PPT Master showed up on GitHub Trending, and the first impression was another AI presentation generator. After reading through the README and the engineering notes under docs/, the difference became clear: it does not place HTML screenshots into a ZIP archive. It creates real .pptx files with python-pptx.
The project has 48,152 stars, is written in Python, and is released under the MIT License. Its sponsors include Kimi, which is promoting K3, as well as PackyCode, APIKEY.FUN, RunAPI, and UCloud AI Computes. The sponsor list has the familiar feel of a Chinese open-source project.
Here is the practical experience and technical breakdown from running the full pipeline.
The Problem It Solves
What do tools such as Gamma, Beautiful.ai, and the many browser-based presentation plugins actually produce? Most render HTML and package the result, or return a collection of images. When opened in PowerPoint, every slide is effectively a single picture. The text cannot be edited, the images cannot be replaced, and the template is locked.
The resulting file is effectively dead inside PowerPoint.
PPT Master takes the opposite approach: the PPTX file is the target artifact from the beginning. Starting with the outline, every page is created with real shapes, charts, tables, and slide layouts through python-pptx. Users can open the file in PowerPoint and immediately edit the text, replace images, or modify the template.
The distinction sounds minor but has major practical consequences. An editable PPTX can enter a company's CI workflow, support batch-generated pitch decks, and connect to an agent pipeline. An image package can only be viewed.
How to Use It
There are two primary ways to use the project.
CLI Mode
# 最快上手:一行命令
pptmaster ./report.pdf --style corporate --tts --narration
# 指定输出路径和 LLM
pptmaster ./report.pdf \
--llm kimi-k3 \
--output ./output/q4-deck.pptx \
--template ./templates/corporate.pptxLibrary Integration for Agents
from ppt_master import PPTMaster
from ppt_master.loaders import PDFLoader
# 1. 准备输入
doc = PDFLoader("./report.pdf").load()
# 2. 初始化(需要 LLM API key)
master = PPTMaster(
llm="kimi-k3",
template="./templates/corporate.pptx",
tts=True,
)
# 3. 一键生成
output_path = master.build(
source=doc,
style="corporate",
slides=18,
narration=True, # 把演讲者备注转成 TTS 嵌入 PPTX
)
print(f"Done: {output_path}")
# 产出 ./output/report-2024-01-15.pptx,PowerPoint 直接打开即可编辑Supported inputs include PDF, DOCX, URLs, and plain-text topics. The LLM layer uses a general interface, so Kimi K3, Claude, and GPT can be selected. TTS is optional. When enabled, speaker notes are converted to audio and inserted into the audio node of slide.notes_slide.
The code samples are taken directly from the repository's docs/usage.md and can be copied and run.
How It Works
The pipeline has four stages:
- DocumentLoader: Accepts PDF, DOCX, or URLs and outputs chunked text.
- OutlinePlanner: Sends the content to an LLM, extracts the key points, and returns a JSON outline. Each page includes
slide_id,title,bullets,speaker_notes, andchart_hint. - PageBuilder: Iterates through the pages, invokes
ChartBuilder,TableBuilder, orImageFetcherbased onchart_hint, and combines the components. It then usespython-pptxto apply the template, animations, and transitions. - TTSWriter: Reads the
speaker_notes, calls the TTS service, and embeds the generated audio into theaudionode ofslide.notes_slide.
One engineering detail stands out: every shape has a custom placeholder name, such as chart_q4_revenue. Downstream components can locate and modify a shape by name. This is particularly useful when using the project as a library. If an agent needs to insert a specific chart at a known position, it can locate the target with placeholder.name == "xxx".
Strengths and Limitations
PPT Master follows a different path from three established approaches:
| Approach | Editable | Cross-platform | CI-friendly | Speed |
|---|---|---|---|---|
| Office COM automation | ✅ | ❌ Windows only | ❌ Requires Office | Slow |
| Frontend HTML export, following the Gamma approach | ❌ | ✅ | ✅ | Fast |
| Direct python-pptx generation with PPT Master | ✅ | ✅ | ✅ | Medium |
PPT Master takes the third path and earns three check marks across the four dimensions. The trade-off is that animations and transitions cover only the subset supported by python-pptx. Advanced effects such as Morph and Push are unavailable, so it cannot produce a highly theatrical Keynote-style presentation. Chart support also does not cover every Office chart type. Advanced charts require a separate route, such as generating them with matplotlib and importing the result as an image. Templates must be standard .pptx files, and custom themes require direct modification of the underlying XML.
Another issue is documentation quality. The docs/ directory contains many engineering notes, while the README leans more toward marketing. Details about selecting an LLM, working with templates, and configuring TTS are mostly buried in the documentation. Following the README alone does not provide the full picture for a first-time user.
Hands-On Test
A 12-page annual-report PDF was run through the full pipeline.
- Total processing time: 87 seconds, including 14 Kimi K3 calls and 12 TTS requests
- Output: an 18-page
.pptx, including pages split from the source - File size: 4.2 MB
Inspection confirmed that every page contains real shapes rather than images. The template inheritance works correctly, and the .m4a files embedded in the speaker notes play directly in PowerPoint.
The 12-page annual report became an approximately 18-slide presentation because PageBuilder automatically splits pages at logical breakpoints. The deck averaged 0.8 charts per slide, although that ratio is difficult to control because it depends on the LLM's judgment about where a chart is needed.
The default TTS provider is Kimi TTS. The voice is acceptable, but its intonation still sounds distinctly synthetic, so the narration should be reviewed before a live presentation.
Who Should Use It
Suitable for:
- Teams that need to produce pitch decks or training materials in batches
- Developers integrating presentation generation into an agent pipeline
- Projects requiring extensive post-generation editing, such as enterprise work with repeated requirements changes
Not suitable for:
- Keynote presentations focused on highly sophisticated visual effects; those still require Keynote and AI-assisted image editing
- Occasional individual use, where a service such as Gamma may be simpler
For teams building AI agents, generating structured presentations at scale, and avoiding dependence on a Windows Office environment, PPT Master is currently one of the most engineering-friendly open-source options.
The code, templates, and documentation are available at https://github.com/hugohe3/ppt-master. Questions can be submitted as issues, and the author responds fairly quickly.