XB-BOX - 🎵 YuE2 翻唱准备

YuE is the open-weights song generator that made everyone say "Suno at home" in early 2025 - Apache-2.0, fine-tunable, and openly generating full songs with vocals. Where ACE-Step reaches for a song in one diffusion pass, YuE2 works like a studio: a planner writes an editable ABC score from your style and lyrics, then a renderer sings that score. That split is the whole reason this node exists, because it means you can hand the renderer a melody instead of a song.

XB-BOX - 🎵 YuE2 翻唱准备 (Cover Prep) is the node that builds those cover conditions. One caveat before anything else, because it's the assumption everybody arrives with: a cover here supplies a symbolic melody. It does not carry the source singer's identity or their waveform. It is not voice cloning, and no amount of tuning in this node makes it one.

What it does

Feed it a score - either the abc output of the official YuE2GenerateABC node or something you brought in through YuE2 Transcription Intake - and it reconditions that score for a cover using two choices.

保留声部 / "Keep Voice" decides which line survives: keep both, keep vocal only, or keep instrument only. Choosing one side doesn't delete the other; the notes are replaced with rests of equal value, keeping the chord symbols and the bar structure intact. That matters downstream - the renderer still sees a well-formed, harmonically marked score, just one with a hole where a voice used to be.

和声处理 / "Harmony" is the choice that silently sets your workflow's mode. Drop the chords and you get a pure melody line, which means the downstream mode is melody - the official definition of a cover: the accompaniment and harmony are regenerated in whatever new style you asked for. Keep the chords and the original harmony rides along, so the downstream mode is full, which is closer to a re-record than a reinterpretation.

That mapping is not a suggestion. melody mode rejects any chord symbol in the score outright, so a mismatch between this node and the renderer's mode setting shows up as a failure, not as a subtly wrong song.

Inputs and outputs

The two dropdowns above are required; abc is the optional input for the source score. Outputs are abc (STRING) - wire it into the official music node's score input, or into Score Workbench for another pass - and 报告 / Report (STRING), which carries the before/after event comparison plus the official cover flow. Read the report. Its headline carries the mode suggestion that the harmony choice implies, and the body has the before/after event comparison.

Note what's not here: instruments. The ABC score has no instrument field at all, and moving a voice around doesn't pick one for you. The lead and supporting instruments come from the YuE2 Style Prompt node's 主奏乐器 / 辅助乐器 widgets. If your cover comes out piano when you wanted strings, that's where you fix it.

The trap the author actually warns about

The official guidance ships with the node's own tooltip: don't drop the Ins track just because the user asked for a cover. In a lot of songs the instrument line carries the theme, the riff, or a solo - strip it and you've thrown away the part people recognize. Keep both when in doubt, and let the renderer figure out the arrangement.

Install

These music nodes are pure string/int/float emitters with a standard-library ABC parser, so they load on any ComfyUI build. But they only do anything next to the official native YuE2GenerateABC / YuE2GenerateMusic nodes, so install a build that ships those and pull the official text-to-music workflow to wire into. Then:

cd ComfyUI/custom_nodes
git clone https://github.com/wjluoxiao/XB_ToolBox.git

or search XB_ToolBox in ComfyUI Manager and restart. The pack's README claims no pip dependencies; the repo's requirements.txt is a longer story, but none of it is needed for the YuE2 nodes.