AI for video editing
Artificial intelligence is already built into video editors: it cleans up voices, transcribes, reframes and finds shots. Used well, it saves you the most tedious tasks. Used badly, it makes everything look the same or gets you into trouble. It's the topic in this course that changes fastest.
AI tools change every few months. This lesson explains the types of tools and how to use them with good judgment; it's best to check specific names, prices and features in each program's current documentation.
What already works really well
| Task | What the AI does | Time saved |
|---|---|---|
| Transcription and captions | Turns speech into timed text | Huge: hours of work in minutes |
| Voice isolation or cleanup | Separates the voice from background noise, music or wind | Rescues recordings that used to be unsalvageable |
| Text-based editing | You delete words in the transcript and the video gets cut | Super fast for interviews and podcasts |
| Scene detection | Finds the cuts in a video that's already edited | Useful for reusing old footage |
| Auto reframe | Follows the subject to go from 16:9 to 9:16 | Turns a horizontal video into a vertical one in seconds |
| Shot search | Finds clips by content ("dog on the beach") | Useful in large archives |
| Masks and cutouts | Separates a person from the background without a green screen | Makes local color adjustments and effects easier |
What to keep an eye on
- Captions: mistakes with names, numbers and homophones. Always check them.
- Voice cleanup: at 100% it usually leaves an artificial sound. Blend in a little of the original signal.
- Reframing: sometimes it follows the person who isn't speaking. Check shot by shot.
- Automatic editing: it can remove silences that were dramatic pauses. The pacing is still yours.
Generating images, voices and video
Generative tools create images, voices, music or video shots from text. They can be useful for storyboards, references, mock-ups or illustrative footage. But they come with risks:
- Real people: don't generate voices or images of real people without their permission. It may be illegal and it's always deceptive.
- Transparency: many platforms require you to label realistic synthetic content, and in journalism and documentary the audience has a right to know.
- Rights: check each tool's terms for commercial use and the legal situation in your country, which is changing fast.
- Sameness: generated content tends to look alike. What sets your work apart is your eye.
In a news or documentary context, a generated image or voice presented as real is a fake. If you use it as an illustration, label it clearly on screen.
A sensible AI workflow
- Transcribe all the raw footage.
- Read the transcript and mark the best lines: that way you make your selects in minutes.
- Do the first cut with text-based editing.
- Move to traditional editing for pacing, J-cuts and L-cuts, and B-roll.
- Clean up the voice with AI, blended with the original.
- Generate the captions and check them.
- Reframe to vertical and check every shot.
- AI saves time on transcription, voice cleanup, search and reframing.
- Always check what it produces.
- No voices or images of real people without permission.
- Label generated content when it could be mistaken for the real thing.
Your turn: save an hour
Take an interview of ten minutes or more and edit it with the help of AI.
- Transcribe it and select the lines by reading the text.
- Do the first cut by text and polish it by hand.
- Note how long it took you and what you had to fix.