cli.md

Using the command line

← Documentation ・ ← SmartCut ・ 日本語

The same engine that runs inside the GUI is also available as a command. The packages install it as smartcut-cli — smartcut-cli.exe on Windows, ./smartcut-cli in the unpacked tar.gz — and smartcut is the GUI. Built from source, the command is called smartcut, which is what the examples on this page use.

It is for repeating the same job, for calling from a script, and for machines with no screen to put a window on. It can do everything the GUI can.

Start with these six

smartcut input.ts --keep 5.3-12.7 -o out.ts   # keep only this range
smartcut input.ts --cut 8.0-20.0  -o out.ts   # drop this range
smartcut input.ts --analyze                   # just show what would happen

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 instead of a file

• --keep and --cut can be given as many times as you like. --keep 0-60 --keep 180-240 keeps those two ranges.
• Times can be plain seconds or the 1:23:45.6 form.
• --analyze writes no cut. It prints the plan — what gets copied and what gets rebuilt. Running it first is the safe way to check before committing. (What other options write is still written: a --seek-index file, say.)
• smartcut --help lists every option, grouped by what it is for.
• smartcut --version prints the version as smartcut 1.2.3 and nothing else, for a script that has to know which one it is running.

Choosing where to cut

Option	Meaning
--keep START-END	A range to keep. Repeatable
--cut START-END	A range to drop. Repeatable. Not with --keep
--no-open-gop	Never start a copy at an open GOP
--clean-joins	Re-encode up to two extra seconds at the start of each range, far enough to reach a point where no picture comes out of the decoder in the wrong order. Off by default — see the note below

What --clean-joins costs. At an ordinary join, the pictures a copy makes are already exact; the only thing wrong is the order one of them comes out in. --clean-joins avoids that by rebuilding two seconds, and those two seconds stop being a copy of the recording (about 51 dB against the original). That trades something certain for something occasional, which is why it is off.

Choosing where the output goes

Option	Meaning
-o OUTPUT (or --output)	Output path. The extension picks the container. It has to be a file on this machine, and not the recording being read (or a link to it): either is refused before anything is written. On Linux a share named the Windows way or as a URL (\\nas\rec\x.ts, smb://nas/rec/x.ts) is written through the place it is mounted, as an input named that way is read; a share that is not mounted is refused. An extension that holds sound alone (.aac, .ac3, .eac3, .mp2, .mp3, .dts, .wav, .aiff, .aif, .aifc, .flac, .opus, .thd, .caf, .w64, .oga, .wv, .spx, .au, .mpa, .tta, .amr, .gsm, .mlp, .sbc) is accepted only with --sound-only. Given twice, -o is refused: one run writes one file. Nor can it be one of the files read by a recording named the way FFmpeg names several files or a piece of one (concat:a.ts|b.ts, subfile,,start,…,end,…,,:x.iso, file:x.ts). A run that fails after it has begun writing over a file already under that name removes what it left there, and says so; a run stopped before it wrote anything leaves the file as it was
--drop-stream INDEX	Leave one of the recording's streams out of the output. Repeatable. The same thing the cut editor's Tracks menu does. The index is a sound, caption or subtitle stream's; one the recording does not have, or the pictures', is an error. A broadcast's sound track with nothing in the ranges kept (a second track the programme before this one had, say) is left out without being asked, and the run says so
--title N	Which recording on a disc (a folder or an .iso) to open. Part of the programme's name works in place of the number. Left out, it lists what is on the disc and stops. If the run also asked for something that needs a recording open (a cut, an output, --analyze, a --seek-index, --as-proxy and the like), the list is followed by an error. A recorder's title cut with no --cut or --keep keeps what the disc's playlist plays. The clip a recorder writes holds more than that: a second or so of the programme before, the moment either side of a break the recorder left out, and the start of the programme after. Those stretches are cut, on the pictures, and the run names them on a plays : line. No chapter is put where they were. A pressed disc, a DVD and a disc made with an authoring tool play the whole clip, and are cut as before

Joining several recordings

Option	Meaning
--join RECORDING	Another recording, written into the same output after this one. Repeatable, and they go in the order they are named. Where the master's sound is TrueHD or DTS, a recording whose sound is another codec is refused before anything is written; make it the master with --master, or name a codec with --audio-codec. A disc, a folder or an .iso, is refused at once, before the first recording is read: name one of its recordings by the path smartcut DISC --title N --analyze prints on its input : line. The refusal gives the first recording's as an example — DISC/BDAV/STREAM/00001.m2ts on a recorder's disc, DISC/VIDEO_TS/VTS_01_1.VOB@…-… with its stretch on a DVD; VTS_01_1.VOB alone would open only the first gigabyte of the title set, not a title. A folder that holds no disc is refused at once as well, and so is an .iso that holds no disc that can be read: it is not a recording either
--master N	Which of them the output takes its shape from — frame size, rate, codec, sound tracks, tables. Counting the first as 1. Left out, it is the first. A number past the last recording is refused

--keep and --cut belong to the first recording; a joined one is written whole. A command line giving every file its own ranges would be a project file with a worse syntax, and the window is where a list with cuts in it belongs.

smartcut part1.ts --join part2.ts --join part3.ts -o whole.ts

A recording that is not the master's shape is decoded and written afresh to fit it — scaled, rate-converted, re-encoded — and the run says which and why before it writes a byte. Everything that matches is smart-rendered as usual.

Transitions at the joins

Option	Meaning
--transition KIND	none, fade-black, fade-white, dissolve, wipe-left|right|top|bottom, slide-left|right|top|bottom. Applied at every join between two clips
--transition-seconds S	How long it runs. Up to 30; 1 when left out
--transition-easing C[:M]	The curve and the end it is applied at: none, back, bounce, circle, elastic, exponential, power, sine, quadratic, cubic, quartic, quintic, each with :in, :out, :in-out or :out-in. Default none:in, which is a straight line. A name not on the list is refused
--transition-image FILE	A still laid over each crossing, coming up and going down with it

All four are about the joins, and so are --join-fade-out and --join-fade-in: given without a --join, they stop the run rather than being ignored.

smartcut a.ts --join b.ts --transition dissolve --transition-seconds 2 \
  --transition-easing sine:in-out -o joined.ts

A fade keeps the output's length — it takes half its time from each side. Everything else takes its own seconds off the output, because both clips are on screen at once for the whole of it.

Every frame a transition covers is written afresh: it is pictures that are in neither recording. Two seconds is two seconds of encoding per join and nothing anywhere else. The sound is not mixed. Where the two clips are on screen together, the clip before plays through the crossing and the clip after starts where it ends; through a fade, the sound changes over at the colour, in the middle. --audio-fade is what softens the change if it needs softening.

The captions change over where the pictures do. Through a dissolve, a wipe or a slide, a broadcast's captions and a 4K recording's subtitles are the clip after's from the crossing's first frame: the line it has up there comes up at once, what it says inside the crossing comes at its own time, and the clip before's line goes when the clip after's first one comes. Where the clip after says nothing inside the crossing, the clip before's lines run on as they would. Through a fade the two follow one another, as the pictures do. Up to 0.8.35 the clip after's lines waited for the crossing to end, and one said and gone inside it was lost. A disc's subtitles, which only the master carries, do the same where the master is the clip after (up to 0.8.36 they waited for the crossing to end as well). See The captions at a crossing are the clip after's.

Audio

By default the sound is treated like the video: only the few frames a join falls inside are rebuilt (smart rendering). Give any of the five settings below a value the recording does not already have and there is no frame left to copy, so the whole track is re-encoded.

Before it starts, the run prints a render: line saying what is done with the sound (audio smart, or audio reencode and how), on the main track -- or, where --drop-stream takes that away and always with --sound-only, the first track kept, which is the one a file of sound alone is written from. It says what the cut will really do, not what was asked: a width asked of AAC, or the track's own codec named with --audio-codec, leaves it smart; a DTS or TrueHD track, which is carried through whole unless another codec is named, is not shown as downmixed or resampled, and asked to be re-encoded with no other codec named it says copy, which is what it gets; a Blu-ray's or a DVD's LPCM going into a .mkv or an .mp4, which hold it only as plain PCM, says reencode and what it is written as (as pcm_s16be, for one); --audio-codec lpcm on a Blu-ray's own LPCM into a .ts or .m2ts stays smart; a .wav of sound alone says it is re-encoded as linear PCM; and FLAC written alone into a .flac, .ogg or .oga says reencode, since such a file opens on a header that copied frames would contradict.

Option	Meaning
--audio-mode smart|copy|reencode	smart is the default: it rebuilds only the frames a boundary falls inside, so nothing from the far side of a cut is heard — and nothing at all when the join falls in silence. copy is lossless to the byte but leaves a little of the far side audible; reencode cuts to the sample at the cost of rebuilding the whole track
--audio-codec source|aac|lpcm|ac3|dts	What the audio is written as. source, the default, keeps the recording's own. AAC plays straight off a phone, AC-3 and DTS are what an AV receiver handles most reliably, and linear PCM is uncompressed
--audio-channels N	Channel count, 1 to 8. Anything but the recording's own is a downmix — 5.1 folded to stereo, for instance
--audio-samplerate RATE	Sample rate, as 48k or 48000. Codecs differ in what they support (AC-3 has three rates, Blu-ray LPCM three others); a rate a codec does not have is taken to the nearest one it does, with a note saying which
--audio-bits 16|24	How wide each sample is. Only meaningful for linear PCM — the other codecs write a description of the sound rather than the sound itself, so there is nowhere to put a width. A recording's own plain PCM written into a .mkv or .mov is written at the width asked for, float and 8-bit samples included. The signed 8-bit PCM an old .mov can carry goes into a .mkv or an .mp4 only as 24 bit; otherwise the run is refused before anything is written, and the message says so
--audio-bitrate RATE	Bits per second when re-encoding, as 192k or 192000. Left out, it follows the recording. A figure the encoder will not accept is raised to what that codec is ordinarily carried at, with a note saying so
--aac auto|mpeg2|mpeg4	Which flavour of AAC the frames SmartCut writes announce themselves as. auto, the default, follows the recording — MPEG-2 AAC for a broadcast
--audio-es	Also write the sound out as a bare stream beside the output. AAC only
--poster TIME	Carry the picture at TIME into the recording as the file's cover (thumbnail): covr in .mp4 and .m4v, an attachment named cover.jpg in .mkv. Other containers have no place for one (FFmpeg writes no covr into a .mov), and the run says it is ignored. A time past the end takes the last picture, with a note. An interlaced recording's cover is taken from one field. Not with --sound-only
--no-chapters	Leave the chapter list out. Otherwise an .mp4, .m4v, .mov, .mkv or .webm gets one chapter where each kept range begins (and where each --join recording begins), plus the chapters the recording came with that survive the cut — the disc's own for a recording off one, otherwise the chapter list an .mp4 or .mkv input already has, and each --join recording's own list, moved to where that recording lands — the same points --bdav writes into a playlist, which this does not change. A cut with only one range and nothing else gets no chapter list. Not with --sound-only
--sound-only	Write the sound and no pictures: the ranges, the joins and the fades exactly as they would be inside the video, and nothing read or written for the frames. -o names the file and its extension picks the container — .aac, .ac3, .eac3, .mp2, .mp3, .dts, .m4a, .mka, or .wav, which is written as linear PCM unless --audio-codec names another codec (a recording's 32-bit float sound stays 32-bit float in a .wav or .aiff where --audio-bits is not given). .aiff also takes linear PCM, and .flac, .opus and .thd take a recording's own FLAC, Opus or TrueHD. A sound the container has no room for stops the run before anything is written. A 4K broadcast's LATM-framed AAC asked for as .aac or .m4a is re-encoded as plain AAC, and the run says so; .mka keeps its own frames. One sound track, taken from the recording --master names, and not with --bdav, --video-share or --audio-es, nor with what only a file with pictures has: --tables, --data-broadcast, --no-data-broadcast, --subtitles, --drop-subpicture, --vc1-quant, --clean-joins, --no-open-gop, --transition-image or --no-chapters. A recording joined on with no sound is left out of the file, and the run says so. On a recorder's title, the time at a seam that no picture is written for is left out of the sound too, so the file is as long as the video would be. With several ranges, each one lines up within a frame with the sound of the video cut of the same ranges
--audio-fade SECONDS	Take the level down into each seam and bring it back out over that many seconds. 0 to 10; 0, no fade, is the default
--join-fade-out SECONDS	How long the sound takes to leave at the end of each clip, where --join writes several into one file. 0 to 10; 0 is the default
--join-fade-in SECONDS	...and how long it takes to come back at the start of the next one

A join between two clips is asked for at each end. How the programme that is ending should end and how the one that is starting should start are two questions: --join-fade-out 3 --join-fade-in 0 takes three seconds to leave and starts the next one at full level. The seams inside one recording take --audio-fade, which is one answer for the run.

What these reach. Only sound this program writes: smart and reencode fade, copy does not, and neither does sound carried through whole because re-encoding it would lose what makes it lossless (TrueHD, DTS-HD MA). Where it cannot be applied the cut says so. The beginning and the end of the output are left alone: a fade is for a place where two pieces meet.

Writing a broadcast .ts

Option	Meaning
--tables partial|broadcast|muxer	How a transport stream describes its own contents. Unsaid, a .ts gets broadcast — the recording's own SDT, EIT and TOT, which is where a player reads the programme name, the station and the clock — and a .m2ts gets partial, the shape a disc's stream is written in. muxer adds nothing. --no-tables is the old name for muxer. With --bdav only partial is taken: a disc's stream is written in that shape, and muxer or broadcast stops the run before anything is read
--no-data-broadcast	Leave out the recording's data broadcast — what is behind the d button. It is otherwise carried into any .ts whose tables are not left to the muxer (broadcast or partial); a .m2ts has nowhere to put one: the modules come out whole and byte for byte, so a receiver draws the same pages. What turning it down buys is size — a carousel is between a hundredth and a fifth of what a broadcast multiplex spends. --data-broadcast asks for it outright, which only changes what is said when it cannot be carried; given together with --no-data-broadcast, the run is refused. With --join, every recording's carousel goes in over its own stretch and is listed as data : carousel among that recording's lines; one whose data component is described otherwise than the same component already in the output (another station's profile or BML version), or whose entry component cannot be announced, is left out whole with a note and listed as not carried; this takes a master that is itself a transport stream, since the tables, the programme guide included, are put back from that recording

Writing a disc (BDAV)

--bdav writes the same shape a recorder writes to a BD-RE — a folder, not a file. The disc decides the filenames, so -o is not used, and giving both is an error. If a disc is already there, the recording is added to it. --iso and everything below it in the table describe a disc, and without --bdav they are refused rather than quietly ignored.

Option	Meaning
--bdav FOLDER	Write a BDAV disc into this folder (one recording becomes BDAV/STREAM/00001.m2ts). The index is built afterwards. Pictures or a kept sound track that a disc's index has no coding type for (Opus, for one) stop the run before anything is cut or a place on the disc is taken; --audio-codec or --drop-stream gets past the sound. A run that stops before the recording is written removes the stream file it set aside on the disc, and takes away the folders it made for it (BDAV and the ones in it, and the folder itself and its parents where the run made them) as long as they are still empty. So does a run whose recording was written but could not be put in the disc's index: that recording is taken off the disc, and the folders go with it the same way, while a disc that was there before keeps its info.bdav as it was. None is taken away while another run's recording is in that disc's STREAM folder, and a folder that was there before stays, empty or not, and so does every folder where the path given goes through ... The folder has to be on this machine, as -o does: one named as an address (http://…) is refused at once, before the recording is read, and by --analyze too. Given twice, --bdav is refused, as -o is. On Windows a drive's own root (E:\) is written into as named, not into the folder last used on that drive
--fit bd25|bd50|bd100|bd128|BYTES	Write the pictures back smaller, by as much as it takes for the output to fit that much room. Where it already fits, nothing is done. The sound, the subtitles and the programme information are untouched; only the pictures give anything up. The sound is counted as it will be written, so --audio-codec lpcm or a track left out with --drop-stream is in the sum. MPEG-2 only -- anything else is written at its full size and says so, and where that does not fit the run stops. What share they will be written at is printed before the writing starts. The rewrite does not always get there -- short ranges at a hard share can stay several points above it -- and where it stops short, a note: says how far, and how many MB more than planned the pictures it rewrote came to (the cut as a whole may still fit). Once the run is over, the output (with --bdav, the whole disc, what was on it already included; with a read-only --iso, the image itself) is weighed against the size --fit priced against (with --iso-access overwritable, what the recorder's layout holds), and where it came out larger a note: says by how much and why -- where the pictures were written smaller, what the run was priced at with them at that share and how far past that it came (the recording weighed more than its index measured, or the pictures would not shrink as far as asked, which a note: above says), or the run was priced to fit as it was, or, for an image whose files alone are under the size, the image's own file system, which --fit does not price; the output is left as it is. Each range kept is priced from the bytes the recording holds there, read off its index, rather than from the recording's average, the way the window's disc gauge prices it. On a disc the stream's own tables are priced by the second, since they go out at a fixed rate however small the pictures become; an AAC track in a transport stream (.ts, .m2ts, a disc) whose stated rate is less than half of what AAC usually takes at that many channels is priced at that usual figure, since there the rate is not recorded and what is stated is the average over the opening libav reads, which a quiet opening makes far too low -- a track really carried that low is priced high, and the pictures come out a little smaller than they needed to. A rate an MP4 or Matroska file states is believed however low. With --bdav naming a folder that already holds a disc, what is already written there is measured and the output is fitted into the room left; the fit line says how much that is. With --iso-access overwritable the output is fitted into what a recorder's layout holds of files -- 24.12 GB on a single layer, 48.34 GB on a dual -- and the fit line says so too. Not with --join or --sound-only. See Fitting a disc
--video-share 0.35..1	The share itself, where you would rather name it than have a size worked out into one. Not with --fit, which works one out. MPEG-2 only, as with --fit: other pictures are copied at their full size, and the run says so. The pictures re-encoded at the ends of each range follow the ones around them -- written at the share where those are written smaller, at the full rate where they are copied. Up to 0.8.35 an H.264 cut's seams were re-encoded at the share all the same, the only pictures in it that came out worse
--iso 2.50|2.60	Wrap the finished disc in an .iso beside it: --bdav ~/disc writes ~/disc/BDAV and ~/disc.iso. The image holds the disc, BDAV, and nothing else in the folder. The folder stays. Refused before anything is written where that .iso is the image the recording is being read out of. An image already under that name that is not of this disc (another disc's, or one of this folder from before it was taken away and started again) is refused before anything is cut, and by --analyze too; an image of this disc as it was before the run is replaced
--iso-access read-only|overwritable	What the image says may be done to the disc it is burned onto. read-only, the default, is the truth about a disc nothing will write to again — a BD-R, or a BD-RE you only play. overwritable is what a recorder writes on a BD-RE, and what it wants to see before it will add a recording to the disc or take one off; the image is then laid out the way a recorder lays out a BD-RE, and is the size of the whole disc (25 GB, or 50 GB where one layer will not hold it — the part with nothing in it takes no room on the disk it is written to). An image that outgrows one layer and comes out the dual-layer size says so. Needs --iso
--iso-only	And then take the folder away, leaving the image on its own. Needs --iso: the image is made of the folder, so the folder is written first and goes once the image holds it. What goes is BDAV, and the folder above it only where that leaves it empty — a disc written into a folder of your own leaves everything else in it alone. A file in BDAV that the image could not carry keeps the folder where it is. Where a .iso of that name is already there and the --bdav folder holds no disc to add to, the run is refused before anything is cut (exit code 1), since that image would be replaced by the new disc's; move it away to keep it, or give --bdav another folder. Where that .iso is the image the recording is read out of, the run is refused as that instead, as --iso says
--disc-title NAME	What the disc is called. Left out, a disc that already has a name keeps it, and one with no name — a recorder's disc nobody named — stays without one; a new disc is named after the series the recording is an episode of: the programme name with the episode number, the episode's own title and the broadcast's marks taken off it
--programme NAME	What this recording is called in the disc's index. Left out, the name its playlist gave it if it came off a disc, otherwise the programme name the broadcast carries
--channel NAME[,N]	The channel, and optionally the three digits a viewer knows it by (--channel "衛星第一,161"). Left out, both come from the recording. A number that is not one is refused
--about TEXT	What the disc's index says the programme was. Left out, the description the recording carries
--made "Y-M-D H:M:S"	When the recording was made. Left out, when the programme went out, where the recording still says

Looking before you write

Option	Meaning
--analyze	Work out the plan and print it. Writes no cut — not even into a --bdav folder. A --seek-index file, a --proxy or a --preview asked for alongside is still written. What the run would refuse before cutting it refuses as well: an --iso that is the image the recording is read out of, an --iso-only image in the way, an -o that is not a file on this machine (http://…) or a --bdav that is not a folder on it, an -o that is a recording of the run or the --transition-image, the .aac that --audio-es would write beside -o where that is a recording of the run, and — only where -o or --bdav says the run would write — a --drop-stream number or a --drop-subpicture id that names nothing. Those two are checked against the first recording, since --analyze does not open the --join ones: not at all with --master past the first for --drop-stream, or with any --join for --drop-subpicture
--detect-cm	Look for the commercial breaks
--logo	Let commercial detection use the station logo as well. Given with --detect-cm
--inserts	...and report the few seconds a subscription channel drops into a programme — its own animated ident. Off by default: the same test catches a programme's own caption card. Given with --detect-cm. See the commercial detection guide
--scenes	List the scene changes. Writes no cut: with -o or --bdav the run is refused, as --detect-cm is
--preview TIME	Decode one picture at TIME and write it as a JPEG (preview.jpg, or -o, which has to end in .jpg or .jpeg or have no ending at all; a .png or .bmp is what --still writes). It prints the time actually decoded beside the time asked for
--still TIME	Save the frame at TIME at the size the recording is shown at, as the PNG, JPEG or BMP that -o names (still.png without one). An interlaced recording is saved from one field. A time past the end is an error, as for --cut-near. The same as Save this frame as an image in the cut editor
--cut-near TIME	Print where the nearest picture-to-picture change is to TIME, in windows of ±0.5, ±1 and ±2 seconds. A time past the end of the recording is an error (exit code 1); the length the input line prints is the end, and typed back it is answered. On a recorder's title, a change found at a seam is the first picture after it
--verify	Once the cut is written, decode it from end to end and line it up against the recording frame by frame: the frame count, that every copied frame is the recording's own bit for bit, the re-encoded frames' quality, decode errors and the length of the sound. How long the pictures run and how much time is left without one are held to what the recordings' own pictures hold in the kept ranges (the length line, and a gaps line where the output has more of that time than they do), and each sound track to what the recording's own packets of it hold there, so a track the recording itself lacks for a while is not called short. Each finding is a line starting verify:. A file whose pictures do not match ends the run with exit code 1; sound that runs long or short, or is laid out more than half a second past the last picture, is only reported. Not with anything that writes no cut, such as --analyze, and refused when there is neither -o nor --bdav. It is the same check as the GUI's
--log FILE	Also write what the run prints into FILE, headed by a line giving the version and the command line. An existing file is added to. FFmpeg's lines go in as SMARTCUT_FFMPEG_LOG lets them. A file the run reads or writes — the recording, -o, the -o name with .si added that a .ts cut is written under while its broadcast tables are put back, the .aac of --audio-es, the subtitle files written beside the cut (.idx and .sub, .sup), the proxy --proxy writes and the .marks it writes beside the proxy, and so on — is refused, and so is a file inside the folders of a disc it reads or adds to (the BDMV, BDAV or VIDEO_TS of a disc opened by its folder, the BDMV or BDAV holding a disc clip given as the input or to --join, also when named from inside that folder as STREAM/00001.m2ts or from inside STREAM as a bare 00001.m2ts, the VIDEO_TS of a DVD title given by its stretch as VTS_01_1.VOB@… the same way, the folder of a title set's piece given as a plain VTS_01_1.VOB or VIDEO_TS.VOB where that folder holds VIDEO_TS.IFO (and where the name is a link, the folder it points into as well, whatever the link is called), and the --bdav target's BDAV). A disc named as the folder the command was typed in is given by its full path in the refusal. With --iso, the image beside the disc and the .part it is written under are refused too. A refused log takes away the folders the run made for it, and only those. A log refused only once the recording is open (the names that depend on what is in it) is removed again where this run created it; where the file was already there, it is cut back to the length it had, so the lines the run had added to it go too

Index and proxy

Option	Meaning
--seek-index PATH	Where to keep the seek index. Written on the first run and read on the next, which saves walking the recording again. One that cannot be read — another version's, or a damaged one — is made again, and so is one made from another recording, or from this one while it was still being written, with a note saying so. A file that is not a seek index at all is left untouched and the run stops, so a mistyped path cannot write over a recording. Nor can it be a file the cut writes beside -o: the .aac of --audio-es, the -o name with .si added, or a subtitle file written beside the cut
--index auto|disc|scan|container	Where the entry points come from. auto, the default, asks whoever already knows and falls back: the disc's own map, then the container's seek table, then the walk over the packets. disc and container ask one of those and stop; scan always walks, which is what a transport stream needs — it has neither table. A container table is taken only where it covers the whole recording. disc, container and scan hold for every --join recording too: one that cannot be read that way stops the run with an error naming it, where up to 0.8.29 it fell back on the other table or the walk
--proxy	Build the editing proxy (a light stand-in file) beside the recording and stop (.proxy.mp4, or -o). Its marks are written beside it under the proxy's name with .marks in place of its extension, so an -o ending in .marks is refused
--as-proxy	Treat the input as a proxy rather than a recording — how a proxy is checked against what it stands in for

A disc's subtitles

Option	Meaning
--subtitles pgs|beside|sup	Where the subtitles a disc draws go. pgs, the default, puts them inside the cut as the kind a transport stream carries — a Blu-ray's own untouched, a DVD's converted to it. beside writes them next to the cut as the .idx and .sub pair every player and subtitle tool reads — a DVD's own untouched, a Blu-ray's read back out of its display sets. sup writes those display sets themselves, into a .sup next to the cut — a Blu-ray's subtitles byte for byte, and what BDSup2Sub and Subtitle Edit read. beside and sup are refused with --bdav: a file next to the clip does not go onto the disc
--drop-subpicture ID	Leave one of a DVD's subtitle streams out, by the id the disc knows it by: one byte in hex, 0x20 or 20. Repeatable

Only a .ts or an .m2ts can hold subtitles inside it. A DVD's subtitles cut to an .mp4 or an .mkv are written as the pair beside it instead, and the run says so. A Blu-ray's are not: cut to anything but a transport stream with the default pgs, they are left out, and the run says so and names the two ways to keep them, a .ts or .m2ts, or beside or sup, which write them next to any output. A .sup holds one stream, so a recording with two of them is written as two files, named by language (cut_title.eng.sup), or by number where the language does not tell them apart. See a disc's subtitles.

Some kinds of subtitles are not carried into any output at all: SubRip, ASS, mov_text or WebVTT in an .mkv or an .mp4, a DVB broadcast's subtitles, and a DVD's subtitles remuxed into an .mkv. The run names each of them on a line of its own, not carried: a subtitle stream (its format is not one a cut carries). A --join recording's are named as well, each under the join : line that names its recording.

VC-1 discs

Option	Meaning
--vc1-quant 3..31	How finely the join pictures of VC-1 material are written; 3 is the finest. The default is 4, which lands around 46 dB against the pictures it replaces. On material that uses overlap smoothing, anything coarser than 8 is written at 8

This setting exists only for VC-1 because there is no VC-1 encoder anywhere — not in FFmpeg, not on a graphics card. SmartCut writes those join pictures itself, and this is how finely. The details are in the Rust core.