List existing Reelwright projects.

Create a new project. width/height/fps define the canvas; fit controls how sources with a different aspect ratio are placed (contain = letterbox, cover = crop).

The full edit: clips with timeline and source times, overlays, music, open notes, assets.

Change canvas settings (resolution, fps, fit, background colour, font file for text).

Undo the last edit.

Redo the last undone edit.

Undo/redo stack descriptions (oldest first).

Add media files (video, audio, images) to the project's asset library. Paths must be absolute paths on the machine running this server.

Contact sheet: count evenly spaced frames from [start, end] of a source asset, each labelled with its source timestamp. Use it to see what is in a clip.

Shot-change timestamps (source time) in a video asset. Lower threshold = more sensitive.

Silent ranges [start, end] (source time) in an asset's audio.

Integrated loudness (LUFS), true peak, and a per-window loudness curve for an asset or a rendered file. YouTube normalises to about -14 LUFS.

Waveform image of an asset's or a rendered file's audio, to spot speech, pauses and peaks.

Transcribe an asset with word timestamps (needs pip install reelwright[transcribe]). Results are cached; read them with get_transcript.

Attach an existing .srt or .vtt file (in source time) as an asset's transcript.

Read a transcript as timestamped lines. With asset_id: source time. Without it: the edited timeline (only words that survive the edit). query filters lines containing text.

Place [src_in, src_out] of an asset on the main track (appends unless index is given). Images default to 5 seconds.

Change a clip's in/out points (source time). Later clips ripple.

Split the clip under a timeline time into two clips.

Remove a clip; everything after it ripples left.

Ripple-delete [start, end] of the timeline (e.g. a pause, a fumbled sentence). Works across clip boundaries; overlays and notes after it shift left.

Move a clip to a new position on the main track (0 = first).

Set a clip's audio gain (1.0 = unchanged, 0 = mute).

Cut dead air across the whole timeline: detects silence in every clip's audio and ripple-deletes it, keeping padding seconds either side so speech isn't clipped. Use dry_run=True to preview the cuts first.

Add a text overlay at timeline times. font_size is relative to a 1080p canvas. Use \n for line breaks.

Change any property of a text overlay.

Delete a text overlay.

Lay background audio under the main track from timeline time start. duck=True lowers it automatically whenever the main track has sound (speech).

Delete a music cue.

Turn captions on/off. Captions are generated from asset transcripts and follow the edit automatically. burn_in=False exports a sidecar .srt instead of drawing them.

Pin a review note to a timeline time (e.g. a question for the human).

Review notes (human feedback such as "cut this pause"), in timeline order.

Mark a review note as handled.

Render frames of the EDITED timeline at the given timeline times (overlays and captions included) and return them as one labelled grid. Use after every edit to verify it.

Evenly spaced frames across the edited timeline (or a range of it) in one image.

Fast low-res MP4 of the edit (or of [start, end]) for review. Returns the file path.

Final render. Presets set resolution and loudness (-14 LUFS for youtube/shorts/square). Use a GPU encoder (h264_nvenc etc.) if the machine has one. Writes a sidecar .srt when captions are enabled.