Mureka AI Stem Song (Official Edition)

Stem separation is one of those things that used to mean installing a specialized tool and babysitting it. This node reduces it to: drop in a song URL, get back the separated parts. It's Mureka's official audio-to-audio separation endpoint - the "Stem Song" node - running through BizyAirPlus, and it quietly unlocks a whole remix workflow inside ComfyUI.

What it is

An "Official Edition" Mureka API node. You give it a mixed song, it returns the audio split into its component stems - the rails, in the node's own words. It's the a2a (audio-to-audio) sibling of the Mureka generation nodes, and it's the one you reach for when you already have music and want to take it apart rather than make something new.

How it works

The audios input takes the song URL (MP3 or M4A, per the tooltip), and the model input - default audio-separation-1 - picks the separation engine. Mureka's service runs the source separation and returns the stems. One caveat worth knowing before you plan a workflow around it: this node returns a single audio output, so check how BizyAir bundles the separated rails - if you need individual vocal/drum/bass files as separate wires, you may need to pull them from urls instead.

The inputs and outputs

• audios - required. The song to process.
• model - required. The separation model; audio-separation-1 is the default.
• outputs - audio, request_id, and urls (direct links to the separated files).

api_config and skip_error are the standard pack options.

Installing

ComfyUI Manager → search "BizyAirPlus", or:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart, toggle the BizyAirPlus button ON, enter your API key from bizyair.ai.

Where people get burned

First, the URL requirement - local audio must be hosted before the AUDIO input will take it. Second, expectations: separation quality on clean masters is genuinely good now, but a compressed, low-bitrate source will produce stems with audible bleed; feed it the best master you have. And mind the legal/practical line - separating stems for remixing your own material is the intended use, and the platform's terms are what they are, so don't assume it's free of constraints just because it's in your graph.