Music & Videoke Creator · Open Skill 公開技能
Music & Videoke Creator
Create complete original music experiences—from song concepts, lyrics, melodies, arrangements, and vocal direction to music videos and karaoke/videoke presentations. Generate original songs across multiple genres and languages, develop instrumental backing tracks, structure songs for easy singing, synchronize original lyrics to musical timing, and create visually engaging karaoke videos with readable on-screen lyrics and professional timing cues. The skill is designed for original entertainment, children's songs, educational music, storytelling, celebrations, social-media content, and personal creative projects.
Music
15 2 4
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4179755/2026-10-02/f7by4gu2.jpg
# Music & Videoke Creator A complete creative music-production assistant. It takes a simple idea — "Make me a funny song about being single" — and develops it into: Concept → Lyrics → Song Structure → Musical Direction → Vocal Direction → Instrumental Arrangement → Videoke Version → Visual Concept. ## When to use - The user wants an **original song** (any genre/language), lyrics, or a song concept. - The user wants an **instrumental / karaoke / videoke version** of a song. - The user wants **timed karaoke lyrics** or a **karaoke/videoke video** with readable on-screen lyrics. - The user wants a **music video** or **visual concept** for a song. - The user wants **children's songs**, **educational songs**, **celebration/event songs**, **call-and-response songs**, or **social-media music**. - The user wants to **revise or redo any single stage** (lyrics, song, instrumental, timed lyrics, karaoke video, music video). ## Core principle When the user requests **only one component** (e.g. just lyrics, just an instrumental), produce only that component — do not generate the entire package unless asked. ## Steps 1. **Clarify the request.** Identify the output mode (see below) and gather what the user specified. If they give only a theme, make reasonable creative choices and state them. If they want a full package, use `write_todos` to list the stages and update it as you progress. 2. **Create the song content** (title, theme, story, genre, mood, tempo, key, structure, original lyrics, hook, vocal direction). Follow `references/song-creation.md`. 3. **Generate the music** with `generate_music` (`model='music'`). Pass the full musical direction in the prompt; pass the original lyrics in the `lyrics` field when the song should be sung. Use `name` = song title, `entity_type='music'`. 4. **Generate the vocal/lead** (optional) with `generate_tts` (`model='tts.design'`, `voice_instruction` in English describing the vocal character — never clone a real artist). Use `name` = "Lead vocal — <song>", `entity_type='voiceovers'`. 5. **Instrumental / karaoke version**: generate a backing track with `generate_music` (`model='music'`) designed for singing (no lead vocal, clear space for the singer). Use `name` = "<song> — instrumental", `entity_type='music'`. 6. **Time the lyrics**: if the user provides audio, run `transcribe_audio` (`model='asr'`) on it to get timestamps; otherwise build an **estimated** timing structure and clearly mark it as estimated (never pretend exact sync exists). Follow `references/videoke.md`. 7. **Create visuals** (backgrounds, characters, scenes) with `generate_image` (`model='t2i'` for new scenes, `model='i2i'` to edit/keep a character consistent). Follow `references/music-video.md` for music-video mode and `references/videoke.md` for karaoke backgrounds. 8. **Generate video** (optional) with `generate_video` — use `model='i2v'` (single image → video) or `model='ref2v'` (reference images, keeps character/subject consistent; pass `audio_url` to drive by the song). Video does not support NSFW/adult content. 9. **Assemble the videoke/music video** with `mix_clips` (`clip-mixer`): concatenate clips and burn in timed lyrics via the `subtitles` field (use `font_name='Noto Sans CJK SC'` for Chinese/Japanese/Korean text). Use `name` = "<song> — videoke" / "<song> — music video", `entity_type='final_videos'`. 10. **Quality-check** against `references/quality-control.md` before finalizing. Summarize the result in natural language (don't paste raw URLs). ## Output modes Recognize which mode the user wants and deliver that output: - **MODE A — SONG**: title, concept, genre, mood, tempo, instrumentation, structure, original lyrics. - **MODE B — MUSIC PRODUCTION**: genre, BPM, key, instrumentation, arrangement, vocal direction, production direction. - **MODE C — VIDEOKE**: song info, instrumental direction, timed lyrics, lyric display instructions, background visual concept, transitions, ending. - **MODE D — MUSIC VIDEO**: song con