Descript

Operator's take

Most video editing software was built for people who think in timelines — producers who can eyeball a waveform and know exactly which 200-millisecond clip to trim. If that's not you, and you're spending your Saturdays manually hunting "um"s in a two-hour interview, you're using the wrong tool for how you actually think. Descript's core bet is that anyone fluent in Google Docs can edit video: upload your recording, wait for the transcript, highlight the parts you don't want, delete them. The media follows.

That core metaphor covers more ground than it first seems. Filler word removal runs automatically on the transcript — you don't pick through a waveform, you skim highlighted text. Studio Sound cleans up audio recorded in a kitchen or a car in one click. If you flubbed a line, Regenerate can fix it in your trained AI voice without re-recording the whole segment. Underlord, Descript's AI co-editor, can write scripts, pick viral-worthy clips, design layouts, and generate B-roll video from a prompt — so the editing work that used to take a Saturday can be directed more than executed. For operators publishing a podcast, a weekly Loom-style update, or social video clips from a webinar, the weekly time savings add up fast.

The tool is genuinely wrong for two groups: video professionals doing color work or multi-camera edits, and anyone working with content that isn't dialogue-driven. Descript's model breaks down when you're cutting to music, doing motion graphics, or working with footage where the transcript isn't the source of truth. It ships as both a web editor and a native desktop app, but the AI layer and project storage stay cloud-dependent — so large project files can lag, you're tied to Descript's uptime for the AI features, and transcription accuracy drops with heavy accents or low-quality source audio. If any of those describe your typical workflow, that's a real friction cost to budget for.

What it's good at

• Text-based editing — cut, rearrange, and delete media by editing the transcript; what you remove from the text disappears from the timeline automatically.
• Automatic filler word removal — highlights every "um," "uh," and awkward pause in the transcript so you can clear them in bulk without scrubbing audio.
• Studio Sound enhancement — one-click audio cleanup that reduces background noise and balances levels without requiring audio engineering knowledge.
• AI Speech and Regenerate — text-to-speech from a custom voice clone, plus per-word audio repair that re-types a flubbed word and re-syncs mouth movement to match; available on Hobbyist and up.
• AI clip selection and export — Underlord picks the highest-engagement moments from a longer recording and exports captioned clips for multiple platforms without touching a separate editor.
• Multi-user collaboration — shared projects with commenting, version history, and real-time editing for content teams.
• API and MCP surface — a published API plus an MCP server let you drive Descript's editing programmatically or from inside an AI assistant, so the editor isn't a closed box for automation-first teams.

What it's not

• Not a professional video editor — limited color correction is available but no serious color grading, motion graphics, or complex multi-camera timelines; wrong choice if your output is cinematic rather than conversational.
• Not built for music or non-dialogue content — the transcript-first model requires speech to anchor edits; it doesn't help when the audio is score, sound design, or B-roll.
• Not reliable under bad audio conditions — transcription accuracy degrades with heavy accents, cross-talk, or poor recording quality, which undermines the whole editing model.
• Not fully offline — available as a desktop app but cloud-dependent for AI features and storage; large projects can still lag, and you're reliant on Descript's uptime for the AI layer.