短剧镜头视频提示词

assets/music-spec.jsonl.md

Music is a timeline layer, not a default bed baked into every generated shot. Create a record only when accepted story or directing material gives music a specific dramatic job.

The file opens with a sources header that declares each upstream snapshot once. A source key is short, lowercase and derived from the artifact basename; it stays stable and unique within the file.

{"record_type":"sources","schema_version":"1.0.0","sources":{"screenplay":{"owner":"short-drama-write","artifact":"剧集/EP001/screenplay.md"}}}

Every record after the header names the snapshot through its key:

{"music_id":"MUS-EP001-01","scope":{"episode_id":"EP001","start_seconds":0,"end_seconds":12},"source_refs":[{"src":"screenplay","record_id":"EP001-SC001"}],"narrative_function":"<what changes for the audience, and why picture/performance alone is insufficient>","prompt":"<provider-neutral style, mood, instrumentation, energy and scene description>","mode":"instrumental | song","lyrics":null,"mix_intent":{"entry":"<relative entry against picture or dialogue>","exit":"<relative exit or transition>","duck_under_dialogue":true,"loop":false},"status":"candidate"}

• source_refs carry src plus the record or field they point at; the artifact's owner and artifact live in the header entry that src names.
• scope is the accepted timeline interval; it does not promise a supplier can generate that exact duration.
• prompt contains musical intent only, never model names, API fields, artist imitation, remote task identifiers or credentials.
• song requires creator-owned or licensed lyrics; instrumental carries no lyrics. Do not ask a provider to silently rewrite accepted lyrics.
• mix_intent is an editorial handoff. Production generates a source track; editing still owns exact fades, ducking and placement.