This project provides a local, cost-free PowerPoint generator that leverages local Large Language Models (LLMs) via the Oobabooga Text Generation WebUI API. It's designed for users who want to create presentations without relying on cloud services or incurring API costs, offering enhanced privacy and customization.

How It Works

The tool prompts the user for presentation details, then uses optimized prompts to generate slide content via the Oobabooga API. The python-pptx library converts this text into a PowerPoint file, incorporating image placeholders and offering 7 design options. This direct generation approach allows for easy post-creation editing within PowerPoint, unlike MARP-based solutions.

Quick Start & Requirements

• Install: pip install python-pptx regex collection
• Prerequisites: Oobabooga Text Generation WebUI instance running with an instruct-finetuned model and the --extensions api flag. 13B+ parameter models are recommended, with Mistral instruct models suggested for 7B.
• Configuration: Modify HOST or URL and model_type (Prompt Template) in powerpointer.py if Oobabooga is not on the default host/port.
• Run: python3 powerpointer.py
• Documentation: PowerPointer (for the OpenAI API version)

Highlighted Details

• Directly generates .pptx files, enabling easy editing in PowerPoint.
• Includes image placeholders and 7 selectable design themes.
• Supports multiple prompt templates (Alpaca, ChatML, Vicuna, Llama2Chat) and allows custom additions.
• Recommended models include Nous Hermes Mixtral 8x7b DPO and Hermes Theta 8B.

Maintenance & Community

• Developed by CyberTimon.
• Open to pull requests for bug fixes and improvements.

Licensing & Compatibility

• License: Not explicitly stated in the README.
• Compatibility: Designed for local execution, compatible with closed-source workflows due to its local nature.

Limitations & Caveats

Local models may exhibit limitations such as hallucination, ignoring specified slide counts, or forgetting additional information. Adding new prompt templates is currently complex.