--- name: song-style-ad description: "Turn an e-commerce product into a song-style ad — a micro music video where an original song carries the story over animated (Pixar-style) visuals, with one or two lip-sync beats of the animated lead singing on camera and the rest silent B-roll. Use when the user asks for a 'song ad', 'song-style ad', 'music video ad', 'musical ad', 'jingle', or wants their product sung. Runs one chat: (1) write 2-3 song scripts (lyrics + style prompt + visual concept) matched to the audience's genre and stage the music draft; (2) transcribe the generated song for pacing, design the animated lead, and storyboard beats onto it; (3) generate the lip-sync + B-roll drafts and assemble. Builds on copywriting, character-design, and image-prompting." --- # Song-style ad A **song-style ad** plays like a micro music video: an original **song** carries the product story over animated (usually Pixar-style) visuals. It wins when the song is a _real song people would actually hear out_ (not a jingle) whose hook lands the product's angle in the first five seconds. **The format's identity (always true):** the song is the spine — it sets the pacing, the mood, and the emotional arc, and the visuals ride on top of it. It is a **Mix**: one or two **lip-sync beats** where the animated lead sings on camera, and the rest **silent B-roll** cut to the music. The pacing is audio-first by construction — the generated song's real timestamps are the clock everything else is timed to. This skill builds on the parent (the one law, the Mix and audio-first speech doctrine, the storyboard and its binding rule, lip-sync via omni-reference, the draft tools, "No captions."), on **`copywriting`** (the angle and awareness craft the lyrics carry), on **`character-design`** (locking the animated lead as a hero character), and on **`image-prompting`** (composing each beat's still). It adds only what is specific to the song format — the genre map, the lyric + style-prompt craft, and the song-as-pacing-spine mechanic. ## The gate, filled by this format - **Fixed:** deliverable = an animated music-video ad · speech = Mix, sung (lip-sync beats + silent B-roll) · pacing = audio-first (the song). - **Ask (this format's open questions):** which product (Stage 1) · genre + duration — inferred from the audience, stated, validated (Stage 1) · the animated lead — one consistent hero character across beats, and what it is (Stage 2) · which lines are the lip-sync-on-camera moments (Stage 2). --- ## Stage map One chat, three stages — **detect the stage from what the user gives you this turn:** | They gave you… | Run | | -------------------------------------------------------------------------------- | --------------------------------- | | A product (brief / link / notes), no song yet | **Stage 1 — Song scripts** | | A generated song ("the song's done", a completed music draft, an uploaded track) | **Stage 2 — Storyboard the song** | | An approved lead design + storyboard | **Stage 3 — Generate the video** | Ambiguous → ask one short question. If they name a stage, obey it. --- ## Stage 1 — Write the song scripts and stage the music draft **Goal:** 2–3 distinct song scripts for the user to pick from; the pick becomes a staged music draft. 1. **Ground in the product** — use `findProducts` first unless current context supplies the exact product, then call `getProductBrief`. Continue only after a `resolved` result; otherwise ask the returned clarification question. Pull the brand voice, the target customer, the main pain, and the hero benefit. 2. **Settle the angle, awareness stage, genre, and duration — infer → state → validate.** Activate `copywriting` for the **angle** (the customer outcome the song is about — a specific transformation, _not_ a brand brag or a feature) and the **awareness stage** (how much the lyric has to teach). Infer the **genre** from the audience with the genre map below, and a **duration** (15 / 30 / 60s). State all four in a line and let the user adjust. 3. **Write 2–3 distinct song scripts** in the output format below. Each must feel genuinely different — vary the POV (product POV / customer POV / narrator), the mood (upbeat / emotional / playful / intimate), or the structural hook. Apply every Lyric rule. 4. **Stage the music draft** for the option the user picks with custom vocal mode so the lyrics remain exact: `draftMusicPlan({ customMode: true, model: "V5_5", instrumental: false, title: selectedOption.title, style: selectedOption.style, prompt: selectedOption.lyrics })`. Keep the concise musical description in `style`, put only the exact lyrics in `prompt`, and supply the selected song title. Simple mode may reinterpret its prompt and must not be used when the lyrics need to be preserved. The user clicks Generate; the finished song is the input to Stage 2. ### The genre map (infer the genre from the audience) | Audience | Genres that resonate | | --------------------------- | ------------------------------------------------- | | Moms (general) | pop, indie pop, light acoustic | | Moms (Gen X / 40+) | country, soft pop, throwback pop | | Gen Z women | pop, hyperpop, R&B, indie pop | | Gen Z men | hip-hop, trap, EDM, indie rock | | Millennial women | pop, indie pop, R&B, alt-pop | | Millennial men | indie rock, hip-hop, alt-rock | | Black men / women (general) | hip-hop, R&B, trap, neo-soul / R&B, neo-soul, pop | | Latino / Latina market | reggaeton, Latin pop, bachata, trap latino | | Fitness enthusiasts | hip-hop, EDM, hard rock, trap | | Skincare / beauty | pop, R&B, indie pop, hyperpop | | Luxury buyers | ambient, modern jazz, cinematic pop | | Pet owners | folk, indie pop, country | | Wellness / spiritual | ambient, indie folk, soft acoustic | | Outdoor adventurers | folk, indie folk, alt-rock, americana | | Gamers / tech | EDM, synthwave, hyperpop / lo-fi, electro-pop | | Foodies | indie pop, jazz-pop, acoustic | No exact match → use the closest fit and say why. ### Lyric + style-prompt rules (apply to every option) - **Hook in the first 5 seconds.** Whatever the duration, the catchiest line lands immediately. - **Lyrics carry the ANGLE, not the brand.** Sing the customer outcome; don't just chant the product name. - **A real song, not a jingle.** Real verses, a real chorus, a real rhyme scheme — never "buy buy buy." - **Short, singable lines** — 4–8 syllables for a pop hook. Long lines feel forced when sung. - **Match vocabulary + energy to the genre + audience.** Hip-hop ≠ country ≠ pop. - **Chorus cadence by duration** — 15s = hook + one chorus pass; 30s = verse-chorus-verse-chorus; 60s = full song with a bridge. - **Keep the style description tight (<25 words):** `[genre], [vocal — e.g. female lead / male rap / choir], [tempo / BPM], [mood], [production — modern polished / lo-fi / vintage]`. Music models do better with a concise style prompt. Never include an artist or band name — Suno rejects it; describe the sound instead. - **The visual concept must be producible** as animated beats — animated objects, a stylized character, simple Pixar-feel visuals. Not "live-action cinematic crane shots." - Each of the 2–3 options is **meaningfully different** — don't ship near-duplicates. ### Stage 1 output format (per option) ``` ### OPTION [N]: "[Hook line / song title]" **Concept:** [1–2 sentences — the song's vibe + the animated visual treatment idea.] **Lyrics:** [Verse 1] … [Chorus] … [Verse 2 — 30s+] … [Bridge — 60s only] **Style:** [genre], [vocal], [tempo], [mood], [production] ← under 25 words **Visual concept:** [2–3 sentences — what's on screen, tied to specific lyric moments; which lines are lip-sync-on-stage vs B-roll.] ``` After the options: _"Pick your favorite (or ask for more variations) and I'll stage the song for you to generate — then I'll storyboard the video to it."_ --- ## Stage 2 — Storyboard the song **Goal:** turn the generated song into a beat plan mapped onto its real timing. Runs once the song exists. 1. **Transcribe the song for pacing** — the generated song is the factual pacing source. Get its word-level timestamps and total duration; the storyboard maps beats onto real timing, not estimates. The song is generated whole, once — music models aren't deterministic, so never regenerate sections separately. 2. **Settle the elements — infer → state → validate:** - **The animated lead** — the singer/character. It's a **hero character**: use `character-design` to design it and lock its hero character reference (one shot, in the chosen animated style). Every lip-sync beat and any beat it appears in points back to that reference. - **The product** — its hero product reference (a clean, accurate shot) for every beat the product appears in. - **The recurring set** — e.g. a neon stage for the lip-sync beats; establish a hero reference only if it must stay consistent (usually just the stage). 3. **Map beats onto the song's timestamps** — one row per beat, each tied to a time range of the song. Two beat kinds: - **Lip-sync beat** — the lead sings a **slice of the song** on camera (the chorus, the hook, the payoff line). Carries the speech slice for that time range. - **B-roll beat** — silent animated visuals illustrating the lyric at that moment (the product, the ingredients, a pastoral setting), generated with no audio and overlaid on the song in the editor. Most of the ad is B-roll with one or two lip-sync beats (the hook and the final chorus are the usual lip-sync moments — see the worked example). Vary the framing across B-roll beats; apply `image-prompting`. 4. **Present the storyboard for approval**, then say generating the stills is next. --- ## Stage 3 — Generate the video **Goal:** the drafts, assembled over the song. 1. **Per-beat start frames** (`draftImagePlan`) — the hero lead reference first, then one start frame per beat: re-attach the lead / product / set references and re-state them, add the beat's action, setting, framing, lighting (`image-prompting`). No text in any still. Attach the product reference in any beat the product appears. 2. **Video drafts** (`draftVideoPlan`, omni-reference) — group beats into generations up to the ~15s cap: - **Lip-sync beats** — pass the beat's **start frame + its song slice** into a generation with the lip-sync capability and have the lead sing that slice. Because of the ~15s cap, cut the song into the slice for each lip-sync generation, cutting on natural musical beats (never mid-word). - **B-roll beats** — generate **silent** (no audio); the model only needs the pacing so the visual lands with its lyric. These are overlaid on the song in the editor. 3. **Assemble** the clips over the full song in the editor; add any captions there — never burned into the footage. --- ## Locked rules - **R1 — The song is the spine.** Everything is timed to the generated song's real timestamps; never storyboard against a guessed duration. The song is generated whole, once (`draftMusicPlan`); never regenerate sections separately. - **R2 — Lyrics carry the angle, not the brand.** The song sings the customer outcome; the product name is not the hook. - **R3 — The lead is a hero character.** One locked reference (via `character-design`), re-attached and re-stated in every beat it appears in; never chain off a previous beat's output. - **R4 — Lip-sync beats carry a song slice; B-roll beats are silent.** Only the on-stage singing beats get the audio (cut to ≤~15s on natural beats); B-roll is generated silent and laid over the song in the editor. - **R5 — Producible animated visuals.** Animated objects, a stylized lead, simple Pixar-feel visuals — no live-action cinematic crane shots. - **R6 — Footage stays clean of text.** Captions are an editor layer, only if the user asks — never in an image or video prompt. --- ## Worked example (the quality bar — not a template to copy) **HolyCow Whipped Tallow Balm** — animated Pixar musical, 9:16, ~42s. Song "Holy Cow (I've Arrived)" (animated pop, female lead, upbeat, 120 BPM). Angle = "something real, finally" (clean tallow vs. labels full of lies). - **Elements (settled first):** the lead character (one shot), the HolyCow jar (hero product), Pixar 3D style, a recurring neon stage. - **Beats mapped onto the song:** **[0:00–0:06] lip-sync** — the lead belts the opening line on the neon stage · **[0:06–0:14] B-roll** — hands bin the old products, open the jar · **[0:14–0:20] B-roll** — clean-vs-bad ingredient cards · **[0:20–0:28] B-roll** — a golden-hour pasture with a friendly cow · **[0:28–0:34] B-roll** — four scent jars rotate into a wheel · **[0:34–0:40] lip-sync** — back on stage, glowing, sings the chorus · **[0:40–0:42] B-roll** — the hero jar lands on a pedestal, CTA lockup. - **Why it works:** two lip-sync beats (hook + final chorus) bookend five silent B-roll beats; every B-roll beat visualizes the lyric at that exact moment; one locked lead and one locked jar across all seven beats. --- ## Tools the parent calls - `findProducts` / `getProductBrief` — resolve and pull the real product before writing (Stage 1); never invent proof. - `extractMediaFromUrl` — when the user references a song-ad they like as inspiration. - `draftMusicPlan` — the song itself (Stage 1): custom vocal mode with a title, the musical description in `style`, and exact lyrics in `prompt`. Transcription for pacing runs on the generated song in Stage 2. - `draftImagePlan` — the hero lead reference and each beat's start frame (Stage 3); the parent sets the default image model. - `draftVideoPlan` — each omni-reference video generation (Stage 3): lip-sync beats pass the song slice; B-roll beats silent. Stage the drafts; the user clicks **Generate**. Don't run past the song generation unprompted — wait for the finished song before storyboarding. _This is a Starpop skill, built to run inside Starpop.ai. Some of its steps assume system prompts and tools that only the Starpop chat has, so for the best experience, use it in Starpop.ai._