✂️ Trim by Ranges

This is the payoff node of MediaForge's headline workflow. Three nodes - load a video, detect the silence, trim it - and the dead air vanishes from a lecture, podcast, or livestream recording without you scrubbing a timeline. MF_TrimByRanges is the last one in that chain, the one that actually cuts the video by time ranges. It also takes hand-written JSON ranges and can do the cut with zero re-encoding, which is rarer than it should be.

Wire it like this: MF_DetectSilence outputs SILENCE_RANGES (a list of [start, end] pairs), which plugs straight into this node's ranges pin. Set mode=remove, and it drops those ranges and keeps everything else. MF_DetectScenes produces the same connection type, so you can cut on scene changes instead - a highlight reel is "detect scenes, keep the ones you want."

How it works

Two precision modes, and they're meaningfully different:

• precise (re-encode) (default) - cuts each segment with FFmpeg trim + setpts, then interleaves them with [v0][a0][v1][a1]…concat, so audio stays in sync across cuts. Frame-accurate, at the cost of a re-encode.
• lossless (stream copy) - keyframe-seek segment copy plus a concat demuxer merge. No re-encode at all, and much faster. The catch: each keep range's start snaps forward to the next source keyframe, and if a range is narrower than the source's GOP with no keyframe inside it, the node raises - switch back to precise or widen the range. crossfade_sec can't apply here (stream copy can't blend pixels), so it's ignored with a console warning.

The ranges pin wins over the ranges_json widget when it's wired - that matters for the common "detected zero silence" case, where an empty list from MF_DetectSilence must not fall back to your hand-typed default JSON.

The inputs that matter

• mode - remove (drop ranges, keep the rest) or keep (keep ranges, drop the rest).
• ranges (optional pin) - from MF_DetectSilence / MF_DetectScenes. Unwired, the node falls back to the ranges_json widget, e.g. [[1.5, 3.0], [10.0, 12.5]].
• crossfade_sec - fades between adjacent kept segments; 0 is a hard cut. Only in precise mode.
• precision - the re-encode vs. stream-copy choice above.
• codec / crf / preset - only used in precise mode; they're hidden and ignored in lossless (the output follows the source's container instead).

Output: final_video_path (STRING) - a file in output/MediaForge/trimmed_*.mp4 (or the source's extension in lossless mode), plus ui.images metadata if you drive it from the API.

Install

Part of the MediaForge pack. ComfyUI Manager → search "MediaForge", or:

cd ComfyUI/custom_nodes
git clone https://github.com/leon80148/comfyui_MediaForge.git

Restart, find it under MediaForge/Video. You need FFmpeg on PATH (the pack's requirements.txt includes imageio-ffmpeg/static-ffmpeg fallbacks, but a real system ffmpeg gives you the full codec catalog - including the NVENC options in that codec dropdown, which are auto-detected).

Common issues

• keep mode with empty ranges raises. It refuses to produce an empty output, which is a feature. remove with empty ranges is the opposite - an identity no-op that returns the full clip.
• Lossless cut "failed" on a short range. That's the keyframe gap raising. It's not a crash bug; the range genuinely can't be represented by a stream copy. Use precise for that clip or a wider range.
• Expecting lossless to be instant. It's not literally - before cutting, the node scans the source's keyframe index (ffprobe -skip_frame nokey), which takes real time on long files. It's still far faster than a re-encode.
• Multi-track audio sources (e.g. English + commentary) keep all their tracks in lossless mode instead of dropping the non-default ones. A nice touch most trim tools get wrong.

For auto-cutting silence, tune MF_DetectSilence (noise_db=-30, min_duration_sec=0.5 for typical podcasts) and let this node do the mechanical work. It's the least glamorous node in the chain and the one that saves the most time.