RT HeartMuLa Preview

The sampler gives you a song, and then you need to actually hear it. RT HeartMuLa Preview is the output node at the end of the pack's workflows that takes the AUDIO result and turns it into something you can play in the ComfyUI browser and find on disk afterward.

It's a deliberately boring node - one input, no knobs, no outputs - and that's fine. When a workflow generates minutes of audio, the node that makes it audible without hunting through output/ for a randomly-named file is the difference between "cool, let me try another seed" and "where did my song even go".

What it does

Wire the AUDIO output of RT HeartMuLa Sampler into its single audio input and hit queue. The node saves the waveform with torchaudio into your ComfyUI output directory as a .wav (48kHz, matching what the sampler produced), names it with a HeartMuLa_ prefix plus an incrementing counter, and returns a UI entry so the player appears right in the browser.

It's an OUTPUT_NODE, which in ComfyUI terms means it's a terminal point - nothing can come out of it, and the graph treats it as a "done" marker. If you're batch-generating (the loader's compile_model mode exists partly for this), the node iterates over the whole batch and writes one file per waveform, each with its own counter.

How it differs from SaveAudio

Core ComfyUI already ships a SaveAudio node that does basically the same job. The practical differences are the automatic HeartMuLa_ filename prefix and the fact that it's already wired into every example workflow this pack ships, so you don't have to think about it. If you want custom filenames or a subfolder, use the core SaveAudio instead - it's genuinely interchangeable here.

Installing it

There's nothing node-specific to install - this node rides along with the rest of the pack. Get the pack once, and you get all seven nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/monnky/ComfyUI-RT-HeartMuLa
cd ComfyUI-RT-HeartMuLa
pip install -r requirements.txt

or search "ComfyUI-RT-HeartMuLa" in ComfyUI Manager. The heavy lifting (models, GPU drivers) is all on the loader side; this node only needs soundfile (in the pack's requirements) and torchaudio - which, amusingly, the requirements.txt omits even though the code imports it. Most ComfyUI environments already have torchaudio; if not, pip install torchaudio fixes it.

Common issues

• The node runs but no player appears - check ComfyUI/output/ for the HeartMuLa_*.wav files. If they're there, it's a frontend hiccup; refresh the page.
• "No audio" - make sure the sampler actually returned something and wasn't bypassed or muted; the sampler needs all four loader handles wired in, or it errors before audio ever exists.
• Wrong sample rate / weird pitch in an external player - the file is 48kHz; some old players assume 44.1kHz. Play it back in the browser or a modern player and it'll be right.