AI & Creative Tools

Plenio 2.0 Turns ComfyUI Into a Full Local Song Studio, With a Claude-Facing API Bolted On

The free, open node pack that writes lyrics, sings them, and masters the result now ships a GPU-aware setup wizard, a job API that survives a ComfyUI restart, and an MCP server so an AI coding assistant can request a song directly.

Most “AI makes a song” tools are a black box: type a prompt, wait, get an MP3. Plenio Music Production System, a free, open-source node pack for ComfyUI, takes the opposite approach — it exposes every stage of making a song (lyrics, vocals, mixing, mastering, cover art) as inspectable, swappable nodes running entirely on your own machine. Version 2.0, released November 14, is less about new sound and more about making that whole pipeline actually usable by people who aren’t already ComfyUI power users.

What it does, start to finish

A local language model writes the lyrics from a brief. YuE2 or MiniMax Music 3 performs them. Plenio then masters the result and exports a finished package: the audio, cover art, timed lyrics, and sheet music. The entire chain runs locally — no cloud render, no per-song fee — which is the project’s whole pitch against hosted music generators.

What’s actually new in 2.0

The headline change is a Plenio Setup tab that reads your GPU, lists exactly which model files a song needs, downloads only what you confirm, and verifies each file against its SHA-256 checksum before generating a one-minute test song. That tab then becomes a Home screen — a button per song type, a player, and your most recent songs — rather than the bare node graph ComfyUI normally hands you. Model fields in the bundled templates now default to auto, so a workflow built on one machine picks sensible files on another instead of hard-failing on a missing path.

The more technically interesting addition is a Song API v2 that covers all five song templates, survives a ComfyUI restart mid-job, and pauses at review points so a human can approve or edit a draft before it continues — useful for anyone scripting Plenio from outside ComfyUI’s own UI. Paired with that is an MCP server (plenio_mcp.py) that lets Claude Desktop or Claude Code request a song directly from a chat session, which then plays back through the Plenio Player; the player also exposes an HTTP API for stream decks and home-automation setups to control playback.

Why this is the interesting ComfyUI story this week

ComfyUI’s ecosystem produces a constant stream of new model-support nodes — this week alone brought official partner-node support for several new image and video checkpoints. Plenio is a different kind of release: it’s not wrapping a new model, it’s building production tooling (setup, job orchestration, external control) around models that already exist, aimed at making local AI music genuinely practical to use rather than just technically possible. That’s the gap most “run it yourself” AI tools get stuck on, and it’s the one Plenio 2.0 is explicitly trying to close.

Plenio 2.0 installs through ComfyUI Manager (search “Plenio Music Production System”) or via comfy node install comfyui-plenio-music, and requires ComfyUI 0.41.0 or newer; workflows saved under 1.x still run after a one-click model-field upgrade.