Save Audio

What it is

An output node that writes ComfyUI AUDIO streams to disk as wav, flac or mp3, with a real choice of sample rate and bit depth instead of whatever the graph happened to carry. It has no outputs at all - it's a terminal node, like Save Image. It runs whenever the graph runs.

The reason to use this over ComfyUI's core audio savers is control. Core's SaveAudio family (FLAC, MP3, Opus) is marked deprecated in favour of SaveAudioAdvanced, and while those work, they expose a filename_prefix with formatting tokens rather than a fixed set of format/rate/depth options. Here you pick the exact thing: 24-bit FLAC, or 320k MP3, or float WAV, at 32/44.1/48kHz.

Inputs

• audio - the stream. Wire a stem from any separator, or the output of Audio Ensemble.
• output_format - wav, flac, mp3.
• sample_rate - 32000, 44100, 48000, default 44100. If the incoming audio is at a different rate it gets resampled on the way out.
• wav_bit_depth - FLOAT, PCM_24, PCM_16. Defaults to FLOAT, which is the right choice while you're still processing; PCM_24 for delivery.
• flac_bit_depth - PCM_24 or PCM_16, default PCM_24.
• mp3_bit_rate - 128k, 192k, 256k, 320k, default 320k.
• filename - optional, and forced: it's an input socket, not a box you can type in. So you have to build the name from something else. Load Audio's audio_name plus a separator's stem_N (String) through core's Concatenate Text is the pattern the pack's own README suggests, and it gets you mysong_vocals.wav instead of a timestamp.

The bit-depth and bitrate widgets are hidden dynamically - pick mp3 and only mp3_bit_rate survives on the node, which keeps it from becoming a wall of irrelevant dropdowns.

Where the files land, and why that surprises people

Straight in ComfyUI's output root - ComfyUI/output/ - not ComfyUI/output/audio/ where core's savers put things. If you're used to finding audio in a subfolder, you'll think the save failed. It didn't.

Naming rules, in order:

1. filename if you wired one and it isn't empty.
2. Otherwise audio_YYYYMMDD_HHMMSS.
3. With a batch bigger than one, each item gets _00000-style indices appended.
4. Existing files are never overwritten. If the target path exists, a numeric suffix is appended (mysong_vocals_00001.wav). This is the pack being careful, and it's why a rerun of the same graph leaves you with a growing pile - the second run doesn't replace the first.

Saving is done through pymss's own AV-based writer, and the list of paths it wrote comes back to the frontend, so the node reports what it did.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/pymss-project/comfy-mss
python -m pip install pymss   # ComfyUI's Python env

Manager → Comfy-MSS works too. Restart after. The av dependency that does the actual encoding arrives with pymss; you don't install it yourself.

Gotchas

• No passthrough. Unlike core's SaveAudio, this node doesn't forward the audio, so you can't chain a preview after it. Branch the wire: one copy to Save Audio, one to a preview node.
• Wire the filename or accept timestamps. It's a forced input on purpose; the timestamp fallback is a safety net, not a filename strategy.
• Sample rate is a real resample, not a label. Saving a 44.1kHz stem as 48000 genuinely resamples it; that's usually what you want when the downstream video pipeline expects 48k.
• MP3 at 320k is still MP3. For stems you intend to process further, save FLOAT WAV or 24-bit FLAC and encode once at the end.