QueryPPT is an innovative automated presentation generation framework designed to directly transform users' concise natural language queries into content-rich and visually coherent presentations. This framework breaks through the traditional reliance on structured documents by implementing a "Query to All" automated process from scratch, achieved through two main stages: "Intelligent Content Generation" and "Adaptive Presentation Creation."
This project consists of two core sub-modules:
- DeepPPT: Responsible for the first stage of intelligent content generation.
- Adaptive Presentation Creation (APC): Responsible for the second stage of adaptive presentation creation.
QueryPPT, From Query to All: Web-Enhanced Content and Intelligent Layout for Stunning Presentations
Automated presentation (PPT) generation is a significant direction for improving productivity. QueryPPT introduces a novel end-to-end framework that can automatically generate content-rich and visually harmonious presentations directly from a user's simple natural language query.
The core idea of QueryPPT is to decouple the complex generation process into two synergistic stages:
-
Intelligent Content Generation: This stage is led by a central "planner agent," which first parses the user's intent and plans the tasks. Subsequently, the planner coordinates multiple "researcher agents" to perform parallel web searches and content refinement. An "imagent agent" analyzes and summarizes the retrieved images to assist in creating chapters with both text and graphics. All content is finally consolidated by a "recorder" module into a structurally complete content outline.
-
Adaptive Presentation Creation: This stage focuses on visual presentation. First, a "painter agent" dynamically generates background images for the title and content pages based on the style and theme of the user's query. Next, a "colorist agent" intelligently infers the optimal font color scheme based on the generated backgrounds. For layout, we offer two flexible strategies: the "architect agent" uses a component-based layout, dynamically adapting pre-set visual components to the content; while the "mapper agent" employs a set of predefined slide combinations, efficiently populating the content into well-designed templates.
This project is divided into two independent modules, each with detailed instructions. Please refer to the corresponding README file according to your needs.
This module is responsible for generating structured Markdown content from your query through web searches and multi-agent collaboration.
📘 For detailed instructions, please refer to:
👉 DeepPPT/README.md
This module is responsible for rendering the Markdown content generated by DeepPPT into a visually appealing and well-laid-out PPT file.
📘 For detailed instructions, please refer to:
👉 APC/README.md
We sincerely thank the following projects and communities for their contributions and support to this project:
- The design and architecture of this project were inspired by ByteDance's excellent project DeerFlow, especially in terms of multi-agent orchestration, research planning, and LLM/VLM integration.
- We also extend our special thanks to the Auto-PPT project, whose innovative design concepts and implementation provided significant inspiration for the visual presentation part of our project.
We welcome you to submit Issues or suggestions. Please bear with us as we are new to this and will continuously improve and optimize the project !
📮 Contacts:
- Jiaren Peng: 1354527247@qq.com
- Jingyang Hu: 191415857@qq.com
If you like the QueryPPT project, please give us a ⭐ Star! Your support is the greatest motivation for the continuous development of the project.
QueryPPT - Making PPT creation smarter and more efficient !