Playbook

How to Clip With Descript: A Text-Based Workflow for Clippers

How to clip with Descript: search the transcript, cut around the line, fix audio pops, build a 9:16 scene, proof captions, export one range. Worked time math.

By Mark WalnutPublished 8 min read

Key takeaways

  • In Descript you clip by searching the transcript, cutting the words around your line, then putting the result in a 9:16 composition with captions.
  • Descript wins on finding the moment in a long talk source; on a 90-minute file the modelled saving is about 6 minutes per clip against a phone editor.
  • Delete ripples the media and Ignore keeps the text recoverable; use Ignore while you work and Delete only once the cut is final.
  • Leave 100 to 200 ms of room tone at every join and you remove most mid-sentence pops.

To clip with Descript, import the long video, let it transcribe, search the transcript for your line, cut the words before and after it, put the result in a 9:16 composition, fix the captions and export. The transcript is the timeline. Cutting a clip means editing text.

That is also the limit. Descript wins when the hook is a sentence you can quote. For a sports play, a reaction face or a visual gag, a timeline or phone editor is quicker. See How to Clip With CapCut for that side.

A note on what is verified. I have not been able to confirm every menu path and shortcut against the current Descript build while writing this, and Descript renames things between releases. Where I give a shortcut or path, check it against Help and the keyboard shortcut list before you build a routine on it. Where I give a number, I say whether it is arithmetic or a rule of thumb.

A walkthrough that shows the failures

This is a constructed example, not a screenshot from a client project. The audio is a made-up podcast line. The point is the order of mistakes, because these are the ones that actually happen with transcript editing.

The transcript reads:

"...so honestly look I think we got the pricing wrong at launch and it cost us two years of growth, and Dan's team, the Dune Labs guys, said the same thing..."

Attempt 1: wrong cut point. You search "pricing wrong", land on the hit and select from "so honestly" to "growth". Play it. The clip opens on filler and sounds like a man warming up. The hook is "we got the pricing wrong at launch", and "so honestly look I think" delays it by about two seconds. In short-form that is the difference between a held viewer and a swipe.

Correction. Start the selection at "we got". Keep the end at "growth". Now the first spoken word is the claim.

Attempt 2: misheard word. The transcript shows "Dune Labs" but the speaker said "Dyn Labs". Captions will print the wrong brand name in 70 pt white letters. Correct the word in the transcript, not on the caption layer, so the fix carries into every caption style.

Attempt 3: the join pops. You cut "...growth" and the next sentence, deleting the gap word "and" in the middle. The join clicks. This is the audio fix below.

If a proper name is mis-transcribed once, search the transcript for other spellings before you export. The same speaker's name tends to be wrong in several places.

The editing moves, with the Descript names

These are the terms Descript uses, with the caveat above on exact labels.

  • Project and Composition. Your imported file lives in a project. A composition is the canvas you output from. Make a separate composition per clip so the master stays untouched.
  • Scene. The visual block in a composition. A scene holds your layout and captions.
  • Layout. Controls where the speaker video sits in a scene. Choose a single-speaker layout and place the face in the upper-middle of the frame.
  • Captions. A caption style applied to the scene, generated from the transcript.
  • Underlord. Descript's AI assistant, which can propose edits including candidate clips and filler removal. Treat its output as a list to review, not finished posts. How to Clip With Opus Clip covers why auto-picks need a scoring rule.

Find. Cmd/Ctrl+F searches the transcript. Step through hits with Enter.

Delete versus Ignore

This is the detail most guides skip.

  • Delete removes the words and ripples the media: later material slides up to close the gap.
  • Ignore strikes the words through. They stay in the document and can be restored, but they are skipped on playback and not exported.

Use Ignore while you are deciding where the clip starts and ends. You can bring a hook word back if the cut feels wrong. Switch to Delete only when the cut is final. If you remember one rule: Ignore to explore, Delete to commit. Confirm the Ignore shortcut in the shortcut list for your version.

Split versus ripple delete

A split divides a clip without removing anything, so nothing shifts. A ripple delete closes the gap. If you want a hard cut to a different shot, split and delete the middle. If you want the speaker to sound continuous, a ripple delete is what you want, and it is also what creates the join you then have to fix.

Fixing mid-sentence pops

A pop appears when a delete lands on a nonzero point in the waveform or removes the breath and room sound that normally sits between words.

The rule I use, a rule of thumb and not a Descript spec:

  1. Leave 100 to 200 ms of room tone at every join. Do not trim to the first consonant. Keep the tiny silence before the next word.
  2. Under about 100 ms of gap, expect audible clicks. If your join sounds cut too tight, extend the neighbouring word edges a touch.
  3. Over about 200 ms, it sounds like a pause. That is fine if it fits the speaker's rhythm, and wrong if you were trying to hide a cut.
  4. If a join still clicks, cover it with a scene change or a zoom so the jump reads as intentional, or use Descript's gap-handling options if your version offers them. Menu name differs by release.

Play every join at least twice with headphones. A pop you cannot hear on laptop speakers is obvious on a phone.

Setting up the 9:16 composition

  1. Make a new composition and set the aspect ratio to 9:16, 1080 x 1920.
  2. Add a scene and choose a single-speaker layout.
  3. Position the face in the upper-middle. Keep it above the bottom 250 px.
  4. Apply a caption style, then override it with the values in the table.
  5. Save it as a template composition and duplicate it for every clip.
Setting Starting value (rule of thumb)
Font size on 1080 x 1920 about 64 to 80 px, test at phone size
Words per line 3 to 4, two lines max
Bottom margin about 250 px clear
Top margin about 150 px clear
Side margins about 60 px
Base and highlight colour white with dark outline, highlight in a clearly different colour

These are derived from the shape of Reels, not platform specs. Check the platform's current layout and test on a real phone.

Exporting one clip without touching the master

  1. Work in a duplicated composition, one per clip. Delete and ignore freely there.
  2. Name it by campaign, source and timestamp so you can find it later.
  3. Open Publish and export that composition only, in MP4 at 1080 x 1920.
  4. Confirm the export range options in your version. Some builds let you export a selected range; if yours does not, the duplicated composition does the same job.

The master composition and the original media are never edited, so a bad cut costs you a duplicate, not the episode.

The time math: 90 minutes, five clips

This is arithmetic from stated assumptions, not a measured benchmark. Replace the inputs with your own stopwatch numbers.

Assumptions: one 90-minute source, five clips wanted, transcription runs in the background and does not count as your time.

Descript

  • Find: roughly 10 searches ("I was wrong", "nobody told me", numbers, names) at about 45 seconds each = 7.5 minutes for the batch, or 1.5 minutes per clip.
  • Cut with Ignore then Delete, join check: 3 minutes per clip.
  • Caption proofread: 4 minutes per clip.
  • Export and phone check: 4 minutes per clip.
  • Total: about 12.5 minutes per clip, 62.5 minutes for five.

CapCut on the same file

  • Find: one pass at 2x speed to mark moments is 45 minutes for the file, shared across five clips, so 9 minutes per clip.
  • Cut: 3 minutes per clip.
  • Captions and proofread: 4 minutes per clip.
  • Export and check: 4 minutes per clip.
  • Total: about 20 minutes per clip, 100 minutes for five.

So on a talk-heavy 90-minute source the modelled gap is about 7.5 minutes per clip, and almost all of it is the find step. On a 10-minute source where you already have timestamps, the find step is near zero in both tools and the gap disappears. Run three real clips through each and keep whichever wins on your source type.

Plans and limits: what a clipper actually needs

Descript's tier names, hours of transcription, export caps and watermark rules change often, and I cannot confirm the current numbers from here. So I will not print figures I cannot stand behind. Use this test instead:

  1. Watermark-free export. If the free plan puts a Descript mark on exports, it fails for campaign work, especially logo campaigns where the brand placement is what you are paid for.
  2. Transcription hours against your weekly source. Add up the hours of source you load per week. If that exceeds the plan's monthly allowance, you will hit a wall mid-campaign.
  3. 1080p vertical export. Anything lower is a rejection risk for Reels.
  4. AI credits. Only matters if you lean on Underlord for picks and filler removal.

If a free plan passes tests 1 to 3, use it. If not, the lowest paid tier that passes them is what you need, and a higher tier is only worth it when the weekly hours or AI credits force it. Run one real export before you commit to a campaign deadline.

Failure modes and fixes

  • Studio Sound over-processing. Can make clean audio thin or watery. Turn it off for that clip and compare against the original.
  • Crosstalk. When speakers overlap, the transcript mangles words and captions follow. Correct the text and speaker labels by hand, or skip the stretch.
  • Slow project on a 2-hour file. Keep one project per episode and work on a duplicate composition, not the master.
  • Over-aggressive filler removal. Remove filler words lightly, review the result, and restore any pause that is the joke.
  • Captions confidently wrong. Read each line against the audio with the caption layer on.

Before you post

  1. The first spoken word is the claim.
  2. Every caption matches the audio, including names.
  3. No caption overlaps any logo or required element in the campaign brief.
  4. Nothing important sits in the bottom margin.
  5. The clip ends on a finished thought.

For a first-week plan, see How to Start Clipping. For choosing what to pull from a long show, read Podcast Clipping.

Descript saves you the search, not the judgment.

When the workflow is stable, join free at FindClout and match it to a campaign that fits your source type.

Frequently asked questions

How do you clip with Descript?

Import the source, let it transcribe, search the transcript for your line, cut the words before and after it, put the result in a 9:16 composition, check the captions against the audio and export. The transcript is the timeline, so cutting a clip means editing text.

Is Descript good for clippers?

It is good when the hook is a sentence you can quote: podcasts, interviews, commentary streams. It is slower than a timeline or phone editor for sports plays, reaction faces and anything where the picture carries the clip.

Can Descript make vertical 9:16 clips?

Yes. Create a composition, set its aspect ratio to 9:16, add a scene and position the speaker with a layout. Preview at phone size before exporting, and keep faces out of the bottom of the frame where Instagram's interface covers the video.

Does Descript add captions automatically?

It builds captions from the transcript, so transcription errors become caption errors. Names, tickers, slang and numbers fail most often. Fix the words in the transcript, not on the caption layer, then read every line against the audio before export.

What Descript plan does a clipper need?

The deciding test is not price. You need watermark-free export, transcription time that covers the hours of source you load per week, and 1080p vertical export. Those allowances change, so confirm them on Descript's pricing page against your own weekly source hours before paying.

Get paid for the views you already get.

Join FindClout free and apply once. We place your page on campaigns that fit it, and you get paid per verified view. Campaigns drop in the Discord first.

Get Aura Farm in your inbox

One email when a new piece drops: clipping, distribution and where attention is going. No spam.

Keep farming

Playbook

How to Start Clipping: A 7-Day Plan for Beginners

How to start clipping this week: a day-by-day 7-day plan covering accounts, campaigns, your first clips, posting cadence and getting paid on verified views.

7 min read · November 8, 2026
Playbook

Podcast Clipping: How Shows Grow Through Clippers

Podcast clipping turns long episodes into short, shareable moments that find new listeners. Here is the workflow, the rights you need, and how shows grow.

8 min read · November 8, 2026
Playbook

How to Clip With Opus Clip: Settings, Scoring Rule and Failure Fixes

How to clip with Opus Clip: pick the model, set length to 20-40s, test the Virality Score against real views, recut the hook in the transcript editor, then post.

7 min read · November 15, 2026
Playbook

How to Clip With InShot: A Phone-Only Workflow for Clippers

How to clip with InShot: exact taps for 9:16 canvas, blur fill, logo sticker and export, plus the pixel math that decides whether a logo clip passes review.

5 min read · November 14, 2026
Playbook

How to Clip With VN Editor: A Free Phone Workflow for Clippers

How to clip with VN editor: get footage in, cut, reframe to 9:16, caption, place a logo with pixel math and export clean. Settings, thresholds and a batch template.

6 min read · November 13, 2026
Playbook

How to Clip With Final Cut Pro: A Mac Clipper's Workflow

How to clip with Final Cut Pro: build a vertical 1080x1920 project, reframe 16:9 source with real math, cut the hook, caption, brand it, export and fix rejected uploads.

7 min read · November 11, 2026