DEV Community

Cover image for I Turned a Lecture PDF Into a Deck Without Rewriting a Line
Kouki
Kouki

Posted on AI-assisted

I Turned a Lecture PDF Into a Deck Without Rewriting a Line

Engineers do not have a slide problem. We have a conversion problem.

The source of truth is already written: a design doc, a postmortem, a lecture PDF, a 40-minute transcript, a README that grew into a talk. The talk is next Tuesday. Nobody wants to re-outline that material into 12 slides by hand, and nobody wants an AI tool that invents a new structure you then have to fight.

I wanted a path that looks like this:

existing artifact  →  editable outline  →  slides  →  PPTX
         ^                    ^
         |                    |
    do not rewrite      do not skip this
Enter fullscreen mode Exit fullscreen mode

PPT Maker is one of the few tools in this category that puts the middle box on the screen before it spends pixels.

The actual workflow

1. Start from the artifact, not from a vibe prompt.

The generator accepts a topic, pasted text, a PDF, an article, or a video. That matters more than the model picker. If your talk already exists as a PDF, the job is extraction and sequencing, not “write me a presentation about Kubernetes.”

Drop the file in. Pick Agent if the deck has to carry an argument (this is the default for technical talks). Save Creative for the one or two pages that would look silly as bullets — a mental model, an architecture beat, a title.

Set page count in a range (the homepage default is 5–10), text density, 16:9, and whether to keep original wording. If the PDF is already in your voice, keep original text on. Letting the model “improve” lecture notes is how you get keynote English on top of precise terms.

Drop the PDF (or paste) here. Agent, not a blank “write me a TED talk” prompt.

2. Read the outline like a PR.

This is the step most tools skip, and it is the whole point.

You get a structured outline before slides exist. Treat it the way you treat a pull request:

  • Kill the section that is really an appendix
  • Move the recommendation / thesis up
  • Merge the headings the model repeated (this happens on long PDFs; more on that below)
  • Rename anything that would read as jargon in the room

You are not pixel-pushing. You are deciding the talk. That is the only leverage you have.

The middle box in the pipeline: thesis moved up while it is still a line in a list.

3. Generate, then patch a page, not the repo.

Once the outline is something you would defend, generate. Walk the deck the way you walk a failing test: find the one page that is wrong.

If slide 6 still sounds like the PDF’s section 4 dumped into a text box, prompt that slide: “Turn this into three numbered steps. Keep the command names. Drop the paragraph.” The rest of the deck stays put.

That is the difference between a generator and a compiler with an edit loop. Regenerating the whole deck after a one-line fix is make clean when you needed to rebuild one object file.

Rebuild one object file: prompt slide 7, leave 6 and 8 untouched.

4. Export into the toolchain you already have.

Export PPTX if the talk has to live in corporate PowerPoint. PDF for a handout. Keynote if that is the machine on stage. PNGs if you are dropping frames into a web page or a recorded talk.

There is also a plain outline download, which is underrated: it is a backup of the argument when the slides are not the artifact you want to keep.

Where it breaks (so you do not find out on Tuesday)

I would not ship this workflow without the failure modes:

  • Long PDFs duplicate headings. A 30-page report can come back with repeated section titles. Merge them in the outline. Do not wait until you have 18 slides named “Introduction.”
  • Charts are not the point of this tool. It will put numbers on a page. It will not give you a charting environment. Export and finish real figures in the tool you already trust.
  • Visuals are clean, not cinematic. Agent output looks like a serious internal deck. It does not look like Gamma. If the talk is the aesthetic, you are in the wrong product.
  • Credits are a budget, not a vibe. The free plan is 50 credits a month, they do not roll over, and a deck is capped at 10 slides. Creative images cost extra per page (on the order of 10–20 credits each, depending on the image model). A visual-heavy Creative run will burn the free month faster than an Agent outline. Plan the generate-and-patch loop; do not spray regenerations.
  • Browser only. No offline, no desktop app, no API in the public product. Fine for a talk. Not fine for an air-gapped docs pipeline.

None of those are surprises if you treat it as a compiler for existing writing, not as a designer.

A reasonable default for technical talks

If you want a recipe instead of a tour:

  1. PDF or pasted doc in, Agent, Keep Original Text on, 8–10 slides, medium density.
  2. Outline review: thesis in the first third, demo or architecture in the middle, ask / takeaways last. Delete anything the PDF needed that the room does not.
  3. Generate.
  4. Prompt-fix the one dense page and, if you need a visual beat, rebuild only that page in Creative.
  5. Export PPTX. Fix charts by hand. Stop.

That is boring on purpose. Tuesday morning talks are boring on purpose.

The category keeps selling “type a sentence, get a TED deck.” If your source of truth is already a PDF, you do not need a TED deck. You need a faithful, reorderable skeleton and a way to patch one slide without invalidating the rest.

That is the job PPT Maker is actually set up to do.

Top comments (0)