MD: Audio Simple Editor ✂️

Generative audio almost never ends exactly where you want it. There's a stray artifact in the first second, a generation tail that cuts off abruptly at the end, or a loop that clicks because the ends don't meet. MD_AudioSimpleEditor does the three things you need for all of those - trim the start, trim the end, and fade the boundaries - right in the graph, no DAW export round-trip.

It takes a standard ComfyUI AUDIO dictionary and edits the waveform samples directly: cut out the first N seconds, cut everything past an absolute end time, then apply fade-in/fade-out ramps. That's the whole feature set, and "simple" is doing honest work in the name.

The inputs

• audio - the AUDIO to edit, from LoadAudio, an ACE-Step decoder, wherever.
• trim_start_sec - seconds to remove from the beginning. The tooltip's suggestion is the practical one: use it to strip generation artifacts at the start of a track.
• trim_end_sec - the absolute end time of the track, measured from the original start, not from the trimmed start. 0.0 keeps the original length; anything above 0 cuts everything past that point. It's an easy detail to get backwards the first time.
• fade_in_sec / fade_out_sec - durations of the ramps, applied after trimming. The tooltip's recommendations are worth taking: even a tiny 0.05s fade-in prevents click/pop artifacts, and 1–3s fade-out smooths an abrupt cut into a natural ending. Essential for seamless loops.
• fade_curve - Linear or Exponential. Linear for short crossfades, Exponential for musical fade-outs - exponential tracks human hearing better and sounds more natural on anything longer than about a second.

Outputs: edited_audio (the trimmed, faded AUDIO, chain it into your save node) and waveform_plot (an IMAGE of the result, handy for confirming the trim visually before you export).

How it fits a workflow

The natural pipeline is generate → Simple Editor (strip the dead start, fade the tail) → Guardian (make sure nothing's blown out) → save. For looped audio, the discipline is: find the loop point, set trim_end_sec to it, and put a small fade-in and fade-out so the seam is inaudible - the tooltip calls fades "essential for seamless loops," and it's right.

Installing it

Part of MD Nodes (MDMAchine/ComfyUI_MD_Nodes). ComfyUI Manager → search MD_Nodes → Install, restart, or:

cd path/to/ComfyUI/custom_nodes
git clone https://github.com/MDMAchine/ComfyUI_MD_Nodes.git
cd ComfyUI_MD_Nodes
pip install -r requirements.txt

The waveform plot needs matplotlib (in the pack's requirements); editing itself is plain tensor math. Python 3.10+, current ComfyUI.

Common issues

Memory is the quiet gotcha the tooltip flags: editing keeps the whole waveform in memory, so very long tracks (an hour+) can get heavy - trim early in your chain rather than after piling up generations. And the trim_end_sec measurement base trips people once: it's absolute from the original start, so if you trim 10s off the start and set end to 30s, you get 20s of audio (10→30), not 30s. That's by design; read the tooltip before you call it a bug.