Apply direct edits to scenes by id

Apply exact edits without a Clypmint model round. Read get_project_context first. Canvas ops: output_frame {frame: {ratio: 9:16, 16:9, 1:1 or 4:5, fit}} reshapes this video ("make it vertical"); fit subject (the default) follows the subject, contain keeps the whole picture with bars (only when they ask to fit or keep everything in), cover fills; optional x/y (0-1, with fit cover only) move the centre of the filled frame; add: true makes the shape an extra output and keeps this one ("a square version too"); frame: null alone (no second op) is the original shape again ("undo the crop"); layout moves/scales/rotates a returned element ID using its units; element changes supported text/color/hidden; put_overlay adds or updates timed text/images with none/fade/slide/pop animation; remove_overlay removes an added element. Use only returned editable capabilities and existing canvas image assetIds. These are the manual editor operations and preserve authored animation. Music: {op:"music",enabled:false} removes only added music; enabled:true with track (one of Clypmint's own tracks, never claimed when posted: bright-house (upbeat), lofi-desk (calm), tech-forward (tech), documentary-pulse (story), gym-phonk (energetic), acoustic-morning (warm), soft-piano (emotional), funk-pop (fun), night-suspense (mystery), rise-up (inspiring)) or with music_staging_key from import_asset (a song of their own). Optional level_db (-36 to 0 relative to source audio) and beat_sync:true for source-shot action montages; other editing clocks refuse beat sync explicitly. Scene ops: remove, remove_layer, replace_screen, set_text, swap_picture (an existing picture key), move, add, cut_words, emphasis, captions, sound, scene_sounds. remove and cut_words take footage time out of the video; to delete a graphic and keep the footage and captions under it use remove_layer {layer: graphic}, to delete a zoom or camera move remove_layer {layer: camera}, and to shorten or lengthen a graphic move its scene to fewer or more words. scene_sounds replaces the scene cue list (empty removes added cues): choose names from scenes.soundOptions, optional on from soundWords.index and gain -36 to -10 dB. It preserves speech; global sound mute still applies. Batch up to 40 ops: all apply or all refuse. Native canvas changes save a playable project; export separately for a video file. Older timeline operations may require rendering; unsupported native operations refuse without rebuilding the project. Set base_version and output_id from context. Hook openings use revise_video. Free, up to 30 a day. Preserve operation_id and exact arguments on retries. Cost: Free: no credits and no change round. Same as the MCP tool apply_edits.