Clip long videos without uploading them to the cloud
You finish a two-hour recording. Somewhere inside it are five minutes worth posting. To find them with most AI clipping tools, you first have to upload the whole file to someone else’s servers and wait.
We think that step is the wrong default, so we are building ClipBomb the other way around: a desktop app that does the heavy work on your own Mac. This post explains what that means in practice, what still leaves your machine, and where it costs you something.
The upload tax
Cloud clippers are convenient, and the convenience has a price you pay every time:
- Size. A long recording is often several gigabytes. Uploading it on a home connection can take longer than the editing you were trying to avoid.
- Privacy. Unreleased episodes, client footage, interviews recorded under NDA, internal all-hands: plenty of video should not sit on a third party’s storage, even temporarily.
- Dependency. If the service is down, queued, or changes its terms, your workflow stops with it.
None of this makes cloud tools bad. It makes them a trade-off that is usually presented as the only option.
What “local-first” means here
Local-first does not mean “nothing ever touches the internet”. It means the video file stays on your machine, and we are specific about which steps run where:
| Step | Where it runs | What leaves your Mac |
|---|---|---|
| Transcription | On your Mac, using Whisper | Nothing |
| Choosing clips | An AI model, using your own API key | Transcript text only, never the video |
| Review and reframe to vertical | On your Mac | Nothing |
| Export with burned-in captions | On your Mac, using FFmpeg | Nothing |
The video, the audio, and the exported clips never go to our servers or to the AI provider. Only text, the transcript, is sent so a language model can judge it.
How it picks clips without watching the video
A transcript is enough to find most good moments in talking-heavy video, because the thing that makes a clip work is usually what is being said. The pipeline has four stages:
- Work out what kind of video it is. A comedy set, a lecture and a podcast interview have different ideas of a good moment, so the model decides the genre first.
- Propose candidate clips from the transcript.
- Judge each candidate on its exact words. Does it make sense without the context before it? Does it open on a strong sentence? Does it pay off inside the clip?
- Rank the survivors across the whole recording and show you the best ones.
You can see this in action in the interactive judge demo on the homepage. It uses sample data, but it follows the same steps.
You stay in control of the final cut. Every clip is reviewed in a visual editor, where you trim it, adjust the 9:16 crop, and approve it before anything is exported.
The honest trade-offs
A local-first design is not free. Here is what you give up:
- The transcript still goes to an AI provider. That is the one piece of your content that leaves your Mac. If even the text is too sensitive, this approach is not fully private for you.
- You bring your own API key. There is no account to sign into, but you do need a key from your AI provider.
- It is not fully offline. Transcription is local, but choosing clips needs a model, so you need a connection for that step.
- It runs on your hardware. Local transcription uses your machine. Apple Silicon Macs come first, and Windows is planned.
- It cannot see the picture. Because clips are chosen from the transcript, a moment that is purely visual and has no words will not be picked up. You can always add one yourself in the editor.
When a cloud clipper is the better choice
If your recordings are short, not sensitive, and you are happy to upload them, a cloud tool can be faster to start with and needs no key. If you are on a machine that is not an Apple Silicon Mac, a desktop app that does not support it yet will not help you today.
ClipBomb is for the other case: long recordings, footage you would rather not upload, and a workflow you want to own.
Join the waitlist — one email when ClipBomb is ready.Where things stand
ClipBomb is in active development and not ready to download yet. This blog is where we will post product updates, feature breakdowns and guides on getting more out of long recordings. If you want one email when it is ready to install, join the waitlist above.