Mureka AI Mureka V9 Generate Song (Official Edition)
Full songs from lyrics on Mureka V9 — the node at the end of the music pipeline
- melody
- reference
- vocal
- api_config
- audio
- request_id
- urls
This is the payoff node for the whole Mureka shelf in BizyAirPlus. Everything else in the family - the lyrics LLM, the Extend Lyrics, the BGM generator - feeds into this one: it takes real lyrics and turns them into a finished song with vocals. If you're building a music pipeline in ComfyUI, this is where it pays out.
What it is
Mureka V9's official text-to-audio full-song generator, surfaced as an "Official Edition" BizyAir node. It's the vocal version of the BGM node: lyrics in, complete sung track out, rendered in the cloud.
How it works
The lyrics input is the song's text - up to 3000 characters, and structure matters. Mureka reads your section markers ([Verse 1], [Chorus]) and sings them as written, so the quality of your lyric scaffolding is the quality of the song. n (1–3) sets how many takes to render per request. Then the fun part - the optional audio inputs that let you steer the sound instead of rolling dice:
- melody - control production with a melody reference.
- reference - control the overall music with a reference track.
- vocal - control the voice itself through a timbre reference.
Want a specific singer's vibe, a melody you hummed, or a production style? That's where you put it. prompt adds up to 1024 characters of extra direction on top.
The inputs and outputs
- lyrics - required (up to 3000 chars).
- n - required, 1–3 tracks.
- melody / reference / vocal - optional audio steering.
- prompt - optional text direction.
- outputs - audio, request_id, and urls (direct links to the rendered tracks).
api_config and skip_error are the pack standard.
Installing
ComfyUI Manager → "BizyAirPlus", or:
cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt
Restart, turn the BizyAirPlus button ON, enter your API key from bizyair.ai.
Where people get burned
The usual suspects plus one big one. URL-hosted audio references (local files won't upload themselves), and n multiplying your bill. But the big one is structure: Mureka sings your lyric file's structure, and if your lyrics are a wall of text, you get a wall of sung text. Build verse/chorus/bridge scaffolding in the cheap LLM node first. And know the platform context - Mureka's generation is watermarked and content-filtered like its peers, so this is a tool for legit production, not a loophole.
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| lyrics | STRING | [Verse 1] 红灯停滞在暗房 像一颗不肯熄灭的盲肠 你留下的侧影 还在银盐液体里升华、膨胀 显影液太吝啬 倒不出整座有你的过往 桌角堆着 没被认领的二月 还有整面哭墙 指尖沾上定影粉 抹一抹 窗外就是北方的霜 [Verse 2] 胶片颗粒在抖动 像心跳过了期的反响 你没带走的底片 正等待着一场永恒的散场 光纤穿过指纹 编织成谁也读不懂的虚妄 我把所有关于你的像素 归档进最深处的抽屉 严严实实封上 [Chorus] 摄影是把瞬间变成不朽 也是把活人变成标本的战场 我们在光与影的缝隙 练习如何体面地遗忘 别回头看 那些已经成灰的影像 只要底片还在燃烧 就算是在这零度的暗房 也能听见光落下的声响 [Outro] 雪落了。 暗房的门 终于锁上。 | Generate lyrics for music, up to 3000 characters. |
| n | INT | 11–3 | Number of songs generated |
| melodyopt | AUDIO | Control of music production through melody | |
| referenceopt | AUDIO | Control music generation with reference music | |
| vocalopt | AUDIO | Control music generation through timbre | |
| promptopt | STRING | Through the input prompt word control music generation, maximum 1024 characters. | |
| api_configopt | BIZYAIR_OPENAPI_CONFIG | 单次请求覆盖base_url和api_key | |
| skip_erroropt | BOOLEAN | false | 开启后遇到错误不中断工作流,输出对应类型的错误占位符 |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| audio | AUDIO | — |
| request_id | STRING | — |
| urls | STRING | — |