single click, solving the problems of traditional methods...
What is BizGen?
BizGen is an AI infographic generation tool jointly developed by Tsinghua University and Microsoft Research, focusing on article-level visual text rendering. It can transform long articles into professional-grade infographics and slideshows with a single click, solving the problems of blurry text and disorganized layout that traditional tools encounter when processing long texts. Based on the high-quality dataset Infographics-650K and an advanced "layout-guided cross-attention mechanism," it can break down long texts into smaller instructions and precisely inject them into different regions of the image.
BizGen's main functions
- High-quality content generationBased on the article content entered by the user, it automatically generates professional-quality infographics and slideshows, solving problems such as blurry text and messy layout when traditional tools process long articles.
- Multilingual and style supportIt supports ten different languages and can generate infographics in various styles to meet different needs.
- Multi-layer transparent information imageIt excels in generating multi-layered transparent infographics, offering more flexible and diverse information presentation.
- High accuracy and typesetting qualityIts text spelling accuracy far exceeds that of other models, and user research shows that its typesetting quality is more favored.
- Strong technical supportBased on the Infographics-650K dataset, a "layout-guided cross-attention mechanism" is introduced to ensure fine control over each visual element and text region.
BizGen's technical principles
- High-quality datasetsThe BizGen team built the Infographics-650K dataset, an unprecedentedly large and high-quality dataset of business content, containing 650,000 exquisite business infographics and slides, each accompanied by detailed layout information and descriptions. This lays a solid foundation for models to learn and understand complex business designs.
- Layout-guided cross-attention mechanismIt can break down lengthy, article-level prompts into "small instructions" for different areas, precisely injecting these instructions into different regions of an image based on a preset ultra-high-density layout. This ensures fine-grained control over each visual element and text area, avoiding the chaos and errors caused by global processing in traditional methods.
- Layout condition control generationDuring the reasoning phase, BizGen uses the "layout condition-controlled generation" method, which acts like a meticulous quality inspector, carefully checking each generated sub-region and correcting any flaws in a timely manner to ensure the perfect presentation of the final work.
BizGen's project address
- Project official website:https://bizgen-msra.github.io/
- Github repository:https://github.com/1230young/bizgen
- HuggingFace model library:https://huggingface.co/PYY2001/BizGen
- arXiv technical paper:https://arxiv.org/pdf/2503.20672
Application Scenarios of BizGen
- Business ReportQuickly generate high-quality business reports and presentations.
- Product ShowcaseCreate attractive product promotional posters and slideshows.
- academic researchGenerate academic reports and presentations.
- social mediaCreate engaging social media content.
- Education: Helps teachers quickly create teaching materials.
- Advertising designAutomatically generates beautiful and theme-appropriate advertising posters.