> For the complete documentation index, see [llms.txt](https://learn.firecut.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://learn.firecut.ai/features/magic-cut.md).

# Magic cut

### ❓ What is it?

Magic cut turns a raw, unedited talking-head recording into a finished first edit with one click.

If you record talking-head or "self-tape" style videos (several takes, false starts, stumbles, and chat between takes) you usually face lots of manual work before the video is watchable: cutting the bad takes, tightening pauses, adding captions, zooms and chapters. 

**Magic cut does all of that in one pass.** It transcribes your footage, works out which parts are false starts, stutters and duplicate takes, removes them, and builds a brand-new sequence with editing extras (zooms, captions, titles, chapters) layered on automatically.

You give FireCut one thing:

* One source **video clip or sequence** (that already contains audio)

…and FireCut produces a new, fully assembled sequence. **1 click is all you need.**

{% embed url="<https://www.youtube.com/watch?v=OskbuFBvRfg>" %}

### 🤔 How does it work?

The feature works in three stages, all automatic after you press **Go**.

#### 1️⃣ Transcribe your footage

**Pick a video clip or sequence from your project tree** (bins expand and collapse; a small icon shows whether each item is a video file or a sequence) and click **Go**.

If you've changed the source (e.g. you moved some things around in a sequence that was previously transcribed), **you can force a re-transcription**. Use the \[…] menu on the Go button → Re-transcribe and Go.

#### 2️⃣ Cut the timeline

**FireCut builds a new sequence containing only the kept parts:**

* Word timings are tightened against the actual audio waveform (using a silence threshold measured from *your* clip, so breaths and trailing sounds are trimmed accurately).
* Long silences *inside* a kept section are cut too.
* A subtle audio crossfade is added at every cut, so edits don't pop or click.

#### 3️⃣ Enhancements

**Magic cut then layers on enhancements.** Which ones depends on the video's format, detected from its shape and length. The extras include zooms, punch-out, captions, titles and chapters, depending on your input video/sequence. These are selected automatically.

Everything lands on its own track, so the result stays easy to edit; just disable the tracks you don't want.

### ✏️ Reviewing & editing the cuts

By default Magic cut applies its cuts immediately. If you'd like to approve them first, open **Advanced options** and turn on **Review cuts before applying**.

With review on, pressing Go runs the analysis and then pauses on a **Review cuts** panel showing your full transcript with shaded sections (to be kept):

* **Adjust a section** — hover a shaded section and drag the chevron handles on its edges to grow or shrink what's kept.
* **Add a section** — drag-select across un-shaded transcript text to force-keep something the AI cut (a take you liked, an ad-lib).
* **Delete a section** — right-click a shaded section and choose **Delete section**.
* **Create timeline** — build the sequence from your (edited) selection.
* **Cancel** — return to the Magic cut screen. Nothing is modified.

### ⚙️ Settings

#### Source

* **Source clip or sequence** — the single item to edit. Expand bins to find it; only clips and sequences (not bins) can be selected, one at a time. Items with a saved transcript show a captions badge.
* **Go** — runs the whole pipeline. If the selected item already has a transcript, the **\[…]** menu on the button offers **Re-transcribe and Go** to force a fresh transcription (this also clears the cached cut decisions).

#### Advanced options

* **Audio language** (default: Auto-detect) — the language of the spoken audio, used for transcription. Auto-detect lets the transcription model identify the language itself; set it explicitly if detection gets it wrong.
* **Review cuts before applying** (default: off) — pause after the analysis so you can approve, adjust or reject the proposed cuts before the timeline is built (see above). Off = the one-click flow.

### 💡 Tips for best outcomes

* **Use clean audio.** Mute background music and sound effects, and keep speech clear — the cleaner the transcript, the better the cut decisions.
* **Record with cuts in mind.** If you flub a line, just pause and start the line again — repeated takes are exactly what Magic cut is built to resolve. Long pauses between takes actually help, both for detection and for accurate trimming.
* **Iterate freely.** The transcript *and* the cut analysis are both cached, so re-running Magic cut on the same source is fast and consistent — experiment with review mode without waiting.
* **Treat the output as a strong first cut.** It's designed to be edited: every enhancement is a normal clip on its own track, so you can tweak or remove any piece in Premiere.

### 🙋‍♂️ Common questions, issues, watch-outs

* **"It cut a take I wanted" / "It kept the wrong take"** → Turn on *Review cuts before applying* and re-run — the analysis is cached, so it's quick. In the review panel, drag-select to force-keep the take you want, or delete/resize sections before building.
* **"The transcription language is wrong"** → Set **Advanced options → Audio language** explicitly, then use **\[…] → Re-transcribe and Go** so the clip is transcribed again in that language (the saved transcript is reused otherwise).
* **"Why didn't I get captions/titles?"** → Captions and titles are part of the *shortform* treatment.
* **"A title covers words that are missing from the captions"** → That's intentional: the caption clips underneath each title are disabled, not deleted. To reject a title, delete the title clip on V5 and re-enable the disabled caption clips on V4.
* **"Will this change my original clip or sequence?"** → No. Magic cut builds a separate new sequence ("FireCut Magic - …"). The source is untouched.
* **"I ran it twice and got the same cut"** → The cut analysis is cached per source so results are consistent (and fast). Use **\[…] → Re-transcribe and Go** to start completely fresh.
* **"Can I run it on multiple clips at once?"** → Not yet — Magic cut works on one clip or sequence at a time. To edit a multi-clip shoot, assemble the clips into a sequence first and select that sequence as the source.
* **"There's an extra copy of my source clip in the FireCut bin"** → That's a side effect of how Premiere imports the assembled sequence; it's a reference to the same media file (not a duplicate on disk) and the new sequence depends on it, so leave it in place.
* **"The transcript has mistakes"** → Because audio is transcribed automatically, some errors are expected, and the cut analysis is built to tolerate them. If results are badly off, check the audio is clean and the language is right; if it persists, share your audio with us at <support@firecut.ai>.