MschA2V Preview Audio

MschA2V Preview Audio is the least glamorous node in the msch-a2v pack, and one of the most useful. It takes any AUDIO value and writes a temporary, playable preview file you can hear in the ComfyUI UI - no Save node, no digging through your output folder, no detour into a video encoder just to check a sound.

Why does that matter in this pack specifically? Because the whole msch-a2v chain is about audio timing. Before you commit GPU time to sampling a sequence of MiniMax H3 shots, you want to know that the audio window you're about to generate against is actually the window you meant - the right section of the song, the right lead-in, the right length to match your authored blocks. The sequencer's audio output is windowed and padded to be exactly as long as the assembled video expects, and this node is how you check that without burning a render. A misjudged window discovered after a multi-minute H3 render is an expensive way to learn a lesson this node teaches in five seconds.

The whole interface

One required input: audio (type AUDIO). That's it. No optional inputs, no outputs - it's marked as an output node, which in ComfyUI terms means it can sit at the end of a wire and render a UI result. What it does internally is honest plumbing: it takes the first batch of the waveform, writes it to a FLAC file (named something like mscha2v_preview_<random>.flac) in ComfyUI's temp directory via soundfile, and hands that back to the frontend as an audio-preview payload, which the UI turns into a playable widget.

The practical routing: wire the audio output of MschA2V Beat Prompt Sequencer (or MschA2V Load Audio Path, or pretty much any other AUDIO source in your graph) into it, run the graph, and click play on the preview that appears. Want to hear the version with your audio_start_seconds/audio_duration_seconds overrides applied? The sequencer's output already reflects them, so previewing that is previewing exactly what will be rendered. That's the workflow the node was built for - audition first, generate second.

Install and notes

Nothing exotic to install beyond the pack itself, which is one of the lightest custom-node installs around:

cd ComfyUI/custom_nodes
git clone https://github.com/mariobilly/msch-a2v.git
cd msch-a2v
python -m pip install -r requirements.txt   # adds numpy, librosa, soundfile

Restart ComfyUI. Soundfile (the dependency that actually writes the FLAC) comes from that requirements file, so don't skip the pip step or the node will fail at the write with an import error. One heads-up: the preview lives in the temp directory, so it's disposable by design - if you want to keep an audio render, this isn't your node, and the pack doesn't ship a dedicated save-audio node, so you'd wire AUDIO into whatever video/audio saving node the rest of your install provides. Also worth remembering that the standalone msch-a2v repo has been superseded by the unified MSCH Nodes pack, so new installs should prefer that (migrate to avoid duplicate nodes). For a utility this small, the best compliment is that once it's in your graph you'll forget it's there - until the day it stops a wasted render.