SloppyAudio Fade

Audio in ComfyUI is the layer nobody designed but everybody ends up needing. You render a Wan clip, drop in a voiceover, lay down a synth loop - and it all starts with a hard, abrupt click and ends the same way. That's the problem this node exists for. SloppyAudio Fade is a fade-in/fade-out you drop on any audio wire so your clip starts from silence, ends in silence, and stops assaulting the listener.

It's not glamorous. It's one of those finishing nodes that makes a whole workflow feel professional for zero effort. If you've ever bounced a video and cringed at the first frame of sound, this is your fix.

What it actually does

Two knobs and a dropdown. fade_in and fade_out are seconds, from 0 to 60 in 0.1 steps - set fade_in to 0.5 and your clip fades up over the first half second, set fade_out to 1.0 and it sinks out at the end. Both default to 0, which means pass-through, so the node does nothing until you ask it to. Set both to 0 and you get your audio back untouched.

The fade_type dropdown is where the character lives: linear, quarter-sine, half-sine, logarithmic, and parabolic. Linear is the default and it's fine for voiceover. For music, a sine curve feels far more natural - the ear reads it as a fade rather than a volume slam. Logarithmic and parabolic are there when you want a faster attack into a long tail. Honestly, half-sine is the one I reach for most; linear can feel abrupt on anything musical.

There's exactly one output, audio, and it wires straight into whatever consumed your original audio - a Save Audio node, a video muxer, your next effect.

How it works under the hood

This pack wraps SoX - the Swiss Army knife command-line audio tool - and Fade is a direct pass-through to its fade effect. The node converts your ComfyUI audio to a temp WAV file, runs sox in.wav out.wav fade <curve> <fade_in> 0 <fade_out>, then reads the result back into the AUDIO format ComfyUI's video nodes expect. The curve names are SoX's own, which is why quarter-sine/half-sine live in the list.

Two behaviors worth knowing because they'll confuse you once. If both durations are under 0.01s, the node short-circuits and returns your audio unchanged. And if SoX fails for any reason, it also returns the original audio - silently, aside from a line in the ComfyUI console. So "nothing happened" and "it failed" look identical in the preview. If a fade just isn't happening, open the console and look for [SloppyAudio] SoX error.

Installing it

This is the shared story for all four nodes in the pack. Easiest: ComfyUI Manager → search sloppyaudio → Install → restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/forcepusher/comfyui-sloppyaudio

Then restart ComfyUI. Dependencies (torch, numpy, soundfile, librosa, einops, safetensors) install automatically via Manager, or pip install -r requirements.txt if you're doing it manually. The good news here: SoX binaries are embedded in the pack - there's a bin/ folder for Windows, macOS, and Linux, so no separate SoX install and no system package dance. That's genuinely rare for an audio pack and it removes the biggest install gotcha in this corner of ComfyUI.

Where people get tripped up

• It needs an AUDIO input from somewhere. This pack can't load an audio file itself; wire in ComfyUI's built-in Load Audio (or whatever node produced the audio you're cleaning up). Feed it a proper AUDIO wire or the node just passes None back.
• Silent failure. As above, SoX errors return the original audio. Watch the console.
• librosa is a chunky dependency. It drags in a fair bit on a fresh install, which is the usual source of "why is my environment suddenly slow to start" - normal for audio nodes, per the broader ecosystem pattern.

Fade is a five-minute add to any workflow that touches sound, and once you've used it you'll never bounce another clip without one.