Introduction

͏    To extract specific part of input: seeking has to be performed on which first.
͏    ͏"-ss" can be used to seek within the input several ways.

Input seeking

Specify "-ss" before "-i":

ffmpeg -ss 00:23:00 -i "Mononoke.Hime.mkv" -frames:v 1 "out1.jpg"

͏    The demo produces 1 image frame at 23 min from the beginning of the movie.
͏    The input will be parsed by keyframe, which is very fast.

͏    As of FFmpeg 2.1, when transcoding with `ffmpeg` (i.e. not stream copying): "-ss" is also "frame-accurate" even as input option.
͏    Previous behavior (seek only to nearest preceding keyframe, despite inaccuracy) can be restored with "-noaccurate_seek".

͏    There's no general rule on setting times for inaccurate seeking appropriately:
͏    For that depends on the keyframe interval used when encoding.

͏    As reference, "libx264" by default uses GOP (Group of Pictures) size of 250:
͏    Which means 1 keyframe every 10 s if 25 FPS.

Output seeking

Specify "-ss" after "-i":

ffmpeg -i "Mononoke.Hime.mkv" -ss 23:00 -frames:v 1 "out2.jpg"

͏    The demo also produces 1 image frame precisely at 23 min from the beginning of the movie.

͏    Here, the input is decoded (and discarded) until it reaches the position indicated by "-ss".
͏    This will be done relatively slow, frame-by-frame.

͏    The main advantage is that when applying filters to output stream: the timestamps aren't reset prior to filtering.
͏    E.g. when burning subtitles onto video, the subtitle timestamps don't have to be modified.
͏    But the drawback is that may take long for excessive decoding: bigger the offset, longer the time.

Combined seeking

Specify "-ss" both before/after "-i":

ffmpeg -ss 22:30 -i "Mononoke.Hime.mkv" -ss 30 -frames:v 1 "out3.jpg"

͏    As of FFmpeg 2.1, combined seeking is still possible.
͏    But I (c-14) have yet to find which a valid use case: for "-ss" as input option shall be both fast and accurate.

͏    The demo seeks via keyframe to ~ 22:30, then accurately to 23:00 (22:30 + 30).
͏    Semantically valid, but mostly useless.

Notes

Cutting small sections

͏    To extract a small segment amid a movie, it can be used in combination with "-t":
͏    Which specifies the duration, like "-ss 60 -t 10" to capture from 60 s to 70 s.
͏    Or use "-to" to specify an out point: like "-ss 60 -to 70" to capture the same.
͏    "-t", "-to" are mutually exclusive: when both specified, "-t" takes precedence.

͏    Note when "-ss" used before "-i" only: ("-ss" as input option) the timestamps will be reset according to 0.
͏    So "-t" and "-to" shall be the same.
͏    To keep original timestamps unmodified: add "-copyts".
͏    E.g.
͏    (outdated, see # Input seeking)

ffmpeg -ss 1:00 -i "video.mp4" -to 2:00 -c copy "cut.mp4"
ffmpeg -ss 1:00 -i "video.mp4" -to 2:00 -c copy -copyts "cut.mp4"
ffmpeg -i "video.mp4" -ss 1:00 -to 2:00 -c copy "cut.mp4"

͏    #1 cut from 1:00 to 3:00 (of original), using faster (inaccurate) seek.
͏    #2 cut from 1:00 to 2:00 intended, using faster (inaccurate) seek.
͏    #3 cut from 1:00 to 2:00 intended, using slower (accurate) seek.
͏    [ #1, #2 are essentially Combined seeking: now mostly meaningless. ]

͏    Enable ͏"-avoid_negative_ts" when "-c copy" alike (stream copy): if intended to reuse with ͏"concat" demuxer.
͏    E.g.

ffmpeg -ss 3:00 -i "video.mp4" -t 60 -c copy -avoid_negative_ts 2 "cut.mp4"

[ "-avoid_negative_ts"
͏    int, output
͏    Shift timestamps so they start at 0.
͏    Default: -1

͏    |-1| "auto": Enable when required by target format.
͏    |0| "disabled": Shift no timestamp. [ Doesn't mean "-copyts". ]
͏    |1| "make_non_negative": [
͏    Shift to make timestamps non-negative. (based on the 1st timestamp)
͏    Affects not following non-monotonic timestamps. ]
͏    .
͏    |2| "make_zero": Shift to make timestamps start at 0.

͏    When enabled, all output timestamps are shifted by equal amount.
͏    Stream desync and relative timestamp differences are preserved as before shifting.

͏    "-copyts" takes precedence of "-avoid_negative_ts" when both specified. ]


͏    [ Unclear Unverified:
​https://trac.ffmpeg.org/wiki/Seeking?action=diff&version=24
͏    If have to re-encode anyway, e.g. to apply filters like ͏"afade" that can be very slow.
͏    Make sure to use `-ss 120 -i "some.mov" -to 60` alike to get the 1 min: from 120 s to (( 120 + 60 )) s.
͏    Not "-to 180" for the 3 min starting at 120 s. ]

Time unit syntax

͏    2 formats supported:
͏    |*| Sexagesimal ("Hour:MM:SS.Millisecond", as in "01:23:45.678")
͏    |*| Seconds

͏    If the fraction part presents (e.g. "02:30.05"): it's interpreted as "2 m 30.05 s", not the 6th frame of 02:30. (per SMPTE timecode)
͏    Equivalent as "150.05" in seconds.

͏    Example of SMPTE timecode:
͏    ​https://trac.ffmpeg.org/attachment/ticket/11055/Ticket_11055.m2ts.xml

Seeking while codec copy

͏    Using "-ss" with "-c copy" alike may not be accurate: since `ffmpeg` may only split on I-frame (keyframe independently decodable) alike.
͏    Though it may, if applicable: auto-adjust the stream's start time to negative to compensate.

͏    E.g. (with typical video) requested timestamp 157 s; but no keyframe until 159 s:
͏    It shall include ~ 2 s audio (no video) at the start, and start from the 1st keyframe.
͏    [ ^ ​https://trac.ffmpeg.org/ticket/11018#comment:2 ]

͏    So be careful when splitting and doing codec copy.