🔊✂️AGSoft Audio Split

🔊✂️AGSoft Audio Split is the node you reach for when you've got a TTS clip that starts with two seconds of silence, or a generated voiceover with a ruined first word, and you just want the good part. It cuts a fragment out of an audio file by start_time / end_time - and, unlike a lot of "audio editor" nodes, it can do it without re-encoding when you want speed and losslessness.

The core choice is the cut_mode:

• fast - stream copy. FFmpeg copies the audio packets without decoding and re-encoding them. It's near-instant and bit-perfect, but the cut lands on a codec-frame boundary, so the exact in/out points may shift by a few milliseconds.
• precise - re-encode. The cut is sample-accurate. Costs time and a tiny generation loss, but the timestamps are real.

format defaults to source, which keeps whatever the input was (with fast this means no re-encode at all). Pick mp3, wav, flac, ogg, or m4a to convert instead - conversion always re-encodes, and bitrate (96k–320k) only applies to the lossy ones. output_name sets the filename (the extension is derived from the effective format), output_path is the folder, empty = ComfyUI/output.

The gotcha that will bite you

The source audio can arrive two ways: through the audio_path string (a file on disk) or through the audio AUDIO socket (from another node). These are not the same thing. If audio comes in through the socket, it has already been decoded to PCM - so "keep the source format" produces a WAV, not your original MP3. True stream copy is only possible when the file is connected directly via audio_path.

So the rule is: for a lossless, instant cut of a file, feed it the path. If you've already got audio objects in a graph (say, after Load Audio), either accept that you're getting a WAV, or explicitly choose the format you want.

Outputs

This node is generous with what it returns:

• audio_path (STRING) - where the result landed on disk.
• audio (AUDIO) - the fragment as an audio object, so you can keep chaining without touching the filesystem.
• duration_seconds, duration_timecode, file_size_mb - handy numbers for downstream logic or a display node.
• sample_rate, channels - the technicals, if a downstream node needs to know what it's getting.

Install and notes

It's part of the comfyui-AGSoft pack. ComfyUI Manager → search "comfyui-AGSoft" → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Art-xmaster/comfyui-AGSoft

Restart ComfyUI and you're done. No model downloads, no extra Python deps for this node - it shells out to the FFmpeg that ComfyUI bundles (or a system ffmpeg on your PATH).

If you find yourself cutting and then wanting fades, normalization, reverse, or looping on the same fragment, that's the "Plus" sibling (AGSoft Audio Split Plus) - this one is the clean, fast, no-frills cutter.