AI Song Cover Studio

Problem: Complexity in Reinterpreting Existing Songs

When a user wishes to cover, adapt the style of (e.g., rock, guofeng, acoustic), or rearrange an existing song, manual methods pose significant challenges: they require professional audio engineering knowledge, an understanding of diverse musical styles, and expertise in performance or arrangement. The traditional approach is time-consuming and has a high barrier to entry.

How the Skill Works: Core Capabilities & Key Steps

This skill specializes in a full-track reinterpretation of a reference song recording. The core workflow is as follows:

1. Analyze the Reference Recording: First, examine the user-provided audio file (supporting FLAC, MP3, or WAV) to identify the song's ID, language, mood, existing style, energy, instrumentation, and vocal characteristics.
2. Build a Production Card & Directive: Based on the analysis and the user's specified reinterpretation direction (e.g., style change, new arrangement), generate a clear production card. This card defines the song's metadata, target style, lyrics details to retain, etc., and writes a positive directive focused on the primary creative change.
3. Model Routing and Generation: Call beatra.models.list to verify if a currently available AI model supports the reference audio's MIME type, duration, and other parameters. By default, model: "auto" is used for vocal reinterpretation. For instrumental adaptations, a specific, explicitly supported model family must be selected.
4. Precise Submission and Polling: Use the bundled mcp_client.py tool to submit a frozen payload (containing the reference recording, production prompt, lyrics, etc.) exactly once to beatra.music.generate. Subsequently, poll the task status via beatra.tasks.get until the final audio asset or link is obtained, and review its performance, arrangement, and structure.

Key Control Points: The entire process heavily relies on the quality of the reference recording and a clear production prompt. Any modification to core inputs (reference audio, prompt, lyrics) triggers a new, paid operation.

Applicable Scope & Caveats

• Routing Limitation: This skill only handles requests for "reinterpreting a reference recording." Requests starting from lyrics, pure instrumental music, or requiring visuals are routed to other specialized workflows.
• Models and Cost: Model selection directly affects generation results and billing. The auto model has length restrictions for prompts and lyrics (e.g., 10–300 characters for the prompt). Real-time requirements for supported input formats, durations, and billing should be confirmed before use.
• Lyric Handling: The skill itself does not transcribe lyrics. If specific lyrics must be accurately retained, they must be actively provided by the user and confirmed before generation. Generation without provided lyrics cannot guarantee lyrical consistency.
• State and Recovery: Task states (queued, running) do not indicate failure. In case of exceptions, recovery should be attempted by replaying with the same stage ID or querying the task list, avoiding duplicate submissions that incur extra costs.