Merge Audio

Two different jobs, one node, and a dropdown that picks which

Audio work in ComfyUI needs two surprisingly different operations. Sometimes you want to layer tracks - dialogue over a music bed, or several takes summed together. Sometimes you want to string them - one voice line followed by the next, end to end. Most packs make you hunt down two separate nodes for that. easy audioMerge does both, and the merge_method dropdown is the whole switch.

It's the pack's audio glue, and it slots into everything: stacking voices for a dubbing pipeline, combining dialogue + SFX before feeding an audio VAE (like LTX-2's audio latent), or cleaning up a multi-track session into one stream.

How it works

The node accepts up to six audio inputs (it's an autogrow socket, so the count flexes), expands any audio lists you connect, and ignores empty list items and unconnected slots - so a list with three real tracks and seven placeholders merges the three and doesn't complain. Then it hands everything to a merge routine with your chosen method:

• add (default) - overlay tracks, summing the waveforms.
• mean - overlay, averaged. Softer than add when you're stacking several loud sources.
• subtract / multiply - overlay with those math ops. Mostly for audio-signal tinkering, not normal music work.
• after / before - concatenate. after appends each next track after the previous one (the usual "one line then the next" order); before reverses the stacking order.

Inputs and output

• audios - up to six audio inputs (autogrow). Lists in are expanded; empties are skipped.
• merge_method - the dropdown above.
• AUDIO out - the merged result, one object with waveform + sample rate.

That's the whole surface. No gain controls, no fades - it's the honest "merge these" node, and if you need per-track volume you'll want gain nodes upstream.

Setup and the traps

Pack install (Manager → ComfyUI-Easy-Media, or git clone + restart). No model files, and no FFmpeg needed for the basic merge - this is tensor math on waveforms.

The ways it bites:

• Overlay vs. concat confusion. If you select add expecting "add another clip at the end," you'll get both tracks at once, doubled and overlapping. add/mean overlay; only after/before extend the timeline. Half the "why is my audio muddy" posts on this node are really this.
• Clipping. add sums amplitudes, so two loud tracks can clip into distortion. That's physics, not a bug - mean or upstream gain staging is the fix.
• Sample-rate mismatch. Overlay methods assume comparable rates; if one source is 16 kHz and another 44.1 kHz, the result can sound off. Keep sources consistent (the pack's easy makeAudioList even warns its silence filler is 16 kHz for this reason).
• Order in concat. Input socket order determines output order. With after, socket 1 plays before socket 2 - put your tracks in the right slots.

For "these voices, in this order, one clip" or "this dialogue over this bed," it's the one node that covers both without a second lookup. The dropdown is the entire tutorial.