Skip to content

Cutting a clip from a recording

Turning a stretch of a long recording into a short, captioned video for each platform - and reading what a narrower shape costs before you commit to it.

Updated

The clips page takes one stretch of one recording and produces a short video project for each platform you name, with the words burned in, an opening line over the head and a call to action over the tail. Describing a clip writes nothing and charges nothing. Cutting one writes the storyboards and one immutable composition version per shape - which is exactly what a render takes.

It does not render. That is deliberate, and it is why there is no "Render it" button on this page: a clip inherits the one render lifecycle the product already has, with its idempotency, its credit reservation, its retry history and its cancellation, instead of growing a second one beside it.

BEFORE YOU START

  • A video in a workspace's media library. Nothing else can be clipped, and a picture or an audio file is refused by name.
  • A transcript of that recording, if you want captions. Captions are cut out of a stored transcript and out of nothing else. This build has no transcription integration at all, so unless a transcript reached your organisation some other way, turn captions off.
  • The "view" or "generate" permission to describe a clip; "create" or "edit" to cut one.
  • Nothing else. Neither describing nor cutting costs credits.

STEPS

  1. Open /app/clips. The quickest route is /app/transcripts: open a transcript, choose "Look for clip moments", and choose "Cut this moment" on one of them - the recording, the transcript and both timecodes arrive filled in.
  2. Choose the workspace, then the recording. Only videos are listed.
  3. Choose the transcript the captions come from, or leave it on "No transcript" and turn captions off.
  4. Set the in and out points in seconds, and give the clip a title.
  5. Add an opening line and a call to action if you want them. Each becomes its own scene over the same footage.
  6. Choose how the clip is fitted to a shape that is not the recording's. The two options and their descriptions come from the renderer, not from this page.
  7. Tick every platform you want it cut for. One video project and one composition version is produced per platform.
  8. Choose "Show me what this would be". Nothing is written. Read the length, the caption standing, and what each shape costs in picture.
  9. Choose "Cut it".

WHAT YOU SHOULD SEE

A plan naming the length, the language, why this stretch was chosen, and a row per shape saying either how much of the picture that shape cuts away or - honestly - that the recording's pixel size was never recorded so the cost cannot be worked out. After cutting, the clip appears in the list with one rendition per shape, each carrying its own state, the composition version id and the delivery profile a render needs.

WHAT THIS WILL NOT DO

  • It does not look at the footage. A clip reshaped for a narrower frame is cropped from the centre. If the speaker is standing off to one side they will be cropped out, and no face or subject tracking exists here to prevent it.
  • It does not guess what it cannot measure. "assets" carries a width and a height only when something recorded them, and they are empty for most of the library. When they are, the page says the cost was not measured rather than printing a plausible percentage.
  • It does not tell you whether a clip will do well. There is no score, no rank and no ordering by anything but recency. A clip surfaced by the transcript matcher says which signals surfaced it; a clip you chose yourself says nobody surfaced it, rather than borrowing reasons it never had.
  • It does not deliver a silent video when you asked for captions. A clip asked for with captions it cannot have is refused, with the reason.
  • It does not caption words that were removed at a speaker's request. Those segments are left out, and the number left out is reported so you can see whether the clip is mostly redaction.
  • It does not render, publish or schedule anything. Turning a rendition into a video is POST /api/v1/render-jobs with modality "composition"; accepting the finished bytes is the promote step on the renders page; putting the result on a post is the content editor.
  • It has no screen in the Flutter app.

IF IT DOES NOT WORK

  • "That recording is not in this organisation's media library." means the id belongs to another tenant or has been removed.
  • "That transcript is of a different recording, so its words are not what is being clipped." A transcript that records no source recording at all is accepted, because most transcripts here were stored against a render job or a project.
  • A refusal naming a platform and a length means that network will not take a video that long. Shorten or lengthen the clip, or drop that platform.
  • If the page says it could not read what this deployment can do, the cutting controls are hidden on purpose. Which platforms exist, which shapes they take and how long a clip may run are all the server's answers, and nothing here keeps its own copy of them - so rather than offer you shapes that may be wrong, it offers nothing. Reload.

COMMON QUESTIONS

Why does it show me the plan before it will let me cut?

Because a refusal found on the pre-flight costs you a sentence, and the same refusal found later costs you a row in your render history and a job that never had a chance. It is also the only moment you can see what a 9:16 crop takes out of a 16:9 recording before you commit to it.

Why is there no render button?

Because there is no clip render endpoint, and adding one would mean a second render lifecycle. Each rendition hands you the composition version id, the document hash and the delivery profile, which is the whole payload the ordinary render submission takes.

Why does my clip say nothing ran and nothing was charged?

Because no compositor is active in this build. The clip is real - the storyboards and the composition versions exist and are immutable - but nothing can turn them into a video here yet.

  • Finding clip moments in a transcript

    Reading and correcting a machine transcript, and asking it for places worth a look - which is a text search, not a judgement.

  • Rendering a composition

    What submitting a cut version does, why every deployment currently answers "no provider active", and what that costs.

  • Following a render

    The renders page, what each render state means, and the rule that a render is charged only when a provider produced something.