Cut a video in one tool and it finishes instantly with perfect quality; cut the same file in another and it grinds through a re-encode. Neither tool is broken — they're making opposite trade-offs against the same constraint: how video compression works.
Video isn't a stack of pictures
A modern codec (H.264, H.265, VP9, AV1) stores only occasional complete frames — keyframes (I-frames) — typically one every 2–10 seconds. Every frame in between is stored as a diff: "like the previous frame, but these blocks moved." The run from one keyframe to the next is a GOP (group of pictures).
This is why video files are small, and it's the entire cutting problem: you can only start playback cleanly at a keyframe. A frame mid-GOP is meaningless without the chain of frames before it.
The two ways to cut
Lossless (stream copy). Copy the compressed data untouched, starting at a keyframe. Instant (it's basically a file copy) and pixel-perfect — but your cut points snap to keyframes. With keyframes every 5 seconds, "cut at 1:23" becomes "cut at 1:20," and the clip opens with up to a few seconds of unwanted footage. Tools: ffmpeg's -c copy, LosslessCut.
Re-encode. Decode the video to raw frames, cut at exactly the frame you chose, encode the result as a fresh stream with its own new keyframes. Frame-accurate and universally compatible — but it costs time and one generation of compression quality.
There is a hybrid ("smart cut": re-encode only the partial GOPs at the edges, copy the middle), but it's finicky across codecs and few tools do it reliably.
What a browser splitter does
The split.tools video splitter takes the re-encode path, using the browser's own capture pipeline: your marked segment plays back and is captured to WebM at up to 8 Mbps in real time. That means frame-accurate in/out points and zero installs, at two honest costs: export takes the clip's duration, and the output is one compression generation removed from the source.
Does the generation loss matter?
Depends on the destination:
- Social platforms: no. TikTok, Reels, Shorts, and YouTube all re-encode every upload to their own delivery formats, usually at bitrates below 8 Mbps for feed video. Your clip was getting re-encoded anyway; one clean intermediate generation is invisible after theirs.
- Archival masters: yes. If the clip is the deliverable — festival submission, master copy, footage for further editing — keep a lossless chain: cut with stream-copy at keyframes, or re-encode once at very high bitrate from the original.
The rule of thumb
Frame-accuracy and convenience for clips headed to platforms that re-encode anyway; lossless stream-copy for masters, accepting keyframe snapping.
For the daily job — slicing a long recording into social clips — frame-accurate browser cutting with no upload and no install is the right trade, and your footage never touches a server while you make it.