AI Song & Music Video Studio · öffentlicher Open Skill
AI Song & Music Video Studio
An idea in. A finished music video out — with an original song. Most music-video tools need a track you already have. This one writes the song too: lyrics, vocals, the beat, and a 90-second multi-shot video cut to that beat. WHAT IT DOES • Writes real song structure — intro, verse, pre-chorus, chorus, bridge, final chorus, outro — with a hook that repeats verbatim • Generates the full track (genre, mood, bpm, key, instrumentation, vocal style) • Designs an artist persona and REMEMBERS it, so your next video opens in the same world • Builds a 12–16 shot list mapped to song sections, then generates one still per shot • Shoots performance shots with lip-sync and consistent identity (i2i against the persona asset — the face does not drift) • Adds b-roll so the video breathes instead of filling the timeline with faces • Cuts on the beat using transcription timestamps, faster in the chorus, slower in the verse • Delivers one stitched MV, portrait 9:16 or landscape 16:9 WHAT MAKES IT GOOD • The chorus is treated as the product — you approve the hook before the expensive rendering starts • Chorus shots repeat a grammar and change the angle, so a chorus feels like a chorus • One movement idea per shot — no shot pushes in and pans and zooms at once • The artist persists across runs, which is why people come back for song two INPUTS IT HANDLES • A theme, a feeling, one sentence, or a few loose lines • Lyrics you already wrote • Your own uploaded audio (it transcribes and cuts to your track instead) • A photo of a performer (kept consistent across every shot) GOOD FOR J-pop, city pop, ballad, hip-hop, EDM, anime-style MV, lyric video, an AI artist's first single, a song for a friend, a vertical short for TikTok / Shorts / Reels. NOT FOR Talking-head explainers with no music, podcast editing, cloning a real artist's voice or likeness, or adult content. Original work only. Give it one line — "a city-pop song about leaving Tokyo at 5am" — and it will write the song, shoot the video, and hand you the MV.
Music
8 1 5
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4267104/2026-10-19/4g2uk8bf.png
--- allowed-tools: - generate_image - generate_video - generate_music - generate_tts - transcribe_audio - recognize_image - recognize_video - split_image - mix_clips - create_document - update_document - memory_whoami - memory_query - memory_store --- # AI Song & Music Video Studio One idea in, one finished music video out — **with an original song**. This skill writes the lyrics, generates the track, locks a performer, shoots 12–16 shots cut to the music, and stitches the deliverable. It also remembers the artist, so the next video in the same world starts half-done. ## When to use - The user gives a **theme, mood, or a few lines** and wants a song about it. - The user wants a **music video** for a track they already have. - The user wants **lyrics turned into a video**, or a **lyric video** with subtitles. - The user wants an **AI artist / persona** that appears across several videos. - The user uploads a **photo of a performer** and wants them to sing. - The user asks "what should I make with this audio?" — route here. Do **not** use this for: a talking-head explainer with no music (that is a plain avatar task), podcast editing, or adult/NSFW content — see Notes. ## Steps ### 0. Intake — decide the one thing that matters Call `write_todos` with the plan: intake → artist bible → lyrics → song → cover → shot list → stills → performance shots → b-roll → mix → deliver. Ask at most **three** questions, and only the ones you cannot infer: 1. **What is the song about?** — a feeling, a person, a place, a story. One sentence is enough. 2. **What genre and language?** — e.g. J-pop / city pop / ballad / hip-hop / EDM; Japanese, English, or bilingual. 3. **Does a performer already exist?** — a photo, a previous video, or should we design one. **If the user gives you almost nothing, do not stall.** Make the three decisions yourself, state them in one line, and proceed. This skill is judged on delivering a finished video, not on asking a good interview. ### 1. Artist bible — memory is the retention feature Continuity is what makes people come back. Before generating anything: ``` memory_query(query='artist persona, voice, visual style, previous songs for this user') ``` - If a persona exists: reuse it exactly — same face, hair, palette, wardrobe language. - If not: design one, then persist it. ``` memory_store(key='artist_persona', value='<name>; <age/typing>; <hair/face/outfit>; <palette>; <genre>; <signature visual motif>; <voice description>') ``` Also store the song so sequels can reference it: ``` memory_store(key='song_<slug>', value='<title>; <theme>; <genre>; <bpm>; <key>; <hook line>') ``` Show the persona back to the user in one line: **"This is <name> — <one descriptor>."** They will recognise her in the next video, and that is the point. ### 2. Lyrics — structure before poetry Write lyrics as a real song structure, not a block of verse. Keep it singable: ``` [Intro] 2 bars [Verse 1] 8 bars [Pre-Chorus] 4 bars [Chorus] 8 bars [Verse 2] 8 bars [Chorus] 8 bars [Bridge] 4 bars [Final Chorus] 8 bars [Outro] 4 bars ``` Rules that make generated lyrics usable: - **One image per line**, not two. Line length 6–12 words (or 10–24 Japanese characters). - **The hook must repeat verbatim** in every chorus. A chorus that changes is not a chorus. - Write the **title into the hook line** so the video's title is already in the song. - Non-English lyrics: keep them plain and singable; avoid dense kanji compounds a singer would trip on. - Save the script: `create_document(title='<song title> — lyrics', content=...)`. Read the chorus back to the user before generating audio. **The chorus is the whole product** — if they do not like it, everything downstream is wasted. ### 3. The song ``` generate_music( model='music', prompt='<genre>, <mood>, <bpm> bpm, <key>, 90 seconds, full song structure with intro, verse, pre-chorus, chorus, verse 2, chorus, bridge, final chorus, outro; <instru