AI Music Generation and BGM Scoring

Problem

When making music for short videos, podcasts, product demos, or brand campaigns, teams often need usable BGM or candidate songs quickly without registering, configuring API keys, or switching between tools. This skill keeps generation, payment, and result retrieval inside the agent session: describe a style, scene, or lyrics, and it calls the locked suno_v5_5 Suno model. After a per-run payment, it returns 2 candidate songs with downloadable CDN audio URLs.

How it works

The flow starts with a prompt, full lyrics, or scene description. The first call triggers the WeChat Agent Pay 402 payment flow: the agent reads WeixinPay-Required, passes the paymentCode to weixinpay_pay, and retries the same request after authorization. The retry returns a project_id. Generation is asynchronous, typically taking 1-3 minutes, then the agent polls with status until data.project.status is completed and reads songs[].audio_url. If Node is unavailable, use curl against the same API; if local order state is lost, restore context with --order or out_trade_no. Important: use status polling, never rerun the creation command, to avoid a new paid run or order_consumed.

Boundaries

It is aimed at small candidate sets, quick comparison, and direct retrieval, not bulk generation, long-term library management, reference-audio remixing, stem separation, or precise chord and arrangement control. Each run produces exactly 2 songs, with no quantity or duration control; results are stochastic. Commercial licensing applies only after completed payment, and generated content is not guaranteed to be legally registerable or fully controllable at the music-theory level.