# Write a useful video goal

> One optional sentence that tells the AI editor what the video is for, so it can judge which digressions are tangents.
>
> Verified against the current RoughCut app on 28 August 2026
> https://www.roughcuteditor.com/docs/video-goal

Video goal is an optional field inside Fine-tune, capped at 500 characters. One sentence saying what the finished video should deliver gives the AI editor the context it needs to tell a genuine tangent from material that belongs. Name the format, the audience, and the outcome; do not paste a script, and do not ask RoughCut to add anything, because it only ever removes. Over the limit, Pro subscribers get a Summarize with AI button that condenses the goal with their own Gemini or OpenAI key, targeting 440 characters and retrying once before failing visibly rather than truncating silently. Everyone else sees the validation error and cannot process until the text fits. Editing the field clears the previous error.

## Prerequisites

- Nothing. The field is optional and every job runs fine without it.
- It only has an effect when `Bad-take removal` is on, because it is context for the AI editor pass.

## Steps

1. On `New Rough Cut`, expand the `Fine-tune` section. It is collapsed by default.
2. Find `Video goal (optional)`. Its placeholder shows the shape of a good answer: `e.g. Explain the new eval workflow end to end`.
3. Write **one sentence** describing what the finished video should deliver — the format, who it is for, and what the viewer should leave with.
4. Keep it under **500 characters**. The field accepts a few lines, but length is not the point; specificity is.
5. Check that no red validation message is showing beneath the controls.
6. Click `Process`. When a goal is present, the collapsed `Fine-tune` summary line ends with `goal set`, so you can confirm at a glance on later jobs.

## Expected result

The job runs as usual. The difference shows up in the review list: fewer good passages proposed as `Tangent`, and better judgement about which digressions genuinely wander away from the point.

## What the goal is for

The app's own help text states the purpose exactly:

```text
One sentence about what the video should deliver — helps the editor judge tangents.
Max 500 characters.
```

Most removal categories need no context. A verbatim retake is a verbatim retake whatever the video is about. **Tangent is the exception.** Whether a two-minute detour into pricing is a tangent depends entirely on whether the video is about pricing — and without a goal, the editor has to infer the subject from the transcript alone, which is exactly the material the digression is embedded in.

One sentence resolves that ambiguity. It is the highest-value optional field on the screen for anyone whose recordings wander.

## Good goals

Each of these is one sentence, names the format, and says what the viewer should end up with.

| Video type | A goal that works |
|---|---|
| Tutorial | `Teach a new user how to set up the app and export their first file, start to finish, with no prior knowledge assumed.` |
| Product review | `An honest review of this microphone for podcasters, covering sound quality, build, and whether it is worth the price.` |
| Corporate update | `A quarterly update for the whole company: what shipped, what slipped, and what the next quarter's priorities are.` |
| Course lesson | `Lesson four of a beginner photography course, covering aperture and depth of field, with the exercise at the end.` |
| Product demo | `A demo for prospective customers showing the reporting dashboard solving a real reporting problem end to end.` |

What makes them work:

- **The format is named** — tutorial, review, update, lesson, demo. That alone tells the editor what structure to expect.
- **The audience is named** — a new user, podcasters, the whole company. Digressions that serve that audience are not tangents.
- **The outcome is named** — what the viewer should be able to do or decide afterwards.

A goal that is merely a title (`Microphone review`) is better than nothing, but it does little work. Add the audience and the outcome and it starts earning its place.

## What not to write

> ! **Do not paste your script.** The transcript already contains everything you said. A pasted script is a long block of text that will not fit, adds no information the editor lacks, and wastes the space where a clear statement of intent could have gone.

> ! **Do not ask RoughCut to create anything.** RoughCut only removes; it never adds, rewrites, reorders, or generates. Instructions such as "add an intro", "make it punchier", "write a summary at the end", or "rearrange this into three sections" cannot be carried out, and phrasing the goal as an instruction makes it worse at its actual job of describing the video.

Also avoid:

- A list of every topic covered. That describes the recording, not the goal, and gives the editor no way to rank anything.
- Notes to yourself about the shoot — which take was good, where the dog barked. That is not editorial intent.
- Anything in a different language from the recording. Write the goal in the language you are speaking.

## Over the 500-character limit

The cap is a hard limit of **500 characters**. Over it, the job will not process:

```text
Video goal must be 500 characters or fewer.
```

What happens next depends on your plan.

### With RoughCut Pro

An over-limit footer appears with the current count:

```text
Over the 500-character limit (<count>).
```

Alongside it is a `Summarize with AI` button, shown when you have a usable Gemini or OpenAI key. Clicking it condenses the goal using **your own key** — RoughCut has no server in the path. While it runs, the button shows `Summarizing…` and the goal field is disabled.

How it behaves:

- It **targets 440 characters**, deliberately under the 500 cap, so ordinary variation in the model's output still lands inside the limit.
- It keeps the goal in the **same language** it was written in, and preserves the concrete specifics an editor needs — topic, audience, what must stay or go.
- If the first attempt is still too long, it **retries once, instructed to cut harder**.
- If the second attempt is still over the limit, it **fails visibly** rather than quietly trimming your sentence:

  ```text
  The model could not fit the goal into 500 characters (got <n>). Try again, or trim it by hand.
  ```

  Nothing is silently truncated. You always know whether the text in the field is what you meant.
- If you edit the goal while the summariser is running, your edit wins. A newer edit is never overwritten by a result that arrived late.

With no usable key, the button is replaced by a hint:

```text
Summarizing needs your Gemini or OpenAI API key.
```

with a link to add it in Settings. See [Create and add a Gemini API key](/docs/gemini-api-key) or [Create and add an OpenAI API key](/docs/openai-api-key).

### Without Pro

There is no summariser. You see the validation error, and `Process` stays unavailable until the goal fits. Shorten it by hand — which, for a field that is meant to hold one sentence, is usually the faster route anyway.

### Clearing the error

**Editing the field clears the previous error.** Both the validation message and a summariser failure disappear as soon as you type, so you are never looking at a stale complaint about text you have already changed.

## Common mistakes

- **Pasting the script.** The single most common misuse, and the one the character limit exists to prevent.
- **Writing instructions instead of a description.** RoughCut removes; it does not add.
- **Listing topics.** Say what the video is *for*, not what it contains.
- **Writing the goal in a different language from the recording.** Match the spoken language.
- **Fighting the limit.** If one sentence will not fit in 500 characters, the sentence is doing too much.
- **Expecting a goal to help a silence-only cut.** With `Bad-take removal` off, no model reads the transcript, so the goal has nothing to influence.

## Troubleshooting

| Symptom | Cause | Fix |
|---|---|---|
| `Video goal must be 500 characters or fewer.` | Over the cap | Shorten it, or use `Summarize with AI` if you have Pro and a key |
| No `Summarize with AI` button appears | Not a Pro subscriber, or no usable Gemini or OpenAI key saved | Shorten by hand, or add a key in Settings |
| `Summarizing needs your Gemini or OpenAI API key.` | Pro, but no key saved | Add a key in Settings |
| `The model could not fit the goal into 500 characters (got <n>). …` | The goal is too dense to compress further | Trim it by hand — usually a sign it is a script, not a goal |
| Good material still removed as `Tangent` | The goal is too vague, or omits the audience | Rewrite it naming the format, the audience, and the outcome, then process again |
| The error stays on screen | It clears on the next edit | Type in the field |