Native SponsorBlock implementation and related improvements (#360)

SponsorBlock options:
* The fetched sponsor sections are written to infojson
* `--sponsorblock-remove` removes specified chapters from file
* `--sponsorblock-mark` marks the specified sponsor sections as chapters
* `--sponsorblock-chapter-title` to specify sponsor chapter template
* `--sponsorblock-api` to use a different API

Related improvements:
* Split `--embed-chapters` from `--embed-metadata`
* Add `--remove-chapters` to remove arbitrary chapters
* Add `--force-keyframes-at-cuts` for more accurate cuts when removing and splitting chapters

Deprecates all `--sponskrub` options

Authored by: nihil-admirari, pukkandan

yt_dlp/__init__.py

@@ -13,7 +13,6 @@ import random
 import re
 import sys
 
-
 from .options import (
     parseOpts,
 )
@@ -307,6 +306,7 @@ def _real_main(argv=None):
     opts.forceprint = opts.forceprint or []
     for tmpl in opts.forceprint or []:
         validate_outtmpl(tmpl, 'print template')
+    validate_outtmpl(opts.sponsorblock_chapter_title, 'SponsorBlock chapter title')
 
     if opts.extractaudio and not opts.keepvideo and opts.format is None:
         opts.format = 'bestaudio/best'
@@ -353,15 +353,34 @@ def _real_main(argv=None):
     if opts.getcomments and not printing_json:
         opts.writeinfojson = True
 
+    if opts.no_sponsorblock:
+        opts.sponsorblock_mark = set()
+        opts.sponsorblock_remove = set()
+    sponsorblock_query = opts.sponsorblock_mark | opts.sponsorblock_remove
+
+    if (opts.addmetadata or opts.sponsorblock_mark) and opts.addchapters is None:
+        opts.addchapters = True
+    opts.remove_chapters = opts.remove_chapters or []
+
     def report_conflict(arg1, arg2):
         warnings.append('%s is ignored since %s was given' % (arg2, arg1))
 
+    if (opts.remove_chapters or sponsorblock_query) and opts.sponskrub is not False:
+        if opts.sponskrub:
+            if opts.remove_chapters:
+                report_conflict('--remove-chapters', '--sponskrub')
+            if opts.sponsorblock_mark:
+                report_conflict('--sponsorblock-mark', '--sponskrub')
+            if opts.sponsorblock_remove:
+                report_conflict('--sponsorblock-remove', '--sponskrub')
+        opts.sponskrub = False
+    if opts.sponskrub_cut and opts.split_chapters and opts.sponskrub is not False:
+        report_conflict('--split-chapter', '--sponskrub-cut')
+        opts.sponskrub_cut = False
+
     if opts.remuxvideo and opts.recodevideo:
         report_conflict('--recode-video', '--remux-video')
         opts.remuxvideo = False
-    if opts.sponskrub_cut and opts.split_chapters and opts.sponskrub is not False:
-        report_conflict('--split-chapter', '--sponskrub-cut')
-        opts.sponskrub_cut = False
 
     if opts.allow_unplayable_formats:
         if opts.extractaudio:
@@ -388,12 +407,26 @@ def _real_main(argv=None):
         if opts.fixup and opts.fixup.lower() not in ('never', 'ignore'):
             report_conflict('--allow-unplayable-formats', '--fixup')
         opts.fixup = 'never'
+        if opts.remove_chapters:
+            report_conflict('--allow-unplayable-formats', '--remove-chapters')
+            opts.remove_chapters = []
+        if opts.sponsorblock_remove:
+            report_conflict('--allow-unplayable-formats', '--sponsorblock-remove')
+            opts.sponsorblock_remove = set()
         if opts.sponskrub:
             report_conflict('--allow-unplayable-formats', '--sponskrub')
         opts.sponskrub = False
 
     # PostProcessors
     postprocessors = []
+    if sponsorblock_query:
+        postprocessors.append({
+            'key': 'SponsorBlock',
+            'categories': sponsorblock_query,
+            'api': opts.sponsorblock_api,
+            # Run this immediately after extraction is complete
+            'when': 'pre_process'
+        })
     if opts.parse_metadata:
         postprocessors.append({
             'key': 'MetadataParser',
@@ -439,16 +472,7 @@ def _real_main(argv=None):
             'key': 'FFmpegVideoConvertor',
             'preferedformat': opts.recodevideo,
         })
-    # FFmpegMetadataPP should be run after FFmpegVideoConvertorPP and
-    # FFmpegExtractAudioPP as containers before conversion may not support
-    # metadata (3gp, webm, etc.)
-    # And this post-processor should be placed before other metadata
-    # manipulating post-processors (FFmpegEmbedSubtitle) to prevent loss of
-    # extra metadata. By default ffmpeg preserves metadata applicable for both
-    # source and target containers. From this point the container won't change,
-    # so metadata can be added here.
-    if opts.addmetadata:
-        postprocessors.append({'key': 'FFmpegMetadata'})
+    # If ModifyChapters is going to remove chapters, subtitles must already be in the container.
     if opts.embedsubtitles:
         already_have_subtitle = opts.writesubtitles and 'no-keep-subs' not in compat_opts
         postprocessors.append({
@@ -462,6 +486,33 @@ def _real_main(argv=None):
     # this was the old behaviour if only --all-sub was given.
     if opts.allsubtitles and not opts.writeautomaticsub:
         opts.writesubtitles = True
+    # ModifyChapters must run before FFmpegMetadataPP
+    remove_chapters_patterns = []
+    for regex in opts.remove_chapters:
+        try:
+            remove_chapters_patterns.append(re.compile(regex))
+        except re.error as err:
+            parser.error(f'invalid --remove-chapters regex {regex!r} - {err}')
+    if opts.remove_chapters or sponsorblock_query:
+        postprocessors.append({
+            'key': 'ModifyChapters',
+            'remove_chapters_patterns': remove_chapters_patterns,
+            'remove_sponsor_segments': opts.sponsorblock_remove,
+            'sponsorblock_chapter_title': opts.sponsorblock_chapter_title,
+            'force_keyframes': opts.force_keyframes_at_cuts
+        })
+    # FFmpegMetadataPP should be run after FFmpegVideoConvertorPP and
+    # FFmpegExtractAudioPP as containers before conversion may not support
+    # metadata (3gp, webm, etc.)
+    # By default ffmpeg preserves metadata applicable for both
+    # source and target containers. From this point the container won't change,
+    # so metadata can be added here.
+    if opts.addmetadata or opts.addchapters:
+        postprocessors.append({
+            'key': 'FFmpegMetadata',
+            'add_chapters': opts.addchapters,
+            'add_metadata': opts.addmetadata,
+        })
     # This should be above EmbedThumbnail since sponskrub removes the thumbnail attachment
     # but must be below EmbedSubtitle and FFmpegMetadata
     # See https://github.com/yt-dlp/yt-dlp/issues/204 , https://github.com/faissaloo/SponSkrub/issues/29
@@ -485,7 +536,10 @@ def _real_main(argv=None):
         if not already_have_thumbnail:
             opts.writethumbnail = True
     if opts.split_chapters:
-        postprocessors.append({'key': 'FFmpegSplitChapters'})
+        postprocessors.append({
+            'key': 'FFmpegSplitChapters',
+            'force_keyframes': opts.force_keyframes_at_cuts,
+        })
     # XAttrMetadataPP should be run after post-processors that may change file contents
     if opts.xattrs:
         postprocessors.append({'key': 'XAttrMetadata'})

yt_dlp/options.py

@@ -28,7 +28,9 @@ from .postprocessor import (
     FFmpegSubtitlesConvertorPP,
     FFmpegThumbnailsConvertorPP,
     FFmpegVideoRemuxerPP,
+    SponsorBlockPP,
 )
+from .postprocessor.modify_chapters import DEFAULT_SPONSORBLOCK_CHAPTER_TITLE
 
 
 def _hide_login_info(opts):
@@ -1218,10 +1220,10 @@ def parseOpts(overrideArguments=None):
             'Give these arguments to the postprocessors. '
             'Specify the postprocessor/executable name and the arguments separated by a colon ":" '
             'to give the argument to the specified postprocessor/executable. Supported PP are: '
-            'Merger, ExtractAudio, SplitChapters, Metadata, EmbedSubtitle, EmbedThumbnail, '
-            'SubtitlesConvertor, ThumbnailsConvertor, VideoRemuxer, VideoConvertor, '
-            'SponSkrub, FixupStretched, FixupM4a, FixupM3u8, FixupTimestamp and FixupDuration. '
-            'The supported executables are: AtomicParsley, FFmpeg, FFprobe, and SponSkrub. '
+            'Merger, ModifyChapters, SplitChapters, ExtractAudio, VideoRemuxer, VideoConvertor, '
+            'Metadata, EmbedSubtitle, EmbedThumbnail, SubtitlesConvertor, ThumbnailsConvertor, '
+            'FixupStretched, FixupM4a, FixupM3u8, FixupTimestamp and FixupDuration. '
+            'The supported executables are: AtomicParsley, FFmpeg and FFprobe. '
             'You can also specify "PP+EXE:ARGS" to give the arguments to the specified executable '
             'only when being used by the specified postprocessor. Additionally, for ffmpeg/ffprobe, '
             '"_i"/"_o" can be appended to the prefix optionally followed by a number to pass the argument '
@@ -1263,11 +1265,19 @@ def parseOpts(overrideArguments=None):
     postproc.add_option(
         '--embed-metadata', '--add-metadata',
         action='store_true', dest='addmetadata', default=False,
-        help='Embed metadata including chapter markers (if supported by the format) to the video file (Alias: --add-metadata)')
+        help='Embed metadata to the video file. Also adds chapters to file unless --no-add-chapters is used (Alias: --add-metadata)')
     postproc.add_option(
         '--no-embed-metadata', '--no-add-metadata',
         action='store_false', dest='addmetadata',
-        help='Do not write metadata (default)  (Alias: --no-add-metadata)')
+        help='Do not add metadata to file (default) (Alias: --no-add-metadata)')
+    postproc.add_option(
+        '--embed-chapters', '--add-chapters',
+        action='store_true', dest='addchapters', default=None,
+        help='Add chapter markers to the video file (Alias: --add-chapters)')
+    postproc.add_option(
+        '--no-embed-chapters', '--no-add-chapters',
+        action='store_false', dest='addchapters',
+        help='Do not add chapter markers (default) (Alias: --no-add-chapters)')
     postproc.add_option(
         '--metadata-from-title',
         metavar='FORMAT', dest='metafromtitle',
@@ -1354,41 +1364,90 @@ def parseOpts(overrideArguments=None):
         '--no-split-chapters', '--no-split-tracks',
         dest='split_chapters', action='store_false',
         help='Do not split video based on chapters (default)')
+    postproc.add_option(
+        '--remove-chapters',
+        metavar='REGEX', dest='remove_chapters', action='append',
+        help='Remove chapters whose title matches the given regular expression. This option can be used multiple times')
+    postproc.add_option(
+        '--no-remove-chapters', dest='remove_chapters', action='store_const', const=None,
+        help='Do not remove any chapters from the file (default)')
+    postproc.add_option(
+        '--force-keyframes-at-cuts',
+        action='store_true', dest='force_keyframes_at_cuts', default=False,
+        help=(
+            'Force keyframes around the chapters before removing/splitting them. '
+            'Requires a reencode and thus is very slow, but the resulting video '
+            'may have fewer artifacts around the cuts'))
+    postproc.add_option(
+        '--no-force-keyframes-at-cuts',
+        action='store_false', dest='force_keyframes_at_cuts',
+        help='Do not force keyframes around the chapters when cutting/splitting (default)')
 
-    sponskrub = optparse.OptionGroup(parser, 'SponSkrub (SponsorBlock) Options', description=(
-        'SponSkrub (https://github.com/yt-dlp/SponSkrub) is a utility to mark/remove sponsor segments '
-        'from downloaded YouTube videos using SponsorBlock API (https://sponsor.ajay.app)'))
-    sponskrub.add_option(
+    sponsorblock = optparse.OptionGroup(parser, 'SponsorBlock Options', description=(
+        'Make chapter entries for, or remove various segments (sponsor, introductions, etc.) '
+        'from downloaded YouTube videos using the SponsorBlock API (https://sponsor.ajay.app)'))
+    sponsorblock.add_option(
+        '--sponsorblock-mark', metavar='CATS',
+        dest='sponsorblock_mark', default=set(), action='callback', type='str',
+        callback=_set_from_options_callback, callback_kwargs={'allowed_values': SponsorBlockPP.CATEGORIES.keys()},
+        help=(
+            'SponsorBlock categories to create chapters for, separated by commas. '
+            'Available categories are all, %s. You can prefix the category with a "-" to exempt it. '
+            'See https://wiki.sponsor.ajay.app/index.php/Segment_Categories for description of the categories. '
+            'Eg: --sponsorblock-query all,-preview' % ', '.join(SponsorBlockPP.CATEGORIES.keys())))
+    sponsorblock.add_option(
+        '--sponsorblock-remove', metavar='CATS',
+        dest='sponsorblock_remove', default=set(), action='callback', type='str',
+        callback=_set_from_options_callback, callback_kwargs={'allowed_values': SponsorBlockPP.CATEGORIES.keys()},
+        help=(
+            'SponsorBlock categories to be removed from the video file, separated by commas. '
+            'If a category is present in both mark and remove, remove takes precedence. '
+            'The syntax and available categories are the same as for --sponsorblock-mark'))
+    sponsorblock.add_option(
+        '--sponsorblock-chapter-title', metavar='TEMPLATE',
+        default=DEFAULT_SPONSORBLOCK_CHAPTER_TITLE, dest='sponsorblock_chapter_title',
+        help=(
+            'The title template for SponsorBlock chapters created by --sponsorblock-mark. '
+            'The same syntax as the output template is used, but the only available fields are '
+            'start_time, end_time, category, categories, name, category_names. Defaults to "%default"'))
+    sponsorblock.add_option(
+        '--no-sponsorblock', default=False,
+        action='store_true', dest='no_sponsorblock',
+        help='Disable both --sponsorblock-mark and --sponsorblock-remove')
+    sponsorblock.add_option(
+        '--sponsorblock-api', metavar='URL',
+        default='https://sponsor.ajay.app', dest='sponsorblock_api',
+        help='SponsorBlock API location, defaults to %default')
+
+    sponsorblock.add_option(
         '--sponskrub',
         action='store_true', dest='sponskrub', default=None,
-        help=(
-            'Use sponskrub to mark sponsored sections. '
-            'This is enabled by default if the sponskrub binary exists (Youtube only)'))
-    sponskrub.add_option(
+        help=optparse.SUPPRESS_HELP)
+    sponsorblock.add_option(
         '--no-sponskrub',
         action='store_false', dest='sponskrub',
-        help='Do not use sponskrub')
-    sponskrub.add_option(
+        help=optparse.SUPPRESS_HELP)
+    sponsorblock.add_option(
         '--sponskrub-cut', default=False,
         action='store_true', dest='sponskrub_cut',
-        help='Cut out the sponsor sections instead of simply marking them')
-    sponskrub.add_option(
+        help=optparse.SUPPRESS_HELP)
+    sponsorblock.add_option(
         '--no-sponskrub-cut',
         action='store_false', dest='sponskrub_cut',
-        help='Simply mark the sponsor sections, not cut them out (default)')
-    sponskrub.add_option(
+        help=optparse.SUPPRESS_HELP)
+    sponsorblock.add_option(
         '--sponskrub-force', default=False,
         action='store_true', dest='sponskrub_force',
-        help='Run sponskrub even if the video was already downloaded')
-    sponskrub.add_option(
+        help=optparse.SUPPRESS_HELP)
+    sponsorblock.add_option(
         '--no-sponskrub-force',
         action='store_true', dest='sponskrub_force',
-        help='Do not cut out the sponsor sections if the video was already downloaded (default)')
-    sponskrub.add_option(
+        help=optparse.SUPPRESS_HELP)
+    sponsorblock.add_option(
         '--sponskrub-location', metavar='PATH',
         dest='sponskrub_path', default='',
-        help='Location of the sponskrub binary; either the path to the binary or its containing directory')
-    sponskrub.add_option(
+        help=optparse.SUPPRESS_HELP)
+    sponsorblock.add_option(
         '--sponskrub-args', dest='sponskrub_args', metavar='ARGS',
         help=optparse.SUPPRESS_HELP)
 
@@ -1457,7 +1516,7 @@ def parseOpts(overrideArguments=None):
     parser.add_option_group(subtitles)
     parser.add_option_group(authentication)
     parser.add_option_group(postproc)
-    parser.add_option_group(sponskrub)
+    parser.add_option_group(sponsorblock)
     parser.add_option_group(extractor)
 
     if overrideArguments is not None:

yt_dlp/postprocessor/__init__.py

@@ -26,7 +26,9 @@ from .metadataparser import (
     MetadataParserPP,
 )
 from .movefilesafterdownload import MoveFilesAfterDownloadPP
+from .sponsorblock import SponsorBlockPP
 from .sponskrub import SponSkrubPP
+from .modify_chapters import ModifyChaptersPP
 
 
 def get_postprocessor(key):
@@ -56,6 +58,8 @@ __all__ = [
     'MetadataFromFieldPP',
     'MetadataFromTitlePP',
     'MoveFilesAfterDownloadPP',
+    'SponsorBlockPP',
     'SponSkrubPP',
+    'ModifyChaptersPP',
     'XAttrMetadataPP',
 ]

yt_dlp/postprocessor/ffmpeg.py

@@ -8,22 +8,22 @@ import time
 import re
 import json
 
-
 from .common import AudioConversionError, PostProcessor
 
 from ..compat import compat_str, compat_numeric_types
 from ..utils import (
+    dfxp2srt,
     encodeArgument,
     encodeFilename,
     get_exe_version,
     is_outdated_version,
+    ISO639Utils,
+    orderedSet,
     PostProcessingError,
     prepend_extension,
-    shell_quote,
-    dfxp2srt,
-    ISO639Utils,
     process_communicate_or_kill,
     replace_extension,
+    shell_quote,
     traverse_obj,
     variadic,
 )
@@ -281,7 +281,8 @@ class FFmpegPostProcessor(PostProcessor):
     def run_ffmpeg(self, path, out_path, opts, **kwargs):
         return self.run_ffmpeg_multiple_files([path], out_path, opts, **kwargs)
 
-    def _ffmpeg_filename_argument(self, fn):
+    @staticmethod
+    def _ffmpeg_filename_argument(fn):
         # Always use 'file:' because the filename may contain ':' (ffmpeg
         # interprets that as a protocol) or can start with '-' (-- is broken in
         # ffmpeg, see https://ffmpeg.org/trac/ffmpeg/ticket/2127 for details)
@@ -290,6 +291,62 @@ class FFmpegPostProcessor(PostProcessor):
             return fn
         return 'file:' + fn if fn != '-' else fn
 
+    @staticmethod
+    def _quote_for_ffmpeg(string):
+        # See https://ffmpeg.org/ffmpeg-utils.html#toc-Quoting-and-escaping
+        # A sequence of '' produces '\'''\'';
+        # final replace removes the empty '' between \' \'.
+        string = string.replace("'", r"'\''").replace("'''", "'")
+        # Handle potential ' at string boundaries.
+        string = string[1:] if string[0] == "'" else "'" + string
+        return string[:-1] if string[-1] == "'" else string + "'"
+
+    def force_keyframes(self, filename, timestamps):
+        timestamps = orderedSet(timestamps)
+        if timestamps[0] == 0:
+            timestamps = timestamps[1:]
+        keyframe_file = prepend_extension(filename, 'keyframes.temp')
+        self.to_screen(f'Re-encoding "{filename}" with appropriate keyframes')
+        self.run_ffmpeg(filename, keyframe_file, ['-force_key_frames', ','.join(
+            f'{t:.6f}' for t in timestamps)])
+        return keyframe_file
+
+    def concat_files(self, in_files, out_file, concat_opts=None):
+        """
+        Use concat demuxer to concatenate multiple files having identical streams.
+
+        Only inpoint, outpoint, and duration concat options are supported.
+        See https://ffmpeg.org/ffmpeg-formats.html#concat-1 for details
+        """
+        concat_file = f'{out_file}.concat'
+        self.write_debug(f'Writing concat spec to {concat_file}')
+        with open(concat_file, 'wt', encoding='utf-8') as f:
+            f.writelines(self._concat_spec(in_files, concat_opts))
+
+        out_flags = ['-c', 'copy']
+        if out_file.rpartition('.')[-1] in ('mp4', 'mov'):
+            # For some reason, '-c copy' is not enough to copy subtitles
+            out_flags.extend(['-c:s', 'mov_text', '-movflags', '+faststart'])
+
+        try:
+            self.real_run_ffmpeg(
+                [(concat_file, ['-hide_banner', '-nostdin', '-f', 'concat', '-safe', '0'])],
+                [(out_file, out_flags)])
+        finally:
+            os.remove(concat_file)
+
+    @classmethod
+    def _concat_spec(cls, in_files, concat_opts=None):
+        if concat_opts is None:
+            concat_opts = [{}] * len(in_files)
+        yield 'ffconcat version 1.0\n'
+        for file, opts in zip(in_files, concat_opts):
+            yield f'file {cls._quote_for_ffmpeg(cls._ffmpeg_filename_argument(file))}\n'
+            # Iterate explicitly to yield the following directives in order, ignoring the rest.
+            for directive in 'inpoint', 'outpoint', 'duration':
+                if directive in opts:
+                    yield f'{directive} {opts[directive]}\n'
+
 
 class FFmpegExtractAudioPP(FFmpegPostProcessor):
     COMMON_AUDIO_EXTS = ('wav', 'flac', 'm4a', 'aiff', 'mp3', 'ogg', 'mka', 'opus', 'wma')
@@ -531,6 +588,11 @@ class FFmpegEmbedSubtitlePP(FFmpegPostProcessor):
 
 class FFmpegMetadataPP(FFmpegPostProcessor):
 
+    def __init__(self, downloader, add_metadata=True, add_chapters=True):
+        FFmpegPostProcessor.__init__(self, downloader)
+        self._add_metadata = add_metadata
+        self._add_chapters = add_chapters
+
     @staticmethod
     def _options(target_ext):
         yield from ('-map', '0', '-dn')
@@ -541,6 +603,46 @@ class FFmpegMetadataPP(FFmpegPostProcessor):
 
     @PostProcessor._restrict_to(images=False)
     def run(self, info):
+        filename, metadata_filename = info['filepath'], None
+        options = []
+        if self._add_chapters and info.get('chapters'):
+            metadata_filename = replace_extension(filename, 'meta')
+            options.extend(self._get_chapter_opts(info['chapters'], metadata_filename))
+        if self._add_metadata:
+            options.extend(self._get_metadata_opts(info))
+
+        if not options:
+            self.to_screen('There isn\'t any metadata to add')
+            return [], info
+
+        temp_filename = prepend_extension(filename, 'temp')
+        self.to_screen('Adding metadata to "%s"' % filename)
+        self.run_ffmpeg_multiple_files(
+            (filename, metadata_filename), temp_filename,
+            itertools.chain(self._options(info['ext']), *options))
+        if metadata_filename:
+            os.remove(metadata_filename)
+        os.replace(temp_filename, filename)
+        return [], info
+
+    @staticmethod
+    def _get_chapter_opts(chapters, metadata_filename):
+        with io.open(metadata_filename, 'wt', encoding='utf-8') as f:
+            def ffmpeg_escape(text):
+                return re.sub(r'([\\=;#\n])', r'\\\1', text)
+
+            metadata_file_content = ';FFMETADATA1\n'
+            for chapter in chapters:
+                metadata_file_content += '[CHAPTER]\nTIMEBASE=1/1000\n'
+                metadata_file_content += 'START=%d\n' % (chapter['start_time'] * 1000)
+                metadata_file_content += 'END=%d\n' % (chapter['end_time'] * 1000)
+                chapter_title = chapter.get('title')
+                if chapter_title:
+                    metadata_file_content += 'title=%s\n' % ffmpeg_escape(chapter_title)
+            f.write(metadata_file_content)
+        yield ('-map_metadata', '1')
+
+    def _get_metadata_opts(self, info):
         metadata = {}
 
         def add(meta_list, info_list=None):
@@ -577,61 +679,27 @@ class FFmpegMetadataPP(FFmpegPostProcessor):
         for key in filter(lambda k: k.startswith(prefix), info.keys()):
             add(key[len(prefix):], key)
 
-        filename, metadata_filename = info['filepath'], None
-        options = [('-metadata', f'{name}={value}') for name, value in metadata.items()]
+        for name, value in metadata.items():
+            yield ('-metadata', f'{name}={value}')
 
         stream_idx = 0
         for fmt in info.get('requested_formats') or []:
             stream_count = 2 if 'none' not in (fmt.get('vcodec'), fmt.get('acodec')) else 1
             if fmt.get('language'):
                 lang = ISO639Utils.short2long(fmt['language']) or fmt['language']
-                options.extend(('-metadata:s:%d' % (stream_idx + i), 'language=%s' % lang)
-                               for i in range(stream_count))
+                for i in range(stream_count):
+                    yield ('-metadata:s:%d' % (stream_idx + i), 'language=%s' % lang)
             stream_idx += stream_count
 
-        chapters = info.get('chapters', [])
-        if chapters:
-            metadata_filename = replace_extension(filename, 'meta')
-            with io.open(metadata_filename, 'wt', encoding='utf-8') as f:
-                def ffmpeg_escape(text):
-                    return re.sub(r'([\\=;#\n])', r'\\\1', text)
-
-                metadata_file_content = ';FFMETADATA1\n'
-                for chapter in chapters:
-                    metadata_file_content += '[CHAPTER]\nTIMEBASE=1/1000\n'
-                    metadata_file_content += 'START=%d\n' % (chapter['start_time'] * 1000)
-                    metadata_file_content += 'END=%d\n' % (chapter['end_time'] * 1000)
-                    chapter_title = chapter.get('title')
-                    if chapter_title:
-                        metadata_file_content += 'title=%s\n' % ffmpeg_escape(chapter_title)
-                f.write(metadata_file_content)
-                options.append(('-map_metadata', '1'))
-
         if ('no-attach-info-json' not in self.get_param('compat_opts', [])
                 and '__infojson_filename' in info and info['ext'] in ('mkv', 'mka')):
-            old_stream, new_stream = self.get_stream_number(filename, ('tags', 'mimetype'), 'application/json')
+            old_stream, new_stream = self.get_stream_number(info['filepath'], ('tags', 'mimetype'), 'application/json')
             if old_stream is not None:
-                options.append(('-map', '-0:%d' % old_stream))
+                yield ('-map', '-0:%d' % old_stream)
                 new_stream -= 1
 
-            options.append((
-                '-attach', info['__infojson_filename'],
-                '-metadata:s:%d' % new_stream, 'mimetype=application/json'
-            ))
-
-        if not options:
-            self.to_screen('There isn\'t any metadata to add')
-            return [], info
-
-        temp_filename = prepend_extension(filename, 'temp')
-        self.to_screen('Adding metadata to "%s"' % filename)
-        self.run_ffmpeg_multiple_files(
-            (filename, metadata_filename), temp_filename,
-            itertools.chain(self._options(info['ext']), *options))
-        if chapters:
-            os.remove(metadata_filename)
-        os.replace(temp_filename, filename)
-        return [], info
+            yield ('-attach', info['__infojson_filename'],
+                   '-metadata:s:%d' % new_stream, 'mimetype=application/json')
 
 
 class FFmpegMergerPP(FFmpegPostProcessor):
@@ -808,6 +876,9 @@ class FFmpegSubtitlesConvertorPP(FFmpegPostProcessor):
 
 
 class FFmpegSplitChaptersPP(FFmpegPostProcessor):
+    def __init__(self, downloader, force_keyframes=False):
+        FFmpegPostProcessor.__init__(self, downloader)
+        self._force_keyframes = force_keyframes
 
     def _prepare_filename(self, number, chapter, info):
         info = info.copy()
@@ -835,13 +906,18 @@ class FFmpegSplitChaptersPP(FFmpegPostProcessor):
     def run(self, info):
         chapters = info.get('chapters') or []
         if not chapters:
-            self.report_warning('Chapter information is unavailable')
+            self.to_screen('Chapter information is unavailable')
             return [], info
 
+        in_file = info['filepath']
+        if self._force_keyframes and len(chapters) > 1:
+            in_file = self.force_keyframes(in_file, (c['start_time'] for c in chapters))
         self.to_screen('Splitting video by chapters; %d chapters found' % len(chapters))
         for idx, chapter in enumerate(chapters):
             destination, opts = self._ffmpeg_args_for_chapter(idx + 1, chapter, info)
-            self.real_run_ffmpeg([(info['filepath'], opts)], [(destination, ['-c', 'copy'])])
+            self.real_run_ffmpeg([(in_file, opts)], [(destination, ['-c', 'copy'])])
+        if in_file != info['filepath']:
+            os.remove(in_file)
         return [], info
 
 

yt_dlp/postprocessor/modify_chapters.py

@@ -0,0 +1,333 @@
+import copy
+import heapq
+import os
+
+from .common import PostProcessor
+from .ffmpeg import (
+    FFmpegPostProcessor,
+    FFmpegSubtitlesConvertorPP
+)
+from .sponsorblock import SponsorBlockPP
+from ..utils import (
+    float_or_none,
+    orderedSet,
+    PostProcessingError,
+    prepend_extension,
+    traverse_obj,
+)
+
+
+_TINY_SPONSOR_OVERLAP_DURATION = 1
+DEFAULT_SPONSORBLOCK_CHAPTER_TITLE = '[SponsorBlock]: %(category_names)l'
+
+
+class ModifyChaptersPP(FFmpegPostProcessor):
+    def __init__(self, downloader, remove_chapters_patterns=None, remove_sponsor_segments=None,
+                 sponsorblock_chapter_title=DEFAULT_SPONSORBLOCK_CHAPTER_TITLE, force_keyframes=False):
+        FFmpegPostProcessor.__init__(self, downloader)
+        self._remove_chapters_patterns = set(remove_chapters_patterns or [])
+        self._remove_sponsor_segments = set(remove_sponsor_segments or [])
+        self._sponsorblock_chapter_title = sponsorblock_chapter_title
+        self._force_keyframes = force_keyframes
+
+    @PostProcessor._restrict_to(images=False)
+    def run(self, info):
+        chapters, sponsor_chapters = self._mark_chapters_to_remove(
+            info.get('chapters') or [], info.get('sponsorblock_chapters') or [])
+        if not chapters and not sponsor_chapters:
+            return [], info
+
+        real_duration = self._get_real_video_duration(info['filepath'])
+        if not chapters:
+            chapters = [{'start_time': 0, 'end_time': real_duration, 'title': info['title']}]
+
+        info['chapters'], cuts = self._remove_marked_arrange_sponsors(chapters + sponsor_chapters)
+        if not cuts:
+            return [], info
+
+        if abs(real_duration - info['duration']) > 1:
+            if abs(real_duration - info['chapters'][-1]['end_time']) < 1:
+                self.to_screen(f'Skipping {self.pp_key()} since the video appears to be already cut')
+                return [], info
+            if not info.get('__real_download'):
+                raise PostProcessingError('Cannot cut video since the real and expected durations mismatch. '
+                                          'Different chapters may have already been removed')
+                return [], info
+            else:
+                self.write_debug('Expected and actual durations mismatch')
+
+        concat_opts = self._make_concat_opts(cuts, real_duration)
+
+        def remove_chapters(file, is_sub):
+            return file, self.remove_chapters(file, cuts, concat_opts, self._force_keyframes and not is_sub)
+
+        in_out_files = [remove_chapters(info['filepath'], False)]
+        in_out_files.extend(remove_chapters(in_file, True) for in_file in self._get_supported_subs(info))
+
+        # Renaming should only happen after all files are processed
+        files_to_remove = []
+        for in_file, out_file in in_out_files:
+            uncut_file = prepend_extension(in_file, 'uncut')
+            os.replace(in_file, uncut_file)
+            os.replace(out_file, in_file)
+            files_to_remove.append(uncut_file)
+
+        return files_to_remove, info
+
+    def _mark_chapters_to_remove(self, chapters, sponsor_chapters):
+        if self._remove_chapters_patterns:
+            warn_no_chapter_to_remove = True
+            if not chapters:
+                self.to_screen('Chapter information is unavailable')
+                warn_no_chapter_to_remove = False
+            for c in chapters:
+                if any(regex.search(c['title']) for regex in self._remove_chapters_patterns):
+                    c['remove'] = True
+                    warn_no_chapter_to_remove = False
+            if warn_no_chapter_to_remove:
+                self.to_screen('There are no chapters matching the regex')
+
+        if self._remove_sponsor_segments:
+            warn_no_chapter_to_remove = True
+            if not sponsor_chapters:
+                self.to_screen('SponsorBlock information is unavailable')
+                warn_no_chapter_to_remove = False
+            for c in sponsor_chapters:
+                if c['category'] in self._remove_sponsor_segments:
+                    c['remove'] = True
+                    warn_no_chapter_to_remove = False
+            if warn_no_chapter_to_remove:
+                self.to_screen('There are no matching SponsorBlock chapters')
+
+        return chapters, sponsor_chapters
+
+    def _get_real_video_duration(self, filename):
+        duration = float_or_none(
+            traverse_obj(self.get_metadata_object(filename), ('format', 'duration')))
+        if duration is None:
+            raise PostProcessingError('ffprobe returned empty duration')
+        return duration
+
+    def _get_supported_subs(self, info):
+        for sub in (info.get('requested_subtitles') or {}).values():
+            sub_file = sub.get('filepath')
+            # The file might have been removed by --embed-subs
+            if not sub_file or not os.path.exists(sub_file):
+                continue
+            ext = sub['ext']
+            if ext not in FFmpegSubtitlesConvertorPP.SUPPORTED_EXTS:
+                self.report_warning(f'Cannot remove chapters from external {ext} subtitles; "{sub_file}" is now out of sync')
+                continue
+            # TODO: create __real_download for subs?
+            yield sub_file
+
+    def _remove_marked_arrange_sponsors(self, chapters):
+        # Store cuts separately, since adjacent and overlapping cuts must be merged.
+        cuts = []
+
+        def append_cut(c):
+            assert 'remove' in c
+            last_to_cut = cuts[-1] if cuts else None
+            if last_to_cut and last_to_cut['end_time'] >= c['start_time']:
+                last_to_cut['end_time'] = max(last_to_cut['end_time'], c['end_time'])
+            else:
+                cuts.append(c)
+            return len(cuts) - 1
+
+        def excess_duration(c):
+            # Cuts that are completely within the chapter reduce chapters' duration.
+            # Since cuts can overlap, excess duration may be less that the sum of cuts' durations.
+            # To avoid that, chapter stores the index to the fist cut within the chapter,
+            # instead of storing excess duration. append_cut ensures that subsequent cuts (if any)
+            # will be merged with previous ones (if necessary).
+            cut_idx, excess = c.pop('cut_idx', len(cuts)), 0
+            while cut_idx < len(cuts):
+                cut = cuts[cut_idx]
+                if cut['start_time'] >= c['end_time']:
+                    break
+                if cut['end_time'] > c['start_time']:
+                    excess += min(cut['end_time'], c['end_time'])
+                    excess -= max(cut['start_time'], c['start_time'])
+                cut_idx += 1
+            return excess
+
+        new_chapters = []
+
+        def chapter_length(c):
+            return c['end_time'] - c['start_time']
+
+        def original_uncut_chapter(c):
+            return '_was_cut' not in c and '_categories' not in c
+
+        def append_chapter(c):
+            assert 'remove' not in c
+            length = chapter_length(c) - excess_duration(c)
+            # Chapter is completely covered by cuts or sponsors.
+            if length <= 0:
+                return
+            start = new_chapters[-1]['end_time'] if new_chapters else 0
+            c.update(start_time=start, end_time=start + length)
+            # Append without checking for tininess to prevent having
+            # a completely empty chapter list.
+            if not new_chapters:
+                new_chapters.append(c)
+                return
+            old_c = new_chapters[-1]
+            # Merge with the previous if the chapter is tiny.
+            # Only tiny chapters resulting from a cut can be skipped.
+            # Chapters that were already tiny in the original list will be preserved.
+            if not original_uncut_chapter(c) and length < _TINY_SPONSOR_OVERLAP_DURATION:
+                old_c['end_time'] = c['end_time']
+            # Previous tiny chapter was appended for the sake of preventing an empty chapter list.
+            # Replace it with the current one.
+            elif not original_uncut_chapter(old_c) and chapter_length(old_c) < _TINY_SPONSOR_OVERLAP_DURATION:
+                c['start_time'] = old_c['start_time']
+                new_chapters[-1] = c
+            else:
+                new_chapters.append(c)
+
+        # Turn into a priority queue, index is a tie breaker.
+        # Plain stack sorted by start_time is not enough: after splitting the chapter,
+        # the part returned to the stack is not guaranteed to have start_time
+        # less than or equal to the that of the stack's head.
+        chapters = [(c['start_time'], i, c) for i, c in enumerate(chapters)]
+        heapq.heapify(chapters)
+
+        _, cur_i, cur_chapter = heapq.heappop(chapters)
+        while chapters:
+            _, i, c = heapq.heappop(chapters)
+            # Non-overlapping chapters or cuts can be appended directly. However,
+            # adjacent non-overlapping cuts must be merged, which is handled by append_cut.
+            if cur_chapter['end_time'] <= c['start_time']:
+                (append_chapter if 'remove' not in cur_chapter else append_cut)(cur_chapter)
+                cur_i, cur_chapter = i, c
+                continue
+
+            # Eight possibilities for overlapping chapters: (cut, cut), (cut, sponsor),
+            # (cut, normal), (sponsor, cut), (normal, cut), (sponsor, sponsor),
+            # (sponsor, normal), and (normal, sponsor). There is no (normal, normal):
+            # normal chapters are assumed not to overlap.
+            if 'remove' in cur_chapter:
+                # (cut, cut): adjust end_time.
+                if 'remove' in c:
+                    cur_chapter['end_time'] = max(cur_chapter['end_time'], c['end_time'])
+                # (cut, sponsor/normal): chop the beginning of the later chapter
+                # (if it's not completely hidden by the cut). Push to the priority queue
+                # to restore sorting by start_time: with beginning chopped, c may actually
+                # start later than the remaining chapters from the queue.
+                elif cur_chapter['end_time'] < c['end_time']:
+                    c['start_time'] = cur_chapter['end_time']
+                    c['_was_cut'] = True
+                    heapq.heappush(chapters, (c['start_time'], i, c))
+            # (sponsor/normal, cut).
+            elif 'remove' in c:
+                cur_chapter['_was_cut'] = True
+                # Chop the end of the current chapter if the cut is not contained within it.
+                # Chopping the end doesn't break start_time sorting, no PQ push is necessary.
+                if cur_chapter['end_time'] <= c['end_time']:
+                    cur_chapter['end_time'] = c['start_time']
+                    append_chapter(cur_chapter)
+                    cur_i, cur_chapter = i, c
+                    continue
+                # Current chapter contains the cut within it. If the current chapter is
+                # a sponsor chapter, check whether the categories before and after the cut differ.
+                if '_categories' in cur_chapter:
+                    after_c = dict(cur_chapter, start_time=c['end_time'], _categories=[])
+                    cur_cats = []
+                    for cat_start_end in cur_chapter['_categories']:
+                        if cat_start_end[1] < c['start_time']:
+                            cur_cats.append(cat_start_end)
+                        if cat_start_end[2] > c['end_time']:
+                            after_c['_categories'].append(cat_start_end)
+                    cur_chapter['_categories'] = cur_cats
+                    if cur_chapter['_categories'] != after_c['_categories']:
+                        # Categories before and after the cut differ: push the after part to PQ.
+                        heapq.heappush(chapters, (after_c['start_time'], cur_i, after_c))
+                        cur_chapter['end_time'] = c['start_time']
+                        append_chapter(cur_chapter)
+                        cur_i, cur_chapter = i, c
+                        continue
+                # Either sponsor categories before and after the cut are the same or
+                # we're dealing with a normal chapter. Just register an outstanding cut:
+                # subsequent append_chapter will reduce the duration.
+                cur_chapter.setdefault('cut_idx', append_cut(c))
+            # (sponsor, normal): if a normal chapter is not completely overlapped,
+            # chop the beginning of it and push it to PQ.
+            elif '_categories' in cur_chapter and '_categories' not in c:
+                if cur_chapter['end_time'] < c['end_time']:
+                    c['start_time'] = cur_chapter['end_time']
+                    c['_was_cut'] = True
+                    heapq.heappush(chapters, (c['start_time'], i, c))
+            # (normal, sponsor) and (sponsor, sponsor)
+            else:
+                assert '_categories' in c
+                cur_chapter['_was_cut'] = True
+                c['_was_cut'] = True
+                # Push the part after the sponsor to PQ.
+                if cur_chapter['end_time'] > c['end_time']:
+                    # deepcopy to make categories in after_c and cur_chapter/c refer to different lists.
+                    after_c = dict(copy.deepcopy(cur_chapter), start_time=c['end_time'])
+                    heapq.heappush(chapters, (after_c['start_time'], cur_i, after_c))
+                # Push the part after the overlap to PQ.
+                elif c['end_time'] > cur_chapter['end_time']:
+                    after_cur = dict(copy.deepcopy(c), start_time=cur_chapter['end_time'])
+                    heapq.heappush(chapters, (after_cur['start_time'], cur_i, after_cur))
+                    c['end_time'] = cur_chapter['end_time']
+                # (sponsor, sponsor): merge categories in the overlap.
+                if '_categories' in cur_chapter:
+                    c['_categories'] = cur_chapter['_categories'] + c['_categories']
+                # Inherit the cuts that the current chapter has accumulated within it.
+                if 'cut_idx' in cur_chapter:
+                    c['cut_idx'] = cur_chapter['cut_idx']
+                cur_chapter['end_time'] = c['start_time']
+                append_chapter(cur_chapter)
+                cur_i, cur_chapter = i, c
+        (append_chapter if 'remove' not in cur_chapter else append_cut)(cur_chapter)
+
+        i = -1
+        for c in new_chapters.copy():
+            i += 1
+            c.pop('_was_cut', None)
+            cats = c.pop('_categories', None)
+            if cats:
+                category = min(cats, key=lambda c: c[2] - c[1])[0]
+                cats = orderedSet(x[0] for x in cats)
+                c.update({
+                    'category': category,
+                    'categories': cats,
+                    'name': SponsorBlockPP.CATEGORIES[category],
+                    'category_names': [SponsorBlockPP.CATEGORIES[c] for c in cats]
+                })
+                outtmpl, tmpl_dict = self._downloader.prepare_outtmpl(self._sponsorblock_chapter_title, c)
+                c['title'] = self._downloader.escape_outtmpl(outtmpl) % tmpl_dict
+            if i > 0 and c['title'] == new_chapters[i - 1]['title']:
+                new_chapters[i - 1]['end_time'] = c['end_time']
+                new_chapters.pop(i)
+                i -= 1
+
+        return new_chapters, cuts
+
+    def remove_chapters(self, filename, ranges_to_cut, concat_opts, force_keyframes=False):
+        in_file = filename
+        out_file = prepend_extension(in_file, 'temp')
+        if force_keyframes:
+            in_file = self.force_keyframes(in_file, (t for r in ranges_to_cut for t in r))
+        self.to_screen(f'Removing chapters from {filename}')
+        self.concat_files([in_file] * len(concat_opts), out_file, concat_opts)
+        if in_file != filename:
+            os.remove(in_file)
+        return out_file
+
+    @staticmethod
+    def _make_concat_opts(chapters_to_remove, duration):
+        opts = [{}]
+        for s in chapters_to_remove:
+            # Do not create 0 duration chunk at the beginning.
+            if s['start_time'] == 0:
+                opts[-1]['inpoint'] = f'{s["end_time"]:.6f}'
+                continue
+            opts[-1]['outpoint'] = f'{s["start_time"]:.6f}'
+            # Do not create 0 duration chunk at the end.
+            if s['end_time'] != duration:
+                opts.append({'inpoint': f'{s["end_time"]:.6f}'})
+        return opts

yt_dlp/postprocessor/sponskrub.py

@@ -17,6 +17,7 @@ from ..utils import (
 )
 
 
+# Deprecated in favor of the native implementation
 class SponSkrubPP(PostProcessor):
     _temp_ext = 'spons'
     _exe_name = 'sponskrub'

yt_dlp/postprocessor/sponsorblock.py

@@ -0,0 +1,96 @@
+import json
+import re
+from hashlib import sha256
+
+from .ffmpeg import FFmpegPostProcessor
+from ..compat import compat_urllib_parse_urlencode, compat_HTTPError
+from ..utils import PostProcessingError, sanitized_Request
+
+
+class SponsorBlockPP(FFmpegPostProcessor):
+
+    EXTRACTORS = {
+        'Youtube': 'YouTube',
+    }
+    CATEGORIES = {
+        'sponsor': 'Sponsor',
+        'intro': 'Intermission/Intro Animation',
+        'outro': 'Endcards/Credits',
+        'selfpromo': 'Unpaid/Self Promotion',
+        'interaction': 'Interaction Reminder',
+        'preview': 'Preview/Recap',
+        'music_offtopic': 'Non-Music Section'
+    }
+
+    def __init__(self, downloader, categories=None, api='https://sponsor.ajay.app'):
+        FFmpegPostProcessor.__init__(self, downloader)
+        self._categories = tuple(categories or self.CATEGORIES.keys())
+        self._API_URL = api if re.match('^https?://', api) else 'https://' + api
+
+    def run(self, info):
+        extractor = info['extractor_key']
+        if extractor not in self.EXTRACTORS:
+            self.to_screen(f'SponsorBlock is not supported for {extractor}')
+            return [], info
+
+        info['sponsorblock_chapters'] = self._get_sponsor_chapters(info, info['duration'])
+        return [], info
+
+    def _get_sponsor_chapters(self, info, duration):
+        segments = self._get_sponsor_segments(info['id'], self.EXTRACTORS[info['extractor_key']])
+
+        def duration_filter(s):
+            start_end = s['segment']
+            # Ignore milliseconds difference at the start.
+            if start_end[0] <= 1:
+                start_end[0] = 0
+            # Ignore milliseconds difference at the end.
+            # Never allow the segment to exceed the video.
+            if duration and duration - start_end[1] <= 1:
+                start_end[1] = duration
+            # SponsorBlock duration may be absent or it may deviate from the real one.
+            return s['videoDuration'] == 0 or not duration or abs(duration - s['videoDuration']) <= 1
+
+        duration_match = [s for s in segments if duration_filter(s)]
+        if len(duration_match) != len(segments):
+            self.report_warning('Some SponsorBlock segments are from a video of different duration, maybe from an old version of this video')
+
+        def to_chapter(s):
+            (start, end), cat = s['segment'], s['category']
+            return {
+                'start_time': start,
+                'end_time': end,
+                'category': cat,
+                'title': self.CATEGORIES[cat],
+                '_categories': [(cat, start, end)]
+            }
+
+        sponsor_chapters = [to_chapter(s) for s in duration_match]
+        if not sponsor_chapters:
+            self.to_screen('No segments were found in the SponsorBlock database')
+        else:
+            self.to_screen(f'Found {len(sponsor_chapters)} segments in the SponsorBlock database')
+        return sponsor_chapters
+
+    def _get_sponsor_segments(self, video_id, service):
+        hash = sha256(video_id.encode('ascii')).hexdigest()
+        # SponsorBlock API recommends using first 4 hash characters.
+        url = f'{self._API_URL}/api/skipSegments/{hash[:4]}?' + compat_urllib_parse_urlencode({
+            'service': service,
+            'categories': json.dumps(self._categories),
+        })
+        for d in self._get_json(url):
+            if d['videoID'] == video_id:
+                return d['segments']
+        return []
+
+    def _get_json(self, url):
+        self.write_debug(f'SponsorBlock query: {url}')
+        try:
+            rsp = self._downloader.urlopen(sanitized_Request(url))
+        except compat_HTTPError as e:
+            if e.code == 404:
+                return []
+            raise PostProcessingError(f'Error communicating with SponsorBlock API - {e}')
+
+        return json.loads(rsp.read().decode(rsp.info().get_param('charset') or 'utf-8'))