Audio Blend

Audio Blend (SET_AudioBlend) is the pack's mixer. It takes two audio signals, applies an independent volume to each, and adds them together into one output. Musically that's called a mix; in signal terms it's "scale and sum." Either way, it's the node you reach for when you want background music under a voice, two sound effects layered, or simply one clip turned up or down.

Inputs

• audio1 - the first signal. This one is required.
• gain1 - volume for audio1, from -10 to 10 in 0.01 steps, default 1.0.
• gain2 - same range, same default, for audio2.
• audio2 - the second signal, and notably optional.

That optional second input is a quiet superpower. With audio2 left unplugged, the node becomes a plain gain stage: wire in one clip and set gain1 to 2.0 to double its volume, or 0.5 to halve it. Since the gains can go negative, it's also a subtractor - plug the same signal into both inputs with gain2 = -1 and you get near-silence, which is a genuinely useful trick for checking whether a signal is the same as another.

The output, audio_out, is the summed signal. When you blend two clips of different lengths, the shorter one gets padded with silence to match - the same alignment philosophy as the rest of the pack.

How it fits a workflow

The obvious use is layering: generate an instrumental, generate a voice track, then blend them with music at 0.3 and voice at 1.0. Because both inputs accept batches, you can also blend batch-to-batch, which is how the pack's generate_and_blend.json example builds a four-note chord - generate four musical notes, blend them together with per-note gains, and out comes a chord.

There's a real-world caveat worth knowing: raw summing clips. If both inputs are at full volume (gain 1.0 each) and both are loud, the peaks add up and can exceed 1.0 - that's clipping, and it sounds bad. This node won't stop you from doing it. If you're mixing multiple sources, either keep the gains modest or run the result through Audio Normalize (Peak) afterward to pull the peak back down. Normalize can't restore information already lost to clipping, but it keeps you from feeding an over-1.0 waveform downstream.

Install

From ComfyUI Manager, search "Audio Batch", or:

cd ComfyUI/custom_nodes/
git clone https://github.com/set-soft/ComfyUI-AudioBatch
pip install seconohe

Restart ComfyUI. No models, no heavy deps.

Gotchas

The main trap is forgetting the clipping math above - two 1.0-gain loud tracks will sum past full scale. The other is expecting real stereo processing: blend adds channel-wise, so a stereo + stereo mix keeps both channels, but there's no panning, ducking, or EQ here. It's a summing mixer, not a DAW. If you need volume automation per clip in a batch, that's what the paired Audio Apply Batched Gain node is for; Blend is the "two signals at fixed levels" tool.