icip-cas/PPTAgent

What it solves

PPTAgent addresses the difficulty of automatically generating high-quality presentations that balance content quality, visual appeal, and structural coherence. It moves beyond simple text-to-slide conversion by mimicking human workflows to create editable, visually reviewed PowerPoint decks.

How it works

The system operates as a skill for coding agents (like Claude Code, Codex, and OpenCode). It uses a two-stage, edit-based approach: first analyzing reference presentations to extract schemas, then drafting an outline and iteratively applying editing actions. To ensure quality, it integrates a multimodal vision model (such as deepseek-v4-flash-vision) to act as a visual reviewer, rendering the slides as images and providing feedback for revisions before exporting the final result as an editable .pptx file.

Who it’s for

It is designed for developers and users of AI coding agents who need to generate professional, editable presentations from documents or prompts without manual formatting.

Highlights

• Agent Integration: Works as a plugin/skill for Claude Code, Codex CLI, and OpenCode.
• Visual Feedback Loop: Uses multimodal vision models to review rendered slides and suggest improvements.
• Editable Output: Exports final decks as standard .pptx files.
• Advanced Generation: Supports free-form visual design, autonomous asset creation, and text-to-image generation via the DeepPresenter upgrade.