EvoPresent automates the creation of aesthetically pleasing academic presentations from research papers. It targets researchers and academics seeking to efficiently generate high-quality slides, leveraging AI for content extraction, scriptwriting, design, and iterative self-improvement of visual appeal.

How It Works

EvoPresent employs a multi-stage AI pipeline. It begins with content extraction and voice generation, followed by storyline and script construction. Content is then enhanced using image generation and knowledge retrieval. The framework designs and renders slides, incorporating an aesthetic checker that guides iterative self-improvement. This refinement process is powered by the PresAesth model, trained on human preferences for aesthetic evaluation tasks.

Quick Start & Requirements

Installation requires Python 3.10 and pip install -r Evopresent/requirements.txt. Key dependencies include API keys for OpenAI, DeepSeek, Gemini, and Claude models, as well as playwright. Inference is initiated via python3 -m evopresent.ppt.ppt_gen_pipeline with parameters for paper path, target slides, style, and model configurations. Video generation requires pip install -r Evopresent/generation_requirements.txt and setup for external libraries FLOAT and MegaTTS3, along with specific model weights and TTS configurations. PresAesth training involves a separate setup script (train/setup.sh) and environment activation.

Highlighted Details

• Features a self-improving aesthetic agent (PresAesth) trained on human preferences for iterative slide refinement.
• Supports customizable presentation styles via theme templates located in `Evopresent