Save Audio Advanced 🔊 🅑🅔🅣🅐

ComfyUI has a growing number of things that emit audio - TTS nodes like Orpheus or Soprano, LTX-2's video-and-audio output, VHS's Load Audio - but until fairly recently, saving that audio to disk in a format you'd actually use was oddly fiddly. Save Audio Advanced exists because the built-in options were thin on format control. It takes an AUDIO input and writes it out as FLAC, WAV, or MP3, with format-specific knobs, into your normal ComfyUI output directory. If you're doing TTS voice work or pulling audio out of a video-gen pipeline, this is the node you reach for.

The mechanism is straightforward: it normalizes whatever audio shape it receives and hands it to torchaudio.save with the right options for the format. That "whatever shape" bit is the part that's genuinely useful - ComfyUI's AUDIO type shows up as a (tensor, sample_rate) tuple from some nodes, a {'waveform': ..., 'sample_rate': ...} dict from others, or even a wrapped one-element tuple. This node accepts all three and converts internally, so you stop debugging type mismatches and just save.

Inputs that matter:

• audio - the AUDIO output from your TTS / Load Audio / video node.
• filename_prefix - prefix for the file (default ComfyUI); ComfyUI appends date + counter, so you get ComfyUI_00001_.flac.
• format - flac, wav, or mp3. Default flac.

Then the format-specific ones: wav_encoding (PCM_16 / PCM_24 / PCM_32 / FLOAT_32 / FLOAT_64) for WAV, flac_compression (0–8, default 5) for FLAC, and mp3_bitrate for MP3. It's an output node, so there are no outputs - the result is a file plus a UI entry showing where it landed.

Install is the pack install: ComfyUI Manager → search "BETA" or "Burgstall" → ComfyUI-BETA-Cropnodes (now renamed ComfyUI-BETA-Helpernodes; either name works). Or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/Burgstall-labs/ComfyUI-BETA-Cropnodes

then restart. No model downloads. One real dependency to know about: MP3 saving requires FFmpeg with libmp3lame on your system and in PATH. WAV and FLAC don't need anything beyond what ComfyUI ships (torchaudio).

The gotchas are honest ones, mostly documented by the author himself. The mp3_bitrate input is currently ignored - the README admits MP3 saving uses the backend's default bitrate because the torchaudio API wouldn't take the hint. So if you set 320kbps and get 128kbps, that's expected, not a bug. And if MP3 saving fails with a backend error, it's almost always the FFmpeg/LAME install. One more: the node is chatty in the console with debug prints on every run. Annoying, but it means when a save silently fails - which torchaudio can do - the node tells you exactly which step died.