Skip to content
Back to Work
Live Productfrontendbackend

Crisp

A native macOS video editor that lets you edit video by editing the words. Crisp transcribes your footage on-device, links every word to its exact moment of video, and lets you cut by editing text — delete an 'um' in the transcript and that slice of video is gone. One-click Auto-clean flags filler words, repeated takes, and dead air. Built for talking-head creators who just want to cut, clean, and ship.

SwiftSwiftUIAVFoundationWhisperKitCoreMLXcodeGen

Problem

Talking-head creators waste hours scrubbing through raw footage to cut filler words, dead air, and repeated takes, because traditional timeline editors force you to edit visuals instead of content.

Approach

A transcript-first editing model: WhisperKit runs on-device via the Apple Neural Engine to produce word-level timestamps, then AVFoundation's AVMutableComposition rebuilds the video output non-destructively from a set of kept text ranges.

Key Engineering Decisions

1

Transcript as the source of truth

Every word links to its exact video moment, so cuts happen in text and the composition is derived from the kept ranges rather than manual timeline edits.

2

Fully on-device transcription

Whisper runs locally via CoreML/ANE so footage never leaves the Mac, making privacy the default and keeping the download free.

3

Pure-Swift detectors

Fillers, near-duplicate takes, and pauses are detected with dictionary and string-similarity heuristics plus timestamp gaps — no extra ML dependencies needed.

Results

100%

On-device privacy

0

Cloud dependency

Technologies

SwiftSwiftUIAVFoundationWhisperKitCoreMLXcodeGen