Composite a generated creator track into a product layer, per beat — in the creator's zone on split beats (either side of a seam), full frame on creator beats, hidden on product beats — with the creator's voice as the audio (loudness -14 LUFS) and an optional ducked music bed, or a music-only or silent master for voiceless formats. Free, local ffmpeg. Pairs with footage-cutlist (the layer) and create-creator-takes-h3 (the creator). Use for split-screen (with an optional zoom and brand-colour seam bar), screen-insert and any creator-over-product format.
Install gooseworks: https://gooseworks.ai/creative-agent/install.md # then, in Claude Code, Cursor, or Codex: /gooseworks use the compose-creator-layer skill
python compose.py --layer layer.mp4 --beats cutlist.aligned.json --creator creator.mp4 \
--out reel.mp4 [--music bed.mp3 --music-db -20] [--head 0.30] \
[--zoom 1.15] [--divider '#RRGGBB' --divider-px 4]--layer: footage-cutlist cut.py output rendered from the same cut list as
--beats (re-render it after align_beats.py). --draft layers are refused.--creator: create-creator-takes-h3 join_takes.py output, reel time from 0.--audio creator|music|none: the voice (+ ducked --music), the --music bed alone
(voiceless paid ads), or a silent master (organic posts: the track is added in-platform
at upload, which is licensed for organic use only).creator beats.--head: where the crop sits vertically when the creator is scaled to cover an area
(0 keeps the top, 0.5 centres). 0.30 keeps the head in the upper third.The split-screen creator look (product on top, the creator talking below) is this atom with split beats; it replaces render-split-screen-creator.
seam
(even, about 52% of the height for product on top) and creator_side: "bottom", fit each 16:9
product clip by width, and set bg: "blur" so the margins are a darkened blur of the same
clip, never black bars. Window each clip to the moment that proves its line; never loop a
short clip.--zoom 1.15 to 1.2 with --head pushes a webcam-distance take in to
head and shoulders. Tune it by eye against the reference; a re-run is free. Zoom cannot rescue
a take shot too close, so the creator still should be a candid medium shot (chest up, real
room), not a plain-background headshot.--divider '#RRGGBB' draws a brand-colour bar centred on the seam (default 4 px,
even) on split beats only.--music for this look.caption-burn, transcribed from this output).--layer: footage-cutlist cut.py output rendered from the same cut list as--creator: create-creator-takes-h3 join_takes.py output, reel time from 0.--audio creator|music|none: the voice (+ ducked --music), the --music bed alone--head: where the crop sits vertically when the creator is scaled to cover an areaResearch supported advertising promises from a specific buyer's context, current product facts, customer language and observed ads. Turn them into static, video or copy ideas; match video ideas to actual recipes and save a product-scoped angle bank for write-video-ad-script. Product-led discovery can use verified facts without reviews. Reference persistence and reach are observations, not proof of performance.
Analyze ad campaign performance data (Google, Meta, LinkedIn) to identify what's working, what's wasting budget, and specific cut/scale/test recommendations. Runs statistical analysis, funnel diagnostics, and multi-channel budget reallocation with specific dollar-amount shift recommendations and scenario modeling.
For paid lead-gen and participant-recruitment ads, replaces vanity CPA with true CAC per qualified lead by joining ad-platform data with downstream funnel events, surfaces tracking gaps, and classifies every creative into Scale / Keep / Investigate / Cut.