SmartCut is a desktop application for cutting Japanese TV recordings. Drop a
night's worth of .ts files onto it and it finds the commercial breaks for you.
You check the boundaries, cut, and export.
What matters is how it exports. A normal video editor takes the whole file apart into pictures and builds it again from scratch. That takes time, and every frame comes out slightly worse than it went in.
SmartCut does not do that. Video is stored as key frames, which are complete pictures, and frames that hold only what changed since the last one. Cut on a key frame and the rest of the file can simply be copied, byte for byte. Only the few frames caught between key frames at a cut point have to be rebuilt.
On real broadcast recordings, more than 99% of the output is an exact copy of the input — and when the cuts land on key frames, which commercial breaks usually do, nothing is rebuilt at all.
This technique is called smart rendering. SmartCut applies it to the video, to the audio, and to the subtitle and programme-information streams a Japanese broadcast carries alongside them.
No quality is lost unnecessarily. Re-encoding a two-hour recording makes every frame in it slightly worse. SmartCut touches a few dozen frames at most, and often none at all.
It is fast. Copying bytes is limited by your disk, not your processor. A 30-minute recording is written out in well under a minute.
Broadcast recordings come through intact. Captions, programme information, the station name, both languages of a bilingual broadcast, interlacing and 2:3 pulldown all survive. The output still looks like a recording, so it opens in whatever you already use for recordings.
A night's cuts can leave as a disc. Not only as files. SmartCut can write a
BDAV folder — the same shape a recorder writes to a BD-RE — with each programme
under the name it went out under, the channel it came off, the night it was
recorded, what the broadcaster said it was about, and a chapter point wherever a
commercial break was taken out. Ask for it and the finished disc is wrapped in a
.iso a burner can take. See Writing a disc.
You can place a cut by the subtitles. The editor's preview will draw them over the picture (off by default): a broadcast's ARIB captions as characters, at the position and colour the broadcaster asked for, and a disc's — PGS and a DVD's subpictures — as the pictures they are. Whether a seam lands in the middle of a line is not something the picture alone will tell you.
Commercial breaks are found for you. SmartCut uses three independent clues: the marks the broadcaster puts in its own subtitle stream at every switch, runs of silence, and whether the station logo is on screen. It places the marks; you decide what to cut.
It handles a whole evening at once. Drop in twenty recordings, press
Ctrl+A then Ctrl+D, and come back later. Reading, detection and editing all
run at the same time, so a batch never stops you from working.
The animation at the top of this page shows two commercial blocks being removed from a 3 minute 45 second recording:
- 133.91 seconds copied bit-for-bit, 0.57 seconds re-encoded
- 17 frames out of 6743 were touched at all
The clip is a practice recording built by
tests/make_demo_media.sh out of ffmpeg's own test
sources: colour cards, a fake station logo and a 15-second commercial grid. No
broadcast material is involved.
Here is the same idea as a diagram. For each range you keep, only the parts that stick out past the key frames have to be rebuilt:
... I ....... I=========================I ....... I ...
^t_in ^k_first ^k_term ^t_out
|<-head->|<--------- body --------->|<-tail->|
re-encode stream copy re-encode
Cut exactly on a key frame and even the head and tail disappear. A 22-minute export of 5 ranges, placed by the automatic commercial detector, came out bit-identical across all 40589 frames.
Builds are on the Releases page.
Every build except the .deb bundles FFmpeg, so there is nothing else to
install.
| Platform | File | Notes |
|---|---|---|
| Linux | SmartCut_0.5.9_amd64.AppImage |
Make it executable and run it |
| Linux | SmartCut-0.5.9-linux-x86_64.tar.gz |
Unpack and run ./smartcut. Use this if you would rather not deal with FUSE |
| Linux (Debian/Ubuntu) | smartcut_0.5.9_amd64.deb |
sudo apt install ./smartcut_0.5.9_amd64.deb. Only 3.4 MB, because it uses the FFmpeg already on your system |
| Windows | SmartCut_0.5.9_x64-setup.exe |
Installer |
| Windows | smartcut-portable-x64-0.5.9.zip |
Unzip and run smartcut.exe |
Requirements. The AppImage and the tar.gz need glibc 2.39 or newer, which
means Ubuntu 24.04, Debian 13, Fedora 40 or later. The .deb needs FFmpeg 7.1,
which means Debian 13 or Ubuntu 25.04 or later; it installs the GUI as
smartcut and the command-line tool as smartcut-cli. The Windows builds are
x64 only and need the WebView2 runtime, which ships with Windows 11 and is
already present on nearly all Windows 10 machines.
To build from source, see Building.
- Add your recordings. Drag them onto the window, or use ＋ Add files. Each row is filled in the moment the file lands, and the preparation runs behind it. You can open a recording before it has finished being read.
- Find the commercials. Press
Ctrl+Ato select everything, thenCtrl+D. SmartCut works through the list and marks the start of every commercial block and every point where the programme resumes. - Cut. Double-click a recording to open the cut editor. The marks are
already in place: click the one at the start of a break and press
I, click the one where the programme resumes, press←thenO, and press ✂ Cut. Press OK when you have finished with that recording. - Check the settings. The output settings tab applies to the whole list: where to write, which format, and what to do with the sound.
- Export. The export tab writes the whole list, top to bottom.
You can save the list at any point with Ctrl+S, and it comes back next time
with your cuts intact. There is a full walkthrough with screenshots in
Using the GUI.
SmartCut opens a disc as it is — as a folder, or as an .iso that never needs
mounting. It reads both halves of the Blu-ray specification: BDAV, what a
recorder writes, and BDMV, what a pressed disc is. It reads DVD-Video
too: the titles, how long each one runs, the chapters the disc set, and the
audio and subtitle tracks it declares.
Drop a disc on the window and SmartCut asks which of the recordings on it you
meant, and which of their tracks to take. That question is worth asking: a
pressed season set holds twelve episodes among fifty logos, warnings and menu
loops, and the disc calls all sixty-two of them 000NN. The ones likely to be
programmes are ticked for you.
Opening one is instant. A disc keeps its own list of the places playback can start from, so SmartCut reads that instead of examining the whole recording: a UHD title of 81 GB and 2 hours 34 minutes went from 8 minutes 45 seconds to under a second. Audio languages come off the same list, so cutting a bilingual disc leaves an output that still says which track is which.
Cuts are written next to the disc under the programme's name. The chapters the disc set are already on the timeline as marks, and on a Japanese recording they are frequently the commercial breaks themselves.
smartcut Anime.iso # what is on it
smartcut Anime.iso --title 2 --cut 8.0-20.0 -o out.tsEncrypted discs are out of scope. See Reading a disc for how a disc is read and what SmartCut does not do with it.
smartcut input.ts --keep 5.3-12.7 -o out.ts # keep this range
smartcut input.ts --cut 8.0-20.0 -o out.ts # drop this range
smartcut input.ts --analyze # show the plan, write nothing
smartcut input.ts --analyze --detect-cm --logo # list the commercial candidates
smartcut input.ts --analyze --scenes # list the scene changes
smartcut input.ts --cut 8.0-20.0 --bdav ~/disc # onto a disc rather than into a file--keep and --cut can be repeated, and accept 1:23:45.6 as well as plain
seconds. Every option is listed in
Using the command line.
Files it can open: .ts .m2ts .mts .m2t .mp4 .mkv .mov .m4v
.vob .mpg .mpeg .m2p, plus discs — Blu-ray (BDAV or BDMV) and
DVD-Video, as a folder or as an unencrypted .iso, read where they lie.
A DVD stores one continuous stream in pieces of about a gigabyte, because the format required it. SmartCut joins the pieces back together and takes the part you asked for out of them, without unpacking or copying anything first. The cut comes out as a transport stream: putting it back into a DVD's own shape would mean writing several kinds of table SmartCut does not write.
Files it can write: MPEG-TS, M2TS, MP4, Matroska, QuickTime — or a BDAV
disc, a folder of recordings with an index a player reads as a list of
programmes, optionally wrapped in a .iso. By default the output uses the same
format and folder as the input.
Video: H.264, HEVC, MPEG-2, MPEG-4 Part 2, VC-1. Interlaced material stays interlaced, and 2:3 pulldown is handled properly rather than flattened.
4K HDR10 cuts too. The few pictures rewritten at a boundary carry the recording's own HDR settings, so the picture does not visibly shift partway through — including the backward-compatible way a 4K broadcast signals HLG, where two parts of the stream say different things on purpose. Dolby Vision is copied intact wherever the pictures are copied; the pictures rewritten at a boundary cannot carry it, and a cut that has to rewrite any says so.
VP9 and AV1 are not supported. They have no form that can be joined end to end, so supporting them would need a different design.
VC-1 is a special case. Most Blu-rays pressed before about 2010 were written in it, and there is no VC-1 encoder anywhere — not in FFmpeg, not on a graphics card. So SmartCut carries its own, purely for the frames at the ends of a range. It writes whole pictures only, which is all a joined fragment needs and only a small part of the format. Those pictures cost more bits than the ones they replace, but a fragment is under a second long, so the trade is worth making. Everything else is copied byte for byte as usual.
Audio: AAC, AC-3, E-AC-3, MP2 and a Blu-ray's linear PCM are all smart rendered. Every track in the file is cut separately, so a bilingual broadcast keeps both languages. 5.1 can be folded down to stereo, and the sound can be written as a different codec entirely — AAC, AC-3, DTS or linear PCM — although that leaves no frame to copy, so the whole track is rebuilt. The same is true of changing the sample rate.
The output settings screen only offers combinations that can actually be written: a rate a codec does not support, or a bitrate below what its frames need, is greyed out there rather than discovered at the end of an export. A disc's lossless audio — DTS-HD and TrueHD — is carried byte for byte and never re-encoded.
Broadcast extras (when writing a .ts): captions are carried across byte
for byte. Programme information, the station name and the broadcast clock are
put back afterwards, and every stream goes back where it arrived. Superimposed
text and data broadcasting cannot be carried on a cut timeline, and SmartCut
says so rather than dropping them quietly.
A disc's subtitles (when writing a .ts or .m2ts): a Blu-ray draws its
subtitles rather than writing them, so what travels is a picture spread over
several packets that mean nothing apart from one another. Those are carried
across whole, and both ends of every kept range are mended: a subtitle that
was already on screen when the range opens is put up again at its first
frame, and one still on screen when the range ends is taken down, instead of
standing there into the next scene. See
the subtitles a disc draws.
Inside the cut, or beside it. A DVD draws its subtitles the same way and a
transport stream has nowhere to put those, so they are converted by
default: written as the kind a Blu-ray carries, the same pixels and the same
colours, so the cut is one file with its subtitles inside. Two other answers
go for either disc, and both put the subtitles next to the cut instead —
.idx and .sub, the VobSub pair every player and subtitle tool already
reads, or .sup, the display sets themselves, which is a Blu-ray's
subtitles byte for byte. Either way both ends of every kept range are mended,
and beside is the one way an .mp4 of a disc keeps its subtitles at all. See
a disc's subtitles, inside the cut or beside it.
One video track per file. See known limitations for the full list.
Your original file is never modified. SmartCut only reads it. The output
goes to a new file, by default in the same folder with cut_ in front of the
name.
Most of the output is provably identical to the input. The copied parts are byte-for-byte the same bytes. That is not an estimate — it is what copying means.
The rebuilt parts are measured, not assumed. The test suite decodes the output and compares it against the source frame by frame. Against real broadcast recordings:
| Material | Result |
|---|---|
| Terrestrial NHK E-Tele (MPEG-2 1440x1080i) | 899/899 frames, 98.2% lossless, interlacing preserved |
| BS11 (MPEG-2 1920x1080i) | 899/899 frames, 98.2% lossless |
| AT-X (MPEG-2 1440x1080, 2:3 pulldown) | 719/719 frames, 99.9% lossless, pulldown pattern preserved |
| A 22-minute commercial cut, 5 ranges | 40589/40589 frames, 100% bit-identical |
| A Blu-ray in VC-1 (1920x1080i animation), 10s between key frames | 308/308 frames, 90% of the video byte-identical, the rest rewritten at 48 dB |
| A Blu-ray in VC-1 (1920x1080p film, heavy grain), same 10s | 246/246 frames, 90% byte-identical, the rest at 45 dB with the grain intact |
The GUI shows you the result before you commit to it. The status line under the timeline is the plan the engine will actually carry out: which ranges are copied, which are rebuilt, and how many frames that is. If it says "Video completely lossless", not one frame will be re-encoded.
"100%" is never rounded up. Two rebuilt frames out of 40000 rounds to 100.0% in ordinary arithmetic, and that is exactly the number a smart renderer must never print. SmartCut shows the frame count instead, and writes 100% only when it means it.
Full results, including the bugs found along the way and the limits inherent in the approach, are in Validation.
The documentation is in two parts. Every page exists in English and Japanese; the switch is at the top of each one.
| User guide How to use it |
Using the GUI ・ Commercial detection ・ Working through a batch ・ Projects ・ Using the command line |
| Technical What it does inside |
Algorithm ・ Validation ・ Audio ・ Broadcast TS ・ Commercial detection internals ・ Reading a disc ・ Writing a disc ・ Rust core ・ Design notes ・ Building ・ Distribution |
If you only read one page, make it the pitfalls: the ten reasons why "just cut on GOP boundaries and concatenate the pieces" does not work, in the order they were hit.
rust/ Rust core (smartcut_core), the VC-1 encoder and the CLI <- the real implementation
gui/ Tauri v2 + vanilla JS GUI
smartcut/ Python reference implementation <- test oracle
tests/ 20 end-to-end suites, 352 checks
docs/ Documentation
The Python implementation is kept as a reference implementation and test oracle.
It is what pinned down the algorithm and its pitfalls in the first place. It
shares the same frame-hash verification as the Rust core, and
tests/run_tests.sh and tests/run_rust_tests.sh report identical lossless
ratios.
x264 and x265 are GPL, and linking against them makes the whole application GPL. Re-encoding can also be switched to a hardware encoder (NVENC, QSV, VideoToolbox, AMF). If you intend to distribute commercially, patent licensing for H.264 and HEVC needs separate consideration.