Paper-Presentation-Tips Adds AI Skill to Generate & Polish Academic Slides

The MLNLP community's Paper-Presentation-Tips repository provides 11 checkable tips for academic paper presentations and a new paper-presentation-skill that integrates AI assistance into the slide-creation workflow, outputting a fully editable .pptx file.

11 Checkable Tips for Clearer Paper Talks

The tips are organized around overall design, background motivation, method explanation, experimental analysis, and closing summary. Each tip follows a consistent structure: common problem , ❌ Bad Slide , ✅ Good Slide , and 💡 Core Principle , using before/after comparisons to show concrete improvements. When drafting or auditing a presentation, users can start from these recurring problems:

Too much text on a slide, key points buried → Tip 1: Keep per-slide word count low

One slide tries to cover background, method, and results → Tip 2: One core task per slide

Long background but audience still misses the problem → Tip 6: Introduce the problem with a concrete example

Formulas appear too early, audience gets lost → Tip 7: Explain with an example first, then show the formula

Many experimental results, core finding not prominent → Tip 9: Let the audience spot the key result at a glance

Abrupt ending without clear recap → Tip 11: Review core content on the summary slide

These tips serve both as a checklist when drafting new talks and as a page-by-page audit tool for existing PPTs to refine content selection, narrative order, and visual emphasis.

New AI Skill: From Outline to Editable PPTX

The paper-presentation-skill embeds the above advice into a three-step automated pipeline:

Step 1 — Outline: How should this talk be structured?

Provide the paper, audience background, and time budget. The agent produces a slide-by-slide content plan. If a PPT already exists, it first lists specific issues (why they hinder understanding) and proposes concrete fixes.

Step 2 — Design: How will each page look?

After confirming the outline or revision plan, the agent renders full-page reference images showing text, main graphics, color scheme, and layout. Users can request adjustments — e.g., "slide 3 has too much text" or "the conclusion on slide 7 needs more prominence."

Step 3 — Build: Produce the editable PPTX.

The agent constructs shapes, tables, and charts per the approved reference images. PowerPoint, WPS, or LibreOffice can then be used to verify the actual rendering. Photos and complex original paper figures are kept as images so they remain movable, scalable, and croppable.

The repository includes before/after examples such as "using input/output cases to illustrate a task" and "unifying color scheme and font sizes."

Installation

git clone https://github.com/MLNLP-World/Paper-Presentation-Tips.git
cd Paper-Presentation-Tips
python -m pip install -r skills/paper-presentation-skill/requirements.txt

Requirements: Git, Python 3.10+, and an agent with file read/write, command execution, and image viewing capabilities (e.g., Codex or Claude Code). Final output is checked in PowerPoint/WPS/LibreOffice; reference images can be generated via the agent's built-in drawing tools, a configured drawing service, or user-supplied images.

Usage Examples

Create a new presentation from a paper

/paper-presentation-skill
请根据提供的论文制作一份中文组会 PPT。
听众：本领域研究生。
时长：约 15 分钟。
重点：讲清研究问题、方法核心思路和主要实验结论。

请先整理汇报大纲与每页内容，等我确认后再展示页面参考图；参考图确认后，制作可编辑的 PPTX，并检查实际显示效果。

Polish an existing PPT

/paper-presentation-skill

请检查并润色这份 PPT。
结合仓库中的汇报建议，逐页指出内容安排、信息密度、标题、配色和图表中的具体问题，并说明怎么改。

先给出问题清单和修改方案，等确认后再生成页面参考图、制作可编辑的 PPTX。

This iterative flow lets users lock down content and design before verifying the final .pptx rendering.

Project Resources

The repository also collects research-presentation videos, courses, experience articles, and related tools.

GitHub:

https://github.com/MLNLP-World/Paper-Presentation-Tips