Cutting Long Videos into Shorts and Reels with AI: How to Automate It
How AI finds strong moments in a long video or stream, cuts them to vertical format, adds subtitles and prepares them for publishing. The pipeline and the pitfalls.
A long video, podcast or stream is dozens of short clips nobody has cut yet. That used to be an editor's job. Now AI does most of it: it finds complete, strong fragments, cuts them to vertical format and adds subtitles. A person just picks the best.
How AI cuts a video
- 01Transcripttimestamped text
- 02Momentscomplete thoughts
- 03Cutat the pauses
- 04Vertical9:16 tracking the face
- 05Subtitlesbig, for muted viewing
Automatic cutting usually has five steps:
- Transcription. Speech becomes timestamped text. Everything else is built on it: the model works with text, not pictures.
- Finding moments. The model reads the transcript and looks for fragments that make sense without context: a complete thought, a story, a strong line, an answer to a question.
- Cutting. Fragments are cut by timestamp, with edges trimmed at pauses so sentences do not break off.
- Vertical format. A 16:9 frame becomes 9:16. Good tools track the speaker's face and move the crop with it.
- Subtitles and styling. Large subtitles, a title, sometimes a cover. Most people watch short videos on mute.
What can be fully automated
- Picking up new videos from a channel or a folder.
- Transcribing and listing clip candidates.
- Cutting, converting to vertical, adding subtitles.
- Saving drafts and sending them to Telegram for approval.
- Publishing the chosen clips on schedule to Shorts, Reels and VK Clips.
Where a person is needed
- The final pick. AI is good at finding complete fragments but does not always know what is funny or important for your audience.
- Checking meaning. A clipped line can sound different out of context. That is a reputation issue.
- Rights. Cut your own videos or ones you have rights to. Other people's videos without permission lead straight to strikes and bans.
A video content factory pipeline
| Step | What happens | Who |
|---|---|---|
| Source | New video on a channel, a stream, a folder | Automatic |
| Transcription | Timestamped text | AI |
| Selection | 5-15 scored candidates | AI |
| Editing | Cutting, vertical crop, subtitles | Automatic |
| Approval | Drafts in Telegram with yes / no buttons | Person |
| Publishing | Scheduled across platforms | Automatic |
You can build this pipeline in n8n with calls to the services you need and your own scripts, or use ready-made clipping services if volumes are small.
Pitfalls
- Bad audio means a bad transcript, and therefore bad selection.
- Platform limits. Posting too often from one account can hurt reach. A steady schedule works better.
- Repetitive clips. Ten similar clips from one stream in a row tire the audience. Mix topics.
- Cost. Video processing uses more resources than text. Estimate spend before launch.
FAQ
Which AI cuts videos into Shorts?
There are ready-made clipping services, and there are pipelines built from parts: transcription, a model for selection, editing by script. A service is simpler to start with; a custom pipeline is cheaper and more flexible at volume.
Can I cut videos into Shorts for free?
Small volumes, yes, on free plans. For regular cutting of dozens of videos, free limits are usually not enough.
How many clips come out of a one-hour video?
It depends on the content. A dense conversational video usually yields from a few to a couple of dozen usable fragments, and fewer truly strong ones.
Can clips be published without review?
Technically yes, but I recommend leaving a person the final "yes". It takes seconds per clip, and an error in meaning can be costly.