Audio & video · checked 2026-10-14

Can I vibecode Descript?

KINDA· weekend project

Descript's central trick is easy to state: transcribe the recording, let you strike out words, and cut the audio or video to match. A local app on faster-whisper for word timestamps and ffmpeg for the cuts does that in a weekend, including silence trimming and filler-word removal. The pieces around it are what you will not rebuild quickly: Studio Sound's audio clean-up, AI voices and avatars, screen and remote recording, a multitrack timeline and shared projects. For a solo podcaster who mostly cuts speech the local version covers most days; for polished video work it falls short.

$24/moprice per seat

$288you'd save a year

weekendbuild time

—people replaced it

Before you start

What you need on hand.

DifficultyIntermediate · Some system setup or keys.

TimeAbout an hour to transcribe and cut; a weekend to tune the exports

Claude Code, Codex or CursorAny of them can run the prompt.
Python 3.11, ffmpeg and Node.js 22ffmpeg does every cut and export, so install a recent build.
A 10 to 20 minute recording to test withClear speech with a few mistakes you would normally cut.
Your Descript exportsThe media files plus the transcript or subtitle files from each composition.

Before you buildDescript Free covers 1 media hour a month with 720p exports and a watermark, which may be enough for an occasional cut.

1. Install one agent: Claude Code, Codex, Cursor.
2. Make an empty folder and open it in the agent.
3. Paste the prompt below, press enter, and answer its questions as it builds.

The prompt

Build me a local transcript-based audio and video editor to replace Descript for cutting speech. Requirements:

- Python 3.11 FastAPI backend and a React + Vite front end. `./run.sh` starts both and serves
  http://127.0.0.1:4900. ffmpeg must be on the PATH; exit with a clear message if it is not.
- Import: drop in an mp4, mov, wav or mp3. Copy it to ./data/projects/<id>/ and extract 16 kHz mono
  audio with ffmpeg.
- Transcription: faster-whisper with word timestamps (model from .env, default `small`). Store words
  (text, start, end, deleted flag) in SQLite at ./data/projects.db.
- Editor: the transcript as clickable words beside an HTML5 player. Click a word to seek; select words
  and press Delete to strike them out; 50 levels of undo. Playback skips struck-out ranges.
- Clean-up: "Remove filler words" (um, uh, you know; list editable in .env), "Shorten pauses" (gaps over
  0.7 s become 0.3 s) and "Remove selected range". Show how many minutes each action saved.
- Export: build a cut list from the kept words (merge ranges less than 80 ms apart) and render mp4 or
  mp3 with ffmpeg, with a 30 ms audio crossfade at every cut. Also export an .srt of the kept words and
  the plain transcript as .txt.
- Check: the exported duration must match the kept time within 0.2 s; show both numbers.
- Single user. No accounts, no telemetry, local-first; models download once from Hugging Face.
- Out of scope: noise removal, AI voices and avatars, multitrack timelines, screen or remote recording,
  burned-in captions.
- README: install steps for ffmpeg and Python, how long a 10 minute file took to transcribe on this
  machine, where projects live and how to back them up.

Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.

Share on X

Max MiskowSponsoredAI agents, automations and MVPs, built for your company.Visit site ›

Then keep going

Next prompts, same folder.

The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Descript.

Add burned-in captions to the transcript editor in this folder. Requirements:

- Build an ASS subtitle file from the kept words, grouping them into lines of at most 7 words or 2.5 seconds.
- Offer three styles in the UI (font, size, colour, position) and an option to highlight the word
  being spoken with ASS karaoke tags.
- Render with ffmpeg's `subtitles` filter and show a 10 second preview before the full export.
- Support 16:9 and 9:16 outputs; for 9:16 crop to the centre.
- Update the README with the font search path.

Add an audio clean-up step to the transcript editor in this folder. Requirements:

- A Clean audio switch in the export dialog that runs DeepFilterNet if installed, otherwise ffmpeg's
  `afftdn` plus `loudnorm` to -16 LUFS.
- Preview: render a 15 second sample before and after and let me flip between them in the player.
- Never touch the original file; write the cleaned audio next to the project.
- Show the loudness before and after in the export summary.
- Update the README with how to install DeepFilterNet and a note that results trail Studio Sound.

Add a clip finder to the transcript editor in this folder. Requirements:

- A button that sends the transcript, with timestamps, to a model through ANTHROPIC_API_KEY in .env and
  asks for up to 8 self-contained segments of 30 to 90 seconds, each with a one-line title.
- Snap every segment to sentence boundaries, list them in a side panel, and export any of them as a
  clip through the existing export code.
- Show the token count before the call and ask for confirmation above 20,000 tokens.
- Update the README with what is sent to the API.

Add a transcript importer to the editor in this folder. Requirements:

- Import an .srt or .vtt file next to a media file instead of running faster-whisper.
- Align each cue to words with WhisperX forced alignment so every word gets its own timestamp.
- If alignment fails for a cue, keep the cue as one block and mark it in the UI.
- Show a summary of cues imported, aligned and kept as blocks.
- Update the README with the Descript export steps for subtitles.

What you lose

Studio Sound, the one-click clean-up of noisy recordings
Typing a replacement word and hearing it in a cloned voice, plus AI avatars
Recording remote guests or your screen inside the same app
A multitrack timeline with styled captions and a stock media library
Shared projects, comments and team brand templates

Why people still pay

Editors pay to cut spoken content at reading speed: fix a stumble by deleting a sentence, strip filler words from an hour of tape, then make clips with captions without leaving the app. The recording, the clean-up, the voice tools and the sharing sit around that one move, so the subscription saves an export-and-reimport round trip on every episode.

Moatexecution polish proprietary models

Do the math

Descript or your own copy?

Seats

Over

Your usage

Count my time12h ×$/h

Descript, 1 yr$288

Your own copy, 1 yr$40running + build session

You keep$248pays for itself in the first month

Estimate · Free on your own computer. A laptop CPU transcribes large models slower than real time, so use a smaller model or rent a GPU by the hour, which is the only running cost. The build is token usage on your agent plan.

Leaving Descript

Take your data, then cancel.

1 · Take your data out of Descript

1Video or audio: open the composition, click Export in the top right, set the destination to Local export and pick the Video or Audio tab. A Free drive adds a watermark and caps video at 720p.
2Text: the same Export panel writes the transcript as text and subtitles as caption files, one composition at a time.
3Timeline: use the timeline export to keep editing in another tool; the help center does not name the formats, so test with your editor before you cancel.
4There is no one-click backup of a whole Drive in the help center, so repeat the export for each composition you want to keep.

2 · Cancel Descript

1Open App Settings (Command or Control plus comma), pick the right Drive if you own several, and go to the Plan tab.
2Click Change plan, choose Downgrade to Free, select a cancellation reason and click Cancel subscription.
3Only the Drive owner's email can cancel. If the owner has left, contact Descript support.

Paid access lasts to the end of the billing cycle, then the Drive moves to Free. Projects stay, but features outside the Free plan stop working.
You can cancel a pending downgrade from the Plan tab before the cycle ends.
Top-up balances are kept after a downgrade but cannot be used on Free, and they still expire 12 months after purchase.
Exports made after the downgrade carry the Free watermark, so export what you need first.
The cancellation page says nothing about refunds, and a third-party cost review states that Descript does not prorate them, so cancel before renewal.

Checked 2026-10-14 against Descript's docs: 1, 2, 3, 4, 5

The escape hatch

Don't feel like building it? These are already free.

KdenliveOpen sourceFull video editor whose Speech Editor transcribes clips with Whisper and builds timeline edits from selected text.★ 5.9k · active Oct 2026 · one-click install
macOS, Windows, LinuxVisit ↗CutScriptOpen sourceEarly-stage Descript-style editor run from source: delete a word in the transcript and the clip is cut.★ 262 · active Mar 2026 · self-host, real ops
macOS, Windows, LinuxVisit ↗LosslessCutOpen sourceCuts and joins audio and video without re-encoding, which is fast but has no transcript.★ 45k · active Oct 2026 · one-click install
macOS, Windows, LinuxVisit ↗

The numbers

Descript pricing

Plan	Monthly	Annual, per month	What you get
Free	$0	$0	1 media hour a month, 100 AI credits (one-time), 720p export with a Descript watermark, 5 GB of cloud storage.
Hobbyist	$24	$16	10 media hours and 400 AI credits a month, 1080p watermark-free export, 100 GB of storage, one person.
Creator	$35	$24	30 media hours and 800 AI credits a month, 4K export, 1 TB of storage, a team of up to 3 billed separately.
Business	$65	$50	40 media hours and 1,500 AI credits a month, a team of up to 5, Brand Studio, dubbing in 30+ languages.
Enterprise	custom	—	Hours and credits sized to the team, SOC 2 Type II, SSO, SCIM, audit logs.

Free tier: 1 media hour a month, 100 AI credits once, 720p with a watermark.
Billing: Monthly or annual; annual saves up to 35%. Cards are charged automatically at the start of each cycle, and invoicing such as ACH needs Enterprise sales.
Hidden costs: Every minute of imported or recorded media counts against the media-hour cap, and AI tools such as Studio Sound, filler-word removal and avatar video spend AI credits. Top-ups for more hours and credits are sold on Creator and Business only, and unused top-up balance expires after 12 months. Extra team members are billed separately per person.
Source: www.descript.com

Also on the list

Riverside$29/mo · PodcastingKINDA

Loom$18/mo · Screen recordingKINDA

HeyGen$29/mo · AI videoNOT REALLY

Questions

Descript, briefly

Not answered here? Tell us and we'll add it.

Ask a question ›

Free gives 1 media hour a month, a one-time 100 AI credits and 720p exports with a watermark.

Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: Studio Sound, the one-click clean-up of noisy recordings, Typing a replacement word and hearing it in a cloned voice, plus AI avatars.

About $24 a month on the Hobbyist plan per seat, $288 a year (checked 2026-10-14).

Studio Sound, the one-click clean-up of noisy recordings; Typing a replacement word and hearing it in a cloned voice, plus AI avatars; Recording remote guests or your screen inside the same app; A multitrack timeline with styled captions and a stock media library; Shared projects, comments and team brand templates. If any of those matter to you, keep paying.