MiniMax Music Cover

Here's the node that makes people do a double-take: give it an existing track and a text prompt, and it returns a full cover version in the style you asked for. This is MiniMax's music cover model (minimax:music@cover) running on Runware's cloud, and it's genuinely one of those "okay, that's kind of wild" demos you can show a friend in thirty seconds - no local model, no GPU, no per-model setup. Just an API key and a song to feed it.

It's the sibling of the MiniMax Music 2.6 node in the same pack: 2.6 composes from scratch, this one reinterprets. Everything else about the pack is identical - install once via ComfyUI Manager (search Runware) or git clone https://github.com/Runware/ComfyUI-Runware into custom_nodes/, add your Runware API key in ComfyUI Settings, and the whole Runware catalog appears in the node menu.

How it works

You hand it a source track and a description of what you want it to become. The model takes the musical content of the input and re-renders it - the same melody and structure, different instrumentation, tempo, vocal treatment, and vibe per your prompt. Think "make this folk song into a synthwave track," not "EQ the drums." It's not remixing in the DJ sense; it's a full reinterpretation.

The one hard constraint to know going in: the input audio must be between 6 seconds and 6 minutes. Outside that window the request is rejected, so trim your source first.

Inputs and outputs

• audio (required) - the source track as a URL, path, or UUID. This is the gotcha everyone hits: it's not an IMAGE socket, it's a string, because the Runware pack routes audio inputs as media references. For a local file you'll need to get it to a URL (host it) or use Runware's media upload to get a UUID, then paste that in.
• positivePrompt (required) - the style you're converting to.
• settings.lyrics - optional, but if your source has vocals you can supply rewritten lyrics for the cover. Structured tags ([Verse], [Chorus]...) work here too.
• numberResults (1–4) - variations, each with its own seed.
• outputFormat - MP3, WAV, FLAC, or OGG; MP3 is the default.
• seed - pin it for a reproducible cover.

The output is a single audio (AUDIO) socket, same as the other Runware audio nodes, so it plugs into SaveAudio or anything expecting the native audio type. The node's title bar reports the cost per run after the fact - cheap enough that a few attempts to land the right take is fine.

The usual suspects

• "Audio input (UUID or URL)" rejections - if you paste a path or a plain filename it won't resolve. The tooltip says it plainly: UUID or URL.
• Duration out of range - under 6 seconds or over 6 minutes is a hard no, and the error surfaces cleanly as a one-liner on the node.
• No API key - the pack's standard failure: "No Runware API key. Set RUNWARE_API_KEY or run runware auth login."

Worth a slightly skeptical take: music covers are where model quality gaps show most, because the source track is a strong prior and the model can drift from it. Don't expect a faithful note-for-note cover every time; treat it as a creative reinterpretation generator and you'll be happy with the results. For zero-setup, pay-per-run restyling inside ComfyUI, nothing else in the pack hits quite like it.