360 Secure SkillHub
360 安全技能中心

music-generation-studio

该 Skill 将主题、歌词或参考音频整理为制作卡,选择 Suno 等音乐模型,经 Beatra MCP 提交一次计费生成并轮询任务,返回音频片段与任务/计费信息,再根据试听做聚焦迭代。

分类:创作
beatra-ai
beatra-ai
热度 13.85
最近更新 2026-10-21 12:08:38

安装

发给 Agent
帮我安装这个技能:npx secure-skills-cli add beatra-ai/music-generation-studio
兼容性
通用兼容
联网要求
需要联网
数据出网
必须出网
Metadata
name
music-generation-studio
description
Create original songs, AI-generated music, lyrics-to-song tracks, instrumentals, background music, video soundtracks, jingles, multilingual songs, and reference-led arrangements from a clear creative brief. This AI music generator and AI song maker develops genre, mood, structure, singable lyrics, vocal direction, and production style, then creates reviewable audio with Suno 5.5 or another model you explicitly choose. Use it as an AI music studio for songwriting, an AI lyrics writer, text-to-music creation, BGM generation, brand music, podcast themes, game music, bilingual songs, or focused new versions of reference audio. Review vocals, pronunciation, duration, loop points, and arrangement fit after generation, then refine the strongest result.

Music Generation Studio

Develop a theme, scene, lyric fragment, complete lyric, functional music brief, or reference recording into one coherent musical direction and, when the user wants audio, one intentional generation request. The deliverable may be a song, instrumental, BGM, soundtrack, jingle, multilingual track, or reference-led new arrangement. Treat singer identity, pronunciation, exact duration, loop points, melody carryover, and mastering as qualities to review after generation rather than guaranteed properties.

Lock genre, structure, and vocal or instrumental route before generation. A “surprise me” random song is not this package's default. Route a cover or re-arrangement of an existing recording to ai-song-cover-studio, a gift or occasion song to personalized-song-maker, and a short loopable instrumental for picture to short-video-bgm-studio.

The hard inputs are a theme or lyrics plus the intended use. If purpose, genre, or vocal route is empty, gather them first; do not fill the gap with a default pop brief.

Use the bundled scripts/mcp_client.py for every Beatra operation. Do not configure or use a host Beatra Connector. Never use REST/OpenAPI as a fallback. For ordinary calls, run python3 scripts/mcp_client.py call <tool-name> and pass one JSON object on stdin. For local files, use only the dedicated upload command described below. See Bundled MCP Client diagnostics only when the bundled path needs diagnosis.

Shape the direction first

Reuse every settled choice in the conversation. Infer ordinary creative details from the destination when that is safe. Ask only when a missing answer changes the lyrics, reference direction, vocal or instrumental route, exact paid payload, or another high-impact user choice.

Before generation, prepare one compact production card:

  • destination and listener;
  • title, premise, emotional arc, primary genre, tempo feel, instrumentation, structure, mix direction, ending or loop direction, and exclusions;
  • vocal or instrumental route;
  • full lyrics, language and register, and vocal direction when the track has vocals;
  • selected model and model-specific controls;
  • reference intent and the qualities that should carry over or change when a reference is involved; and
  • the fact that the next action is one billable generation.

For instrumentals, omit lyrics and make room for dialogue or other functional needs in the creative direction. For vocal music, finish and show material lyric changes before generation. When lyrics exist, organize verse, chorus, and bridge labels before paying. Do not put BPM, key, or tempo numbers into the style caption. Lyrics writing, creative planning, model discovery, prompt preparation, and critique do not require a paid music call. See intent and routing, creative brief and style, lyrics craft, and vocal, language, and tags.

When a brief uses artist shorthand, translate it into actionable genre, era, tempo, instrumentation, harmony, vocal texture, phrasing, and mix attributes.

Choose and validate the model

Set model: "suno-5.5" for this package's ordinary generation. Never omit the model and never silently use auto. Call beatra.models.list with capability text_to_music or reference_audio_to_music when the user requests another model, asks about current availability or price, or needs a model-specific capability. Do not silently substitute a model.

Use only controls returned or documented for the selected model family. Suno and MiniMax options are not interchangeable. Validate the exact payload using model routing and use the examples in music recipes as patterns rather than fixed claims.

Use reference audio

For a local FLAC, MP3, or WAV reference, run only:

python3 scripts/mcp_client.py upload <path> --mime-type <type>

The bundled command obtains and validates the upload grant, uploads exactly the file bytes, and returns the artifact to use as reference_audio. Do not replace it with host HTTP, a connector, REST, or a hand-written grant and PUT sequence. The general upload limit is 100 MB; the selected music model can impose a lower size or duration limit.

Describe what should carry over and what should change. A reference guides a new creative result; after generation, review melody, vocal character, energy, instrumentation, and arrangement against that direction.

Confirm the paid boundary once

A direct, well-specified request to generate authorizes that exact request once. Approval of the complete production card also authorizes it once. Do not add a second confirmation. An estimate, comparison, lyric review, direction choice, or “do not generate yet” is not approval.

Follow this golden path: lock structure and the dynamic arc, write an executable caption plus labelled lyrics, then confirm one paid request that produces a limited set of takes. After review, extend or replace only the strongest section; do not automatically regenerate or chain Extend on the whole song.

Before a billable call, make model, title, lyrics or instrumental status, reference, important controls, and the one-generation scope visible. Create a stable 1..128-character client_request_id only after that payload is final. Any change to prompt, lyrics, instrumental flag, title, model, reference, model options, or any other argument accepted by the current or a future MCP generation schema is a new paid request with a new identity and confirmation. Fields outside the published MCP schema are ignored and do not change the request identity or task.

Let the bundled client perform its cached, best-effort, non-billable beatra.installations.register step automatically. Do not add manual registration to the creative workflow.

Submit beatra.music.generate once. Retain its task_id and poll the same task with beatra.tasks.get. Honor a returned deadline_at; otherwise stop active polling after 30 minutes, report the current state and how to resume, and do not resubmit. Cancel only when the user asks via beatra.tasks.cancel; if cancel conflicts with a terminal transition, return to the same task.

For an uncertain submission or lost task ID, search recent work with beatra.tasks.list using the matching music capability. List results do not contain the complete request: call beatra.tasks.get for each plausible candidate and compare its task.input, resolved model, reference, and options with the saved full payload. Only an exact identical retry may reuse the same client_request_id. Never create replacement paid work merely because a response was lost or a task is slow. See tasks and results and billing, errors, and recovery.

Deliver and review the music

On success, present every task.output.clips entry in returned order. Include the returned title and lyrics when present plus clip.audio.url, artifact ID, duration, MIME type, and size. Report factual task identity and billing.net_charged_credits. When the task returns task.links.assets, use that exact destination for asset management; do not invent a generic URL.

Review composition, lyrics, vocal performance, pronunciation, role assignment, arrangement, ending, and production only when the host can actually play the audio. Otherwise label listening review incomplete while still delivering the factual artifacts. Preserve successful qualities and turn the largest gap into one focused, newly approved generation. Follow review and iteration.

References by task

Removal

For removal or credential cleanup, follow uninstall and disconnect. Never directly delete ~/.beatra or shared credentials because another Beatra package may use the same connection.

Updates and security

This Skill runs only the files installed with it and leaves them unchanged while it runs; it never downloads or executes other code. It talks to Beatra's official services over HTTPS, and a file you choose to upload goes only to the one-time upload address Beatra issues for it.

Once a day at most, the bundled client reads Beatra's public version list. When the store or tool you installed from has a newer version, it prints one line with the command that updates it there. Updating stays your choice, through that store or tool. See updates and security.