# Create your first rough cut

> Every control on the New Rough Cut screen, and the safest defaults for a first successful result.
>
> Verified against the current RoughCut app on 28 August 2026
> https://www.roughcuteditor.com/docs/first-rough-cut

Choose New Rough Cut in the sidebar, drop in a video, and leave the recommended defaults: Standard recording environment, On-device transcription, Bad-take removal on, Second Look off, Auto-detect language, and no video goal. Click Process and leave your source file where it is. RoughCut never copies or re-encodes your footage — the exported timeline references the original in place. A completed rough cut spends one of your five included runs, but only when it actually finishes: failed and cancelled jobs stay free, and Transcribe Only never counts. Every other control on the screen has its own page for when the defaults are not right for your recording.

## Prerequisites

- The speech model installed, unless you are using Pro [Cloud transcription](/docs/transcription-options). See [Install the local speech model](/docs/speech-model).
- An AI provider key **only if** you want bad-take removal. Silence-only cuts need no key.
- A supported video file. See [Choose video and audio inputs](/docs/file-inputs).

## Steps

1. Choose `New Rough Cut` in the sidebar, under `Studio`.
2. Drop a video onto the drop zone, or click `Browse Video…`. The zone reads `Drop a video here`, and `Drop to replace the video` once a file is chosen.
3. Optionally click `Browse Audio (optional)…` if you recorded a **separate, cleaner microphone track**. RoughCut analyses and transcribes that audio while the exported timeline still references your original video.
4. Leave `Recording environment` on `Standard`. That is −25 dB and suits typical home or office recordings.
5. Leave `Transcription` on `On-device`. Nothing is uploaded for transcription on this path.
6. Leave `Bad-take removal` on if you have a provider key. Turn it off for a silence-only cut that needs no key.
7. Leave `Second Look` off for a first run. It is a Pro refine pass and adds another AI call.
8. Leave `Fine-tune` collapsed. Its defaults are 0.5 s minimum silence, `Standard` aggressiveness, `Auto-detect` language, and no video goal.
9. Click `Process`.
10. Keep the source file where it is until you have exported and imported the result.

## Expected result

The job moves through its stages and finishes with a review list of proposed cuts, then a `Your rough cut is ready` banner with an `Export for Final Cut Pro…` button.

## The controls, in one place

| Control | Default | What it changes | Pro |
|---|---|---|---|
| `Recording environment` | `Standard` (−25 dB) | How quiet counts as silence. `Quiet room` −35, `Noisy` −20, `Custom dB`, or `Smart` | `Smart` only |
| `Transcription` | `On-device` | Whether audio is transcribed locally or by OpenAI | `Cloud` only |
| `Bad-take removal` | On | Whether the AI editor pass runs at all | No |
| `Second Look` | Off | A second AI read of the surviving footage | Yes |
| `Min silence (sec)` | 0.5 | Pauses shorter than this survive | No |
| `Bad-take aggressiveness` | `Standard` | How boldly the editor removes suspected bad takes | No |
| `Language` | `Auto-detect` | Transcription language | No |
| `Video goal (optional)` | empty | Editorial context, max 500 characters | Summarising only |

`Second Look` is disabled unless `Bad-take removal` is on, because it re-reads the result of the first AI pass.

## Your original file is never touched

RoughCut states this on the screen itself: files are never copied, and the FCPXML references your original file in place. Nothing is transcoded, nothing is re-rendered, and no second copy of your footage is created. What RoughCut produces is a set of cut decisions.

## How the trial counts

While you have not purchased the unlock, the screen shows `<n> free cuts left`. Five complete rough cuts are included.

A run is spent **only** when a job actually reaches a completed state. Specifically:

- **Failed jobs do not count.**
- **Cancelled jobs do not count.**
- **Transcribe Only never counts**, and is never gated by purchase at all.

The counter lives in the macOS Keychain, so it survives deleting and reinstalling the app, and nothing in the app lowers it.

## Common mistakes

- **Adding a separate audio file that is not synchronised.** The audio track is used for analysis and transcription; if it drifts from the video, every cut time drifts with it.
- **Moving or renaming the source file before importing.** Final Cut Pro then reports the media as offline.
- **Turning on `Second Look` for a first job.** Learn what the standard pass does before adding a refine pass.
- **Setting a custom dB value before trying the presets.** The three presets cover most recordings.
- **Expecting `Process` to work with no model and no Pro.** On-device transcription needs the model; the button stays disabled and tells you so.

## Troubleshooting

`Process` disabled? Check in order: a video is selected; no red validation message is showing; the speech model is ready (or `Cloud` is selected); you have free cuts remaining or have purchased the unlock.

If the job fails, read the message before deleting anything — it names the cause. See [Troubleshooting by symptom](/docs/troubleshooting).

![The RoughCut New Rough Cut screen with its media drop zone and cut controls.](/assets/docs/en/studio.webp){1600x1000}
Caption: The New Rough Cut screen. The defaults shown are the recommended first run.