Forward-Future/scarlet

Overview

Scarlet is an open-source desktop video editor for macOS and Windows aimed at dialogue-heavy footage such as interviews, podcasts and talking-head takes. Instead of trimming clips by hand, you describe a change in writing or by voice to Claude Code or Codex, and the requested edits land on a working timeline; the original footage stays intact. It is suited to editors who want agent-assisted first cuts and pause or repetition cleanup while keeping review control.

Try the sample and edit your own footage

Scarlet is early in development. Install Git, Node.js 22.x or 24.x, pnpm 12.16.1, Rust and the platform build tools from the Tauri prerequisites. The minimal build is:

git clone https://github.com/Forward-Future/scarlet.git
cd scarlet
pnpm install --frozen-lockfile
pnpm build

To explore the native editor with a simulated assistant and an in-memory project, run pnpm desktop:demo. To work on your own footage, install CMake, then fetch FFmpeg, build local transcription support and start the app:

pnpm desktop:ffmpeg
node apps/desktop/scripts/build-whisper.mjs
pnpm desktop:dev

Prebuilt early-access downloads are also listed for macOS and Windows on the early access releases page.

First edit workflow

1. Create a project, name it and import video, separate audio or stills. Connect Claude Code or Codex with your existing subscription under Settings → General.
2. Select a timeline point, then type or hold V to speak. A whole-interview pass can use /dialogue-pass, which removes false starts, repeated takes and pause cuts; other workflows appear when you type /.
3. Review each proposed cut against the transcript, approve, skip or ask for a revision, and use \ to hear removed footage and the join. Earlier versions remain in project history.
4. Export the whole video or a selected range as MP4, or export the timeline as OpenTimelineIO (.otio) for DaVinci Resolve. Available workflows are documented in the skills guide.

Requirements and data handling

Projects and rendering run locally, with CrisperWhisper as the default local transcription model; until its separately licensed weights are installed, Scarlet downloads Whisper large-v3-turbo on first use. See the transcription model guide for model choices. Your connected AI provider receives the editing request and relevant project context, which may include transcript excerpts and video frames. Packaged releases send anonymous usage, review-decision and task/export events to PostHog; this can be turned off in Settings → About or with SCARLET_TELEMETRY_ENABLED=false, and the telemetry documentation lists the events.

HotRepo's take

Scarlet entered the HotRepo new-project list for November 14, 2026 with 102 stars and 19 forks, all recorded within a growth window of about seven hours on its launch day. There is no daily all-language rank and no multi-day star series yet, so the record supports only that early pickup, not a trend. The technical focus is also clear from the material: dialogue cleanup through agent-driven cuts, local transcription and render, and an OpenTimelineIO handoff for editors who finish in Resolve.