AI

EditPPT: Faithful Long-Deck Slide Editing via Structured Tool-Using Multi-Agent with Dual-Modal Validators

Researchers have developed a new system called EditPPT to automate the editing of PowerPoint presentations. Unlike existing systems that rely on idealized representations or open-ended code generation, EditPPT uses a multi-agent framework that reformulates slide editing as a constrained tool-selection problem. This approach allows for localized shape-level operations through the native PowerPoint interface, preserving the structure of user-authored decks and reducing errors i
Researchers have developed a new system called EditPPT to automate the editing of PowerPoint presentations. Unlike existing systems that rely on idealized representations or open-ended code generation, EditPPT uses a multi-agent framework that reformulates slide editing as a constrained tool-selection problem. This approach allows for localized shape-level operations through the native PowerPoint interface, preserving the structure of user-authored decks and reducing errors in long presentations. The system also includes dual-modal validation to assess both instruction fidelity and visual quality. Experiments show that EditPPT achieves high execution rates and accuracy on short, medium, and long deck tiers. --- Why it matters: This matters because current AI-powered slide editing tools often struggle with long presentations due to cascading errors. EditPPT's approach provides a more robust solution for automating slide editing in real-world scenarios. Source: https://arxiv.org/abs/2608.20381

This article was originally published at: https://arxiv.org/abs/2608.20381