How to turn a PPT into a lecture video — no script, no mic, no re-takes
2026-10-16 · 9 min read
A 10-slide microlecture takes 4–6 hours the screen-recording way, and most of it goes into re-takes and caption cleanup, not into teaching. The other route: import the deck, let AI write the narration, synthesise the voice, cut captions from the script — about 20 minutes. The price is fine-grained editing, and Apple-only hardware.
Basis: timings and credit figures come from SlideTalk v1.7's in-app billing rules, checked 2026-10-26. The 20-minute figure is the official reference for a 10-slide deck; your real time depends on how many script revisions you make. The traditional-route range reflects common values from public tutorials and teacher interviews, not a controlled measurement.
Recording a lecture sounds like teaching work. Most of it is not. For a ten-minute class the real ledger reads: 1–2 hours writing, 3–5 takes, 1–2 hours cutting filler and silence, 30–60 minutes fixing captions. However well you teach, those four stages turn one class into one day.
So the question is not "how do I record faster" but "which of these steps should never have needed a human". Re-takes, caption alignment and per-slide animation share three properties: mechanical, verifiable, and expensive to get wrong. That is exactly what automation is for. What follows is the full path from a 10-slide deck to a finished video — and what it cannot do.
First, the time ledger
Before choosing a tool, look at where the cost sits. The table assumes a 10-slide deck and a 10-minute video. The difference between the two routes is not who explains better; it is which steps need you in the room.
| Step | Screen record + edit | AI narration pipeline |
|---|---|---|
| Script | 1–2 hours, written while recording | AI drafts per slide; you fix the pages you dislike |
| Recording | 3–5 takes; slips, stalls, room noise | Synthesised voice, no slips; edit a sentence, regenerate one page |
| Captions | Speech-to-text, then 30–60 min of manual cleanup | Cut from the script itself, burned in plus a sidecar .srt |
| Effects | Manual keyframes, page by page | AI marks the emphasis region, 6 effect types applied |
| Assembly | Export and compress, 10–20 min | Rendered on-device up to 4K, saved with the project |
| Total | about 4–6 hours | about 20 minutes |
The outputs are not equivalent: the traditional route allows freer editing, this one wins on the number of steps that do not need a human. Which to pick depends on whether your deck is already finished.
Step 1 — treat the deck as source material
The reflex is to rebuild the PPT before recording. You do not need to. .pptx / .ppsx / .pdf drop straight in; layout, images and charts are rebuilt close to their original positions, and the slide order is preserved.
Ten minutes spent checking information density per slide is still worth it. That is a deck-quality question, not a tooling one, but it decides how the video reads:
- One idea per slide; split anything you cannot explain in a minute. Ten slides for ten minutes is a comfortable pace
- Replace paragraphs with bullet points and let the narration carry the detail — do not make viewers read the screen
- Give complex charts their own slide so the emphasis effect has room
- Decks heavy on gradients, SmartArt or native animation reconstruct better if you export to PDF first
Step 2 — the script is the one place worth your time
This pipeline collapses four stages into one, and the survivor is the script. Give the AI one line of context — "first-year undergraduates, slower pace, one everyday example per slide" — and it drafts slide by slide; you rewrite in the workspace.
It is worth the time because the script sets the ceiling. Voice, captions and effects are downstream products of it. A good script makes everything after it automatic; a lazy one produces a fluent video that says nothing.
One practical habit: only fix what would mislead a student. Do not chase style. AI drafts skew written; making them sound spoken is worth more than making them sound elegant.
Step 3 — voice, captions and effects in one pass
Pick a voice once the script is final. 42 Chinese voices cover news reading, warm instruction and deeper narration; previews are free. Keep one voice across a lesson — switching mid-way breaks attention.
Captions are neither typed nor transcribed: they are cut from the script, segmented on Chinese phrase boundaries, burned into the frame, and written out as a sidecar .srt for platform upload. Five presets ship; for lectures, white text with a black outline survives both a projector and a phone.
Effects are not manual either. The AI marks what deserves emphasis on each slide — heading, chart key point, conclusion line — and applies one of six effect types. If it guesses wrong, drag to correct it in the editor.
Step 4 — export, archive, compliance
Rendering happens on your own machine (AVFoundation). Resolution is open from 480p to 4K, long videos can export in per-slide segments, and H.264 compression is optional. The project — deck, script, audio, finished cut — lives in the app sandbox, and the same Apple ID picks it up on another device.
Two compliance questions come up constantly. Can you submit the result to a competition or review? Yes: you rendered it, the content is yours, and the watermark is removed permanently after any first top-up. Can you use confidential material? The page text needed for AI generation passes through SlideTalk's own gateway, which does not store it — but fully offline generation is not possible, so judge confidential decks against your institution's rules.