- Your speaker notes are the brief. Slides with clear notes get the most accurate narration.
- Read the script before you generate audio. Editing text is free; regenerating audio is not.
- Fix names, acronyms and product terms once with a pronunciation rule, and the fix applies to every slide.
- PowerPoint animations do not carry over. Split a slide that relies on builds into two or three slides.
- The export is a 720p MP4 with SRT/VTT captions and a transcript, ready for an LMS or YouTube.
- A free account can export up to 3 minutes with a watermark, so you can test on a real deck first.
The short answer
To narrate a PowerPoint with AI, you upload the deck, let the tool write a narration script for each slide, review that script, choose a voice, and export the video. The AI does the writing and the voicing. Your job is the review, because that is where a narrated deck goes from "fine" to something you would put your name on.
This guide walks through the whole process: how to prepare the deck so the narration comes out right, what to check before you spend anything on audio, and what you get at the end.
Three ways to put a voice on a deck
There are three common ways to turn a PowerPoint into a narrated video. They suit different situations.
| Record it yourself in PowerPoint | Hire a voice-over artist | Narrate it with AI | |
|---|---|---|---|
| Who speaks | You, into a microphone | A professional narrator | An AI voice you choose |
| Who writes the script | You | You | The AI drafts it; you edit |
| Fixing one sentence | Re-record that slide | Book another session | Edit the line, regenerate that slide |
| Other languages | Record each one | Hire a narrator per language | Pick a narration language |
| Best for | A one-off talk in your own voice | A flagship video that will not change | Training and course content that gets updated |
Recording in PowerPoint is free and keeps your own voice, and its built-in export keeps your animations. The cost shows up later: every change to the deck means recording again. AI narration moves the cost to the edit. Change a line, regenerate that slide, export again.
Before you start: prepare the deck
Ten minutes of preparation saves an hour of fixing.
- Write speaker notes for the slides that matter. The AI writes each slide's narration from what is on the slide and in its notes. Notes do not need to be polished prose; a few clear sentences on what the learner should understand from the slide are enough.
- Keep one idea per slide. A slide that tries to cover three points gets a long, rushed narration. Split it.
- Split slides that rely on animation. Animations and transitions do not carry over to the video, because each slide becomes a still scene. If a slide reveals four bullets one at a time, make it four slides, or accept that all four appear at once.
- Spell out what you want said. If an acronym should be read as letters ("S-Q-L") or as a word ("sequel"), note it. You can also fix this later with a pronunciation rule.
- Remove slides you would skip when presenting. Backup slides, appendix tables and "questions?" slides make a video longer without making it better.
How to narrate a PowerPoint with AI, step by step
The steps below use Unmixr's PowerPoint to video workflow as the example. The order is the same in any serious tool.
Step 1: Upload the deck
Start a course or training video project and upload the .pptx. A .ppt or a PDF works too. Unmixr reads the deck and splits it into one scene per slide.
Check: if the deck was made in Google Slides or Keynote, export it as .pptx rather than PDF. A PDF keeps the slides but loses the speaker notes, and the notes are what the narration is written from.
Step 2: Choose the output and the language
Pick Narrated course video as the output. Then choose the narration language. It does not have to match the slides: an English deck can be narrated in Spanish, for example, without translating the slides first.
You can also add a short note for the AI about the audience and terminology, for example: "New support agents. Use our product names. Explain acronyms on first use." That note shapes how every slide is written.
Step 3: Choose the narrator
Choose one narrator for the whole deck, or different narrators for different parts. Pick a voice that suits the content: calm and clear for compliance training, warmer and quicker for a product walkthrough. If you have permission to use an instructor's voice, you can clone it and narrate in their voice.
Step 4: Review the script before you generate audio
Choose Review in Scene Studio. Unmixr writes the narration for every slide, and nothing is charged until you generate the audio. This is the step that decides the quality of the video.
Read each slide's line with the slide next to it and check:
- Accuracy. Does the narration say what the slide means? Fix anything the AI framed differently from how you would.
- Length. A slide on screen for 90 seconds with one sentence of narration will feel empty; a dense paragraph over a title slide will feel rushed.
- Names and numbers. Product names, people, figures and dates are where errors hurt most.
Step 5: Fix pronunciation once
Listen to the first few slides. If the narrator mispronounces a drug name, a framework or your company name, add a pronunciation rule for that word. The rule applies to every slide in the project, so you fix it once.
Step 6: Generate the audio and export
Generate the audio, play the scenes back, and adjust anything that sounds off: a pause before a key point, a slower pace on a definition, or a different voice for one slide. Then export.
What you get
The export is a 720p MP4 (1280×720 for a standard 16:9 deck), plus SRT and VTT caption files and a transcript. That covers an LMS, YouTube, an intranet page or a video editor. On paid plans, course projects can also be packaged as SCORM for any SCORM-compliant LMS.
If you need a 1080p master, use PowerPoint's own Export to Video with narration you record yourself. For training and course content, 720p keeps slide text readable on a laptop screen and keeps files small.
Common mistakes
- Skipping the script review. Generating audio first and reading later costs time and credits. Read first.
- Narrating the slide text verbatim. Learners can read. Good narration explains the slide; it does not recite it.
- One giant deck. A 120-slide deck becomes a video nobody finishes. Split long material into modules of 5 to 15 minutes.
- Forgetting the captions. The caption files come with every export. Upload them with the video; they help accessibility and search.
- Mixed voices by accident. Pick a narrator for the course and keep it, so lesson 12 sounds like lesson 1.
Free and paid
A free Unmixr account narrates a deck and exports videos up to 3 minutes long with a small watermark, and no card is required. That is enough to test the whole process on a real deck before you decide. Paid plans remove the length cap and the watermark and add SCORM packaging; see pricing for current plans.
Next step
If you have a deck ready, start a narrated training video and run its first few slides through. For the bigger picture, including scripts, documents and multi-language courses, see the AI training video maker.