Seed-Nodes: SeedSaveAudio

ComfyUI has been doing audio for a while now - music generation, TTS, voice cloning, sound effects - but the save-your-audio story has always been thinner than the image side. SeedSaveAudio is a straightforward "write this AUDIO tensor to a file" node that lands in the audio category, and it's genuinely useful precisely because it's boring: it just works, and it gives you three format options to boot.

How it works

You feed it an audio input (the AUDIO type that audio-generating nodes output), pick a filename_prefix, pick a format, and it writes the file into your ComfyUI output/audio folder. Every batch item gets its own file, named your_prefix_00001.flac and so on, with a counter that keeps incrementing past files that already exist, so you won't clobber earlier renders.

The format choice actually changes behavior under the hood:

• flac (default) - lossless, and the only format that embeds your workflow metadata. The node writes a Vorbis comment block containing the prompt and extra PNG info, so your saved audio carries its generation recipe like a PNG does.
• wav - lossless, no metadata, and it normalizes the waveform to [-1, 1] before saving.
• mp3 - lossy, also normalized, and the one that depends on your torchaudio build having an MP3 encoder available.

There are no outputs, and it's marked as an output node, which is just ComfyUI-speak for "this is where the graph ends" - it'll appear in your queue results with a link to the file.

The inputs that matter

• audio - the AUDIO tensor from a generation node. The only thing you must wire in.
• filename_prefix - default "ComfyUI"; change it or every render lands in one folder with incrementing numbers.
• format - flac / wav / mp3. Pick flac unless you specifically need mp3's small size.

How to install it

Standard pack install - it lives in ComfyUI-Seed-Nodes along with everything else:

cd ComfyUI/custom_nodes
git clone https://github.com/Aerse/ComfyUI-Seed-Nodes
# restart ComfyUI

Or via ComfyUI Manager (search "ComfyUI-Seed-Nodes"). No model files involved. One dependency note: the node imports torchaudio, which is not listed in the pack's own requirements file. In practice ComfyUI's bundled Python environment ships torchaudio, so it usually just works - but if you run a hand-rolled environment, make sure torchaudio is actually installed or the whole pack will fail to load, not just this node.

Common issues

• Pack fails to load with an ImportError on torchaudio - missing dependency in your Python env. pip install torchaudio (matching your torch version) fixes it.
• mp3 export errors - MP3 encoding in torchaudio needs an FFmpeg backend; if it throws, either switch to flac/wav or get FFmpeg available to torchaudio.
• "Can't find the file" - it's in ComfyUI/output/audio/, not the root output folder. That subfolder trips everyone once.

If your workflow ends in audio, this is a decent, no-drama terminal node - and the flac metadata embedding is a quietly nice touch that most alternatives skip.