Assemble a song-driven video ad from a config, cutting supplied scene clips to the song’s lyric timing and adding synchronized captions and a brand end card. The song serves as the narration, with no separate voiceover. Song, keyframes, and clips must be supplied by upstream generation steps.
Assemble a song-driven music-video ad from a config — a generated sung track carries the whole narration across N tableaux (one keyframe -> one i2v clip per lyric beat) with NO separate voiceover, captions synced to the song's OWN word timings (script-window, never Whisper) and the hook word landing on the chorus drop, closed on a PIL brand end card. This is the FREE deterministic assembly stage (clip cut-to-timeline + captions + end card + FFmpeg composite); the song, keyframes, and clips come from create-music-elevenlabs / create-image-fal / create-video-fal. Use for the song-driven-music-video format.
Run score Not tested
See how to installReview the instructions the agent will receive before you install.
Assemble a song-driven music-video ad from a config: a purpose-written, sung song is
the entire narration (no separate voiceover), and every visual beat is timed to the lyrics.
The delivered song sets the timeline; N tableaux (one keyframe → one image-to-video clip per
lyric beat, all in a single look pack) are cut to their lyric windows and hard-concatenated
on the beat, captions are built from the song's OWN word timings with the hook line landing
on the chorus drop, and the spot closes on a PIL brand end card. It reads like a tiny animated
music video, not a demo. scripts/config.example.json is one worked example (Fernhollow "Bring The
Green Back Home", 28s paper-craft 9:16) — its style, song, vocalist, mood, protagonist and
setting are that demo's answers, not defaults; scripts/PIPELINE.md maps every config block
to its step and scripts/README.md documents the free assembly.
The creative calls are made upstream by the user (the recipe's choices) and arrive in the config;
this assembly never picks them.
look_pack, clip_engine.motion_opener. The demo used a paper-craft diorama
on a rainy morning.song.prompt / song.bpm. The demo used breezy bedroom-pop at ~80 BPM.song.prompt. The demo used a warm, airy male lead.song.prompt, the lyrics, tableaux. The demo went grey and stuck → fresh and hopeful.look_pack.style_opener. The demo used a young-man paper-doll.tableaux[].keyframe_prompt. The demo went grey apartment → paper indoor jungle.This is the FREE, deterministic assembly stage — it spends nothing. The three paid
inputs are separate capabilities: the sung song (create-music-elevenlabs, music_v1,
force_instrumental FALSE — the lyrics ARE the script, returns mp3 + words.json), one
keyframe per tableau (create-image-fal), and one Kling 3.0 i2v clip per tableau
(create-video-fal). Given the delivered song + words.json + one clip per beat,
render-song-mv cuts each clip to its lyric window, hard-concats on the beat, builds the
lyric-synced captions, composites the PIL end card, and muxes → the master. Re-cuts reuse
the existing song / keyframes / clips and spend nothing.
force_instrumental false); do not add a spoken voiceover or a
second music bed.timeline.json) — never trim the song to a pre-planned grid.audio/words.json (~3 words at lyric boundaries); accent words get captions.accent_color.
Whisper on sung audio returns "🎵 Music Playing 🎵", so it can't caption lyrics.is_hook) is timed so the
payoff word (song.hook_word) sits on the chorus drop; accent that word in the captions.style_opener + negative_tail + palette drives
every keyframe so N beats read as one film; no morph within a clip.audio_mix.climax_beat_id — set it to
this run's hook tableau; the demo's T08 is example-only), loudnorm to −14 LUFS → 1080×1920
h264+aac. No paid calls, no keys.No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
The supplied material contains only placeholder text for the skill’s purpose, inputs, steps, and completion criteria. It does not provide enough information to describe a task or result reliably.
Research supported advertising angles using buyer context, current product facts, customer language, and observed ads. Turn the evidence into static ad, video, or copy ideas, matching video concepts to available recipes. Scores rank ideas for testing; they do not predict results.
Analyze campaign data from Google, Meta, and LinkedIn to find strong and wasteful spend. The skill checks funnel performance and test results, then recommends which campaigns to pause, scale, or test and how to shift budget across channels.
Compare ad spend and signups with downstream outcomes to calculate the cost of each qualified lead. The skill checks tracking coverage, accounts for how long leads need to mature, and groups creatives into actionable categories with confidence notes.
Prepare the intake for a Meta ad campaign by reviewing available brand information first, stating reasonable assumptions, and asking only questions whose answers could affect budget, claims, or legal exposure. It records stable brand facts, a campaign brief, and what was asked versus assumed.
Review ad copy, image text, visuals, and people against platform and brand category rules. It suggests honest fixes and needed disclosures, then flags issues requiring human or legal review, with a separate result for each ad and platform.