AIOps SRE
Search

Descript · media-editor

Descript

Official documentation

Descript edits recorded audio and video through a transcript linked to the media, alongside timeline controls, scenes and captions. That makes it a different starting point from generating a new shot: import an authorized recording, correct its transcript, then remove or rearrange words while checking the resulting cuts. Studio Sound and the Underlord co-editor add cleanup and assisted editing, but the exported file still needs a full review for pacing and factual continuity. It is not one foundation model. Its current speech guide lists ElevenLabs Multilingual v2, v3 and v4; tone tags are supported by v3/v4. Media minutes and AI credits are distinct budgets, so an editing session should account for both imported footage and generated or assisted work.

Product profileDocumentation checked

VideoAudio

In this profile

When it fits

Useful for

  • Editing interviews, podcasts and recorded explainers through their transcript
  • Combining captions, scenes and optional synthetic narration in one media project

Limits to consider

  • Generating a wholly new cinematic shot as the primary task
  • Changing a recorded speaker’s meaning without factual and consent review

Capabilities and platforms

  • Transcript-linked word edits cut or rearrange the associated spoken media.
  • Scenes, timeline controls and captions organize a recorded explainer for export.
  • Text-to-speech supports ElevenLabs Multilingual v2, v3 and v4, with tone tags for v3/v4.

Where it runs

  • Web browser
  • macOS application
  • Windows application

Access and setup

Account with plan-dependent media minutes and AI credits. Underlord chat/reasoning and tool use can consume credits; current help says AI credits do not roll over.

Current access and pricing
  1. Open the official web or desktop editor and create a project; import an authorized recording or record a synthetic training script.
  2. Review speaker boundaries and correct names in the transcript before making editorial cuts.
  3. Check media-minute and AI-credit allowances; if generating speech, choose the model in App Settings > AI models and confirm the speaker permission.

Proposed exercise · not hands-on tested

A shorter synthetic onboarding explainer

Input

A self-recorded fictional training explanation with intentional pauses and a supplied correct script; it contains no customer records or other speaker’s voice.

  1. Import the recording, transcribe it and correct names/speaker boundaries against the supplied script.
  2. Remove the planned pauses or redundant words through the transcript, arrange scenes and add captions; compare modest cleanup with the original audio.
  3. Review timeline cuts and export a draft, then watch/listen to the exported file before accepting the shorter explainer.

Artifact to inspect

A proposed edited explainer with captions and an editable project; no recording or edit was executed here.

What to verify

Check that cuts preserve meaning, captions match the spoken words, consonants/breaths remain natural and the export has no gaps, clipping or unintended scene changes.

Data handling and permissions

Descript treats projects as confidential under its terms, subject to stated processing and access exceptions. Optional project-improvement use can be disabled through Share Data with Descript; separate AI-Speaker provisions allow de-identified training data and sample-quality review. This is not a blanket promise of no training or human access. Section 8.4 treats generated inputs and outputs as user content, with ownership only to the extent legally protectable; third-party rights, service conditions and input permissions remain. Partner generative tools can process inputs/outputs under applicable agreements. Review project-sharing and voice settings before sensitive uploads, and obtain the necessary permission before synthesizing or replacing someone’s speech.

Other options

ElevenLabs
Compare speech synthesis when voice/model control is the main deliverable.
Runway
Use shot generation when the missing material must be newly created rather than transcript-edited.

Sources and scope

Documentation-based; not hands-on tested. Features and policies are vendor statements checked against primary documentation on October 2, 2026. Fit, alternatives and the synthetic workflow are editorial proposals, not measured outcomes, quality rankings or legal clearance. Model, plan, regional and account entitlements may differ.

Source review: . Product access and terms can change.

Suggest a correction