Best For
Researchers, consultants, and anyone who needs to hand off a .pptx file that the recipient can actually edit. If the person opening your slides needs to change a title, adjust a chart color, or swap a logo, this is the only AI tool that produces native PowerPoint elements instead of page-sized images. Specifically valuable for thesis defenses, client proposals, investor decks, and annual reports.
How I Actually Use It
Feed it a source document: PDF paper, DOCX proposal, Excel data, EPUB, Jupyter Notebook, or just a Markdown outline. The pipeline's Strategist role asks eight design questions (canvas format, page count, audience, style, palette, icon library, font, image strategy) and locks them into a spec file. Then the Executor generates each slide as SVG, sequentially, checking visual consistency against the spec after every page. Finally, a post-processing stage converts the SVG into native DrawingML shapes — round-corner rectangles become real PowerPoint shapes, text stays as text boxes, charts render as editable chart objects.
The output is a .pptx that behaves exactly like one you'd build by hand in PowerPoint or Keynote. Click any element, drag it, change the text, recolor it.
v2.7.0 adds a live preview workflow: during generation the pipeline opens localhost:5050 where you can click any SVG element, type a revision note, and submit annotations. The AI rewrites the SVG and re-exports the PPTX on the spot. There's also a /create-template command that reverse-engineers any existing .pptx into a reusable template (theme colors, fonts, masters, layouts).
Where It Is Strong
- Every element in the output PPTX is individually selectable and editable. No more "the whole slide is one image" frustration
- spec_lock mechanism prevents visual drift during long generation runs. Page 15 stays visually consistent with page 3 because the design contract is re-injected before each page
- Page Rhythm forces variety. The Strategist tags some pages as "breathing" — big whitespace, full-bleed images, oversized numbers — breaking the "every page is a 3-column card grid" pattern that plagues AI-generated decks
- Supports narration in 90+ languages (TTS via edge-tts, ElevenLabs, MiniMax, CosyVoice) with voice cloning support, plus native OOXML PowerPoint animations. You can export a rehearsal-ready .pptx or a self-playing .mp4
- 14+ AI image generation backends (OpenAI gpt-image-2, Gemini, Qwen, Volcengine, Stability AI, FLUX, Ideogram, Fal.ai, and more) plus free stock photo search via Pexels, Pixabay, Openverse, and Wikimedia
Where It Fails
- Token-heavy. A full run burns roughly 100K tokens. Large decks can double or triple that. This is not a budget-conscious choice
- Strictly sequential. Each page must finish before the next begins. A 20-page deck takes 20x the time of a single page, with no parallelism possible
- Environment setup is non-trivial. Python 3.10+, python-pptx, PyMuPDF, cairosvg, and optionally pandoc. Not a one-command install
- Template triggering is strict. You must provide an actual file path to a template directory. Saying "use a business style" won't activate template mode
- Font support limited to system-installed fonts. No @font-face, no web fonts
Pricing, Difficulty, and Risk
Completely free, MIT license. The real cost is token consumption: budget ~$2-5 per deck at typical API pricing. Setup difficulty is high: multiple Python dependencies, optional system libraries (cairo, pandoc), and you need to understand the multi-role pipeline concept. v2.7.0 adds a Skill Marketplace install path (npx skills add hugohe3/ppt-master) that simplifies initial setup, though you still need pip install -r requirements.txt for the post-processing scripts. Privacy risk is minimal since everything runs locally; the only external calls are to your chosen LLM API. Stability risk: single maintainer (Hugo He), 17,010 stars, 1,633 forks, active development as of May 2026.
Verdict
The only open-source tool that produces genuinely editable PPTX from AI. Worth the complexity if you regularly deliver formal presentations where recipients need to modify the slides. Not worth it for quick internal shares or visual-first showcases — other tools handle those faster and prettier. Think of it as the "final output" stage of your presentation workflow, paired with lighter tools for drafting and ideation.