yt-dlp/yt_dlp/extractor/tiktok.py

1309 lines
57 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

import itertools
import json
import random
import re
import string
import time
from .common import InfoExtractor
from ..compat import compat_urllib_parse_urlparse
from ..networking import HEADRequest
from ..utils import (
ExtractorError,
LazyList,
UnsupportedError,
UserNotLive,
determine_ext,
format_field,
int_or_none,
join_nonempty,
merge_dicts,
qualities,
remove_start,
srt_subtitles_timecode,
str_or_none,
traverse_obj,
try_call,
try_get,
url_or_none,
)
class TikTokBaseIE(InfoExtractor):
_APP_VERSIONS = [('26.1.3', '260103'), ('26.1.2', '260102'), ('26.1.1', '260101'), ('25.6.2', '250602')]
_WORKING_APP_VERSION = None
_APP_NAME = 'trill'
_AID = 1180
_UPLOADER_URL_FORMAT = 'https://www.tiktok.com/@%s'
_WEBPAGE_HOST = 'https://www.tiktok.com/'
QUALITIES = ('360p', '540p', '720p', '1080p')
@property
def _API_HOSTNAME(self):
return self._configuration_arg(
'api_hostname', ['api16-normal-c-useast1a.tiktokv.com'], ie_key=TikTokIE)[0]
@staticmethod
def _create_url(user_id, video_id):
return f'https://www.tiktok.com/@{user_id or "_"}/video/{video_id}'
def _get_sigi_state(self, webpage, display_id):
return self._search_json(
r'<script[^>]+\bid="(?:SIGI_STATE|sigi-persisted-data)"[^>]*>', webpage,
'sigi state', display_id, end_pattern=r'</script>', default={})
def _get_universal_data(self, webpage, display_id):
return traverse_obj(self._search_json(
r'<script[^>]+\bid="__UNIVERSAL_DATA_FOR_REHYDRATION__"[^>]*>', webpage,
'universal data', display_id, end_pattern=r'</script>', default={}),
('__DEFAULT_SCOPE__', {dict})) or {}
def _call_api_impl(self, ep, query, manifest_app_version, video_id, fatal=True,
note='Downloading API JSON', errnote='Unable to download API page'):
self._set_cookie(self._API_HOSTNAME, 'odin_tt', ''.join(random.choices('0123456789abcdef', k=160)))
webpage_cookies = self._get_cookies(self._WEBPAGE_HOST)
if webpage_cookies.get('sid_tt'):
self._set_cookie(self._API_HOSTNAME, 'sid_tt', webpage_cookies['sid_tt'].value)
return self._download_json(
'https://%s/aweme/v1/%s/' % (self._API_HOSTNAME, ep), video_id=video_id,
fatal=fatal, note=note, errnote=errnote, headers={
'User-Agent': f'com.ss.android.ugc.{self._APP_NAME}/{manifest_app_version} (Linux; U; Android 13; en_US; Pixel 7; Build/TD1A.220804.031; Cronet/58.0.2991.0)',
'Accept': 'application/json',
}, query=query)
def _build_api_query(self, query, app_version, manifest_app_version):
return {
**query,
'version_name': app_version,
'version_code': manifest_app_version,
'build_number': app_version,
'manifest_version_code': manifest_app_version,
'update_version_code': manifest_app_version,
'openudid': ''.join(random.choices('0123456789abcdef', k=16)),
'uuid': ''.join(random.choices(string.digits, k=16)),
'_rticket': int(time.time() * 1000),
'ts': int(time.time()),
'device_brand': 'Google',
'device_type': 'Pixel 7',
'device_platform': 'android',
'resolution': '1080*2400',
'dpi': 420,
'os_version': '13',
'os_api': '29',
'carrier_region': 'US',
'sys_region': 'US',
'region': 'US',
'app_name': self._APP_NAME,
'app_language': 'en',
'language': 'en',
'timezone_name': 'America/New_York',
'timezone_offset': '-14400',
'channel': 'googleplay',
'ac': 'wifi',
'mcc_mnc': '310260',
'is_my_cn': 0,
'aid': self._AID,
'ssmix': 'a',
'as': 'a1qwert123',
'cp': 'cbfhckdckkde1',
}
def _call_api(self, ep, query, video_id, fatal=True,
note='Downloading API JSON', errnote='Unable to download API page'):
if not self._WORKING_APP_VERSION:
app_version = self._configuration_arg('app_version', [''], ie_key=TikTokIE.ie_key())[0]
manifest_app_version = self._configuration_arg('manifest_app_version', [''], ie_key=TikTokIE.ie_key())[0]
if app_version and manifest_app_version:
self._WORKING_APP_VERSION = (app_version, manifest_app_version)
self.write_debug('Imported app version combo from extractor arguments')
elif app_version or manifest_app_version:
self.report_warning('Only one of the two required version params are passed as extractor arguments', only_once=True)
if self._WORKING_APP_VERSION:
app_version, manifest_app_version = self._WORKING_APP_VERSION
real_query = self._build_api_query(query, app_version, manifest_app_version)
return self._call_api_impl(ep, real_query, manifest_app_version, video_id, fatal, note, errnote)
for count, (app_version, manifest_app_version) in enumerate(self._APP_VERSIONS, start=1):
real_query = self._build_api_query(query, app_version, manifest_app_version)
try:
res = self._call_api_impl(ep, real_query, manifest_app_version, video_id, fatal, note, errnote)
self._WORKING_APP_VERSION = (app_version, manifest_app_version)
return res
except ExtractorError as e:
if isinstance(e.cause, json.JSONDecodeError) and e.cause.pos == 0:
if count == len(self._APP_VERSIONS):
if fatal:
raise e
else:
self.report_warning(str(e.cause or e.msg))
return
self.report_warning('%s. Retrying... (attempt %s of %s)' % (str(e.cause or e.msg), count, len(self._APP_VERSIONS)))
continue
raise e
def _extract_aweme_app(self, aweme_id):
feed_list = self._call_api(
'feed', {'aweme_id': aweme_id}, aweme_id, note='Downloading video feed',
errnote='Unable to download video feed').get('aweme_list') or []
aweme_detail = next((aweme for aweme in feed_list if str(aweme.get('aweme_id')) == aweme_id), None)
if not aweme_detail:
raise ExtractorError('Unable to find video in feed', video_id=aweme_id)
return self._parse_aweme_video_app(aweme_detail)
def _get_subtitles(self, aweme_detail, aweme_id):
# TODO: Extract text positioning info
subtitles = {}
# aweme/detail endpoint subs
captions_info = traverse_obj(
aweme_detail, ('interaction_stickers', ..., 'auto_video_caption_info', 'auto_captions', ...), expected_type=dict)
for caption in captions_info:
caption_url = traverse_obj(caption, ('url', 'url_list', ...), expected_type=url_or_none, get_all=False)
if not caption_url:
continue
caption_json = self._download_json(
caption_url, aweme_id, note='Downloading captions', errnote='Unable to download captions', fatal=False)
if not caption_json:
continue
subtitles.setdefault(caption.get('language', 'en'), []).append({
'ext': 'srt',
'data': '\n\n'.join(
f'{i + 1}\n{srt_subtitles_timecode(line["start_time"] / 1000)} --> {srt_subtitles_timecode(line["end_time"] / 1000)}\n{line["text"]}'
for i, line in enumerate(caption_json['utterances']) if line.get('text'))
})
# feed endpoint subs
if not subtitles:
for caption in traverse_obj(aweme_detail, ('video', 'cla_info', 'caption_infos', ...), expected_type=dict):
if not caption.get('url'):
continue
subtitles.setdefault(caption.get('lang') or 'en', []).append({
'ext': remove_start(caption.get('caption_format'), 'web'),
'url': caption['url'],
})
# webpage subs
if not subtitles:
for caption in traverse_obj(aweme_detail, ('video', 'subtitleInfos', ...), expected_type=dict):
if not caption.get('Url'):
continue
subtitles.setdefault(caption.get('LanguageCodeName') or 'en', []).append({
'ext': remove_start(caption.get('Format'), 'web'),
'url': caption['Url'],
})
return subtitles
def _parse_aweme_video_app(self, aweme_detail):
aweme_id = aweme_detail['aweme_id']
video_info = aweme_detail['video']
def parse_url_key(url_key):
format_id, codec, res, bitrate = self._search_regex(
r'v[^_]+_(?P<id>(?P<codec>[^_]+)_(?P<res>\d+p)_(?P<bitrate>\d+))', url_key,
'url key', default=(None, None, None, None), group=('id', 'codec', 'res', 'bitrate'))
if not format_id:
return {}, None
return {
'format_id': format_id,
'vcodec': 'h265' if codec == 'bytevc1' else codec,
'tbr': int_or_none(bitrate, scale=1000) or None,
'quality': qualities(self.QUALITIES)(res),
}, res
known_resolutions = {}
def audio_meta(url):
ext = determine_ext(url, default_ext='m4a')
return {
'format_note': 'Music track',
'ext': ext,
'acodec': 'aac' if ext == 'm4a' else ext,
'vcodec': 'none',
'width': None,
'height': None,
} if ext == 'mp3' or '-music-' in url else {}
def extract_addr(addr, add_meta={}):
parsed_meta, res = parse_url_key(addr.get('url_key', ''))
if res:
known_resolutions.setdefault(res, {}).setdefault('height', int_or_none(addr.get('height')))
known_resolutions[res].setdefault('width', int_or_none(addr.get('width')))
parsed_meta.update(known_resolutions.get(res, {}))
add_meta.setdefault('height', int_or_none(res[:-1]))
return [{
'url': url,
'filesize': int_or_none(addr.get('data_size')),
'ext': 'mp4',
'acodec': 'aac',
'source_preference': -2 if 'aweme/v1' in url else -1, # Downloads from API might get blocked
**add_meta, **parsed_meta,
'format_note': join_nonempty(
add_meta.get('format_note'), '(API)' if 'aweme/v1' in url else None, delim=' '),
**audio_meta(url),
} for url in addr.get('url_list') or []]
# Hack: Add direct video links first to prioritize them when removing duplicate formats
formats = []
width = int_or_none(video_info.get('width'))
height = int_or_none(video_info.get('height'))
if video_info.get('play_addr'):
formats.extend(extract_addr(video_info['play_addr'], {
'format_id': 'play_addr',
'format_note': 'Direct video',
'vcodec': 'h265' if traverse_obj(
video_info, 'is_bytevc1', 'is_h265') else 'h264', # TODO: Check for "direct iOS" videos, like https://www.tiktok.com/@cookierun_dev/video/7039716639834656002
'width': width,
'height': height,
}))
if video_info.get('download_addr'):
download_addr = video_info['download_addr']
dl_width = int_or_none(download_addr.get('width'))
formats.extend(extract_addr(download_addr, {
'format_id': 'download_addr',
'format_note': 'Download video%s' % (', watermarked' if video_info.get('has_watermark') else ''),
'vcodec': 'h264',
'width': dl_width or width,
'height': try_call(lambda: int(dl_width / 0.5625)) or height, # download_addr['height'] is wrong
'preference': -2 if video_info.get('has_watermark') else -1,
}))
if video_info.get('play_addr_h264'):
formats.extend(extract_addr(video_info['play_addr_h264'], {
'format_id': 'play_addr_h264',
'format_note': 'Direct video',
'vcodec': 'h264',
}))
if video_info.get('play_addr_bytevc1'):
formats.extend(extract_addr(video_info['play_addr_bytevc1'], {
'format_id': 'play_addr_bytevc1',
'format_note': 'Direct video',
'vcodec': 'h265',
}))
for bitrate in video_info.get('bit_rate', []):
if bitrate.get('play_addr'):
formats.extend(extract_addr(bitrate['play_addr'], {
'format_id': bitrate.get('gear_name'),
'format_note': 'Playback video',
'tbr': try_get(bitrate, lambda x: x['bit_rate'] / 1000),
'vcodec': 'h265' if traverse_obj(
bitrate, 'is_bytevc1', 'is_h265') else 'h264',
'fps': bitrate.get('FPS'),
}))
self._remove_duplicate_formats(formats)
auth_cookie = self._get_cookies(self._WEBPAGE_HOST).get('sid_tt')
if auth_cookie:
for f in formats:
self._set_cookie(compat_urllib_parse_urlparse(f['url']).hostname, 'sid_tt', auth_cookie.value)
thumbnails = []
for cover_id in ('cover', 'ai_dynamic_cover', 'animated_cover', 'ai_dynamic_cover_bak',
'origin_cover', 'dynamic_cover'):
for cover_url in traverse_obj(video_info, (cover_id, 'url_list', ...)):
thumbnails.append({
'id': cover_id,
'url': cover_url,
})
stats_info = aweme_detail.get('statistics') or {}
author_info = aweme_detail.get('author') or {}
music_info = aweme_detail.get('music') or {}
user_url = self._UPLOADER_URL_FORMAT % (traverse_obj(author_info,
'sec_uid', 'id', 'uid', 'unique_id',
expected_type=str_or_none, get_all=False))
labels = traverse_obj(aweme_detail, ('hybrid_label', ..., 'text'), expected_type=str)
contained_music_track = traverse_obj(
music_info, ('matched_song', 'title'), ('matched_pgc_sound', 'title'), expected_type=str)
contained_music_author = traverse_obj(
music_info, ('matched_song', 'author'), ('matched_pgc_sound', 'author'), 'author', expected_type=str)
is_generic_og_trackname = music_info.get('is_original_sound') and music_info.get('title') == 'original sound - %s' % music_info.get('owner_handle')
if is_generic_og_trackname:
music_track, music_author = contained_music_track or 'original sound', contained_music_author
else:
music_track, music_author = music_info.get('title'), music_info.get('author')
return {
'id': aweme_id,
**traverse_obj(aweme_detail, {
'title': ('desc', {str}),
'description': ('desc', {str}),
'timestamp': ('create_time', {int_or_none}),
}),
**traverse_obj(stats_info, {
'view_count': 'play_count',
'like_count': 'digg_count',
'repost_count': 'share_count',
'comment_count': 'comment_count',
}, expected_type=int_or_none),
**traverse_obj(author_info, {
'uploader': 'unique_id',
'uploader_id': 'uid',
'creator': 'nickname',
'channel_id': 'sec_uid',
}, expected_type=str_or_none),
'uploader_url': user_url,
'track': music_track,
'album': str_or_none(music_info.get('album')) or None,
'artist': music_author or None,
'formats': formats,
'subtitles': self.extract_subtitles(aweme_detail, aweme_id),
'thumbnails': thumbnails,
'duration': int_or_none(traverse_obj(video_info, 'duration', ('download_addr', 'duration')), scale=1000),
'availability': self._availability(
is_private='Private' in labels,
needs_subscription='Friends only' in labels,
is_unlisted='Followers only' in labels),
'_format_sort_fields': ('quality', 'codec', 'size', 'br'),
}
def _parse_aweme_video_web(self, aweme_detail, webpage_url, video_id):
video_info = aweme_detail['video']
author_info = traverse_obj(aweme_detail, 'authorInfo', 'author', expected_type=dict, default={})
music_info = aweme_detail.get('music') or {}
stats_info = aweme_detail.get('stats') or {}
channel_id = traverse_obj(author_info or aweme_detail, (('authorSecId', 'secUid'), {str}), get_all=False)
user_url = self._UPLOADER_URL_FORMAT % channel_id if channel_id else None
formats = []
width = int_or_none(video_info.get('width'))
height = int_or_none(video_info.get('height'))
for play_url in traverse_obj(video_info, ('playAddr', ((..., 'src'), None), {url_or_none})):
formats.append({
'url': self._proto_relative_url(play_url),
'ext': 'mp4',
'width': width,
'height': height,
})
for download_url in traverse_obj(video_info, (('downloadAddr', ('download', 'url')), {url_or_none})):
formats.append({
'format_id': 'download',
'url': self._proto_relative_url(download_url),
'ext': 'mp4',
'width': width,
'height': height,
})
self._remove_duplicate_formats(formats)
thumbnails = []
for thumb_url in traverse_obj(aweme_detail, (
(None, 'video'), ('thumbnail', 'cover', 'dynamicCover', 'originCover'), {url_or_none})):
thumbnails.append({
'url': self._proto_relative_url(thumb_url),
'width': width,
'height': height,
})
return {
'id': video_id,
**traverse_obj(aweme_detail, {
'title': ('desc', {str}),
'description': ('desc', {str}),
'duration': ('video', 'duration', {int_or_none}),
'timestamp': ('createTime', {int_or_none}),
}),
**traverse_obj(author_info or aweme_detail, {
'creator': ('nickname', {str}),
'uploader': (('uniqueId', 'author'), {str}),
'uploader_id': (('authorId', 'uid', 'id'), {str_or_none}),
}, get_all=False),
**traverse_obj(stats_info, {
'view_count': 'playCount',
'like_count': 'diggCount',
'repost_count': 'shareCount',
'comment_count': 'commentCount',
}, expected_type=int_or_none),
**traverse_obj(music_info, {
'track': 'title',
'album': ('album', {lambda x: x or None}),
'artist': 'authorName',
}, expected_type=str),
'channel_id': channel_id,
'uploader_url': user_url,
'formats': formats,
'thumbnails': thumbnails,
'http_headers': {
'Referer': webpage_url,
}
}
class TikTokIE(TikTokBaseIE):
_VALID_URL = r'https?://www\.tiktok\.com/(?:embed|@(?P<user_id>[\w\.-]+)?/video)/(?P<id>\d+)'
_EMBED_REGEX = [rf'<(?:script|iframe)[^>]+\bsrc=(["\'])(?P<url>{_VALID_URL})']
_TESTS = [{
'url': 'https://www.tiktok.com/@leenabhushan/video/6748451240264420610',
'md5': '736bb7a466c6f0a6afeb597da1e6f5b7',
'info_dict': {
'id': '6748451240264420610',
'ext': 'mp4',
'title': '#jassmanak #lehanga #leenabhushan',
'description': '#jassmanak #lehanga #leenabhushan',
'duration': 13,
'height': 1024,
'width': 576,
'uploader': 'leenabhushan',
'uploader_id': '6691488002098119685',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAA_Eb4t1vodM1IuTy_cvp9CY22RAb59xqrO0Xtz9CYQJvgXaDvZxYnZYRzDWhhgJmy',
'creator': 'facestoriesbyleenabh',
'thumbnail': r're:^https?://[\w\/\.\-]+(~[\w\-]+\.image)?',
'upload_date': '20191016',
'timestamp': 1571246252,
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
'artist': 'Ysrbeats',
'album': 'Lehanga',
'track': 'Lehanga',
},
'skip': '404 Not Found',
}, {
'url': 'https://www.tiktok.com/@patroxofficial/video/6742501081818877190?langCountry=en',
'md5': '6f3cf8cdd9b28cb8363fe0a9a160695b',
'info_dict': {
'id': '6742501081818877190',
'ext': 'mp4',
'title': 'md5:5e2a23877420bb85ce6521dbee39ba94',
'description': 'md5:5e2a23877420bb85ce6521dbee39ba94',
'duration': 27,
'height': 960,
'width': 540,
'uploader': 'patrox',
'uploader_id': '18702747',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAAiFnldaILebi5heDoVU6bn4jBWWycX6-9U3xuNPqZ8Ws',
'channel_id': 'MS4wLjABAAAAiFnldaILebi5heDoVU6bn4jBWWycX6-9U3xuNPqZ8Ws',
'creator': 'patroX',
'thumbnail': r're:^https?://[\w\/\.\-]+(~[\w\-]+\.image)?',
'upload_date': '20190930',
'timestamp': 1569860870,
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
'artist': 'Evan Todd, Jessica Keenan Wynn, Alice Lee, Barrett Wilbert Weed & Jon Eidson',
'track': 'Big Fun',
},
}, {
# Banned audio, only available on the app
'url': 'https://www.tiktok.com/@barudakhb_/video/6984138651336838402',
'info_dict': {
'id': '6984138651336838402',
'ext': 'mp4',
'title': 'Balas @yolaaftwsr hayu yu ? #SquadRandom_ 🔥',
'description': 'Balas @yolaaftwsr hayu yu ? #SquadRandom_ 🔥',
'uploader': 'barudakhb_',
'creator': 'md5:29f238c49bc0c176cb3cef1a9cea9fa6',
'uploader_id': '6974687867511718913',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAAbhBwQC-R1iKoix6jDFsF-vBdfx2ABoDjaZrM9fX6arU3w71q3cOWgWuTXn1soZ7d',
'channel_id': 'MS4wLjABAAAAbhBwQC-R1iKoix6jDFsF-vBdfx2ABoDjaZrM9fX6arU3w71q3cOWgWuTXn1soZ7d',
'track': 'Boka Dance',
'artist': 'md5:29f238c49bc0c176cb3cef1a9cea9fa6',
'timestamp': 1626121503,
'duration': 18,
'thumbnail': r're:^https?://[\w\/\.\-]+(~[\w\-]+\.image)?',
'upload_date': '20210712',
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
},
}, {
# Sponsored video, only available with feed workaround
'url': 'https://www.tiktok.com/@MS4wLjABAAAATh8Vewkn0LYM7Fo03iec3qKdeCUOcBIouRk1mkiag6h3o_pQu_dUXvZ2EZlGST7_/video/7042692929109986561',
'info_dict': {
'id': '7042692929109986561',
'ext': 'mp4',
'title': 'Slap and Run!',
'description': 'Slap and Run!',
'uploader': 'user440922249',
'creator': 'Slap And Run',
'uploader_id': '7036055384943690754',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAATh8Vewkn0LYM7Fo03iec3qKdeCUOcBIouRk1mkiag6h3o_pQu_dUXvZ2EZlGST7_',
'channel_id': 'MS4wLjABAAAATh8Vewkn0LYM7Fo03iec3qKdeCUOcBIouRk1mkiag6h3o_pQu_dUXvZ2EZlGST7_',
'track': 'Promoted Music',
'timestamp': 1639754738,
'duration': 30,
'thumbnail': r're:^https?://[\w\/\.\-]+(~[\w\-]+\.image)?',
'upload_date': '20211217',
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
},
'params': {'skip_download': True}, # XXX: unable to download video data: HTTP Error 403: Forbidden
}, {
# Video without title and description
'url': 'https://www.tiktok.com/@pokemonlife22/video/7059698374567611694',
'info_dict': {
'id': '7059698374567611694',
'ext': 'mp4',
'title': 'TikTok video #7059698374567611694',
'description': '',
'uploader': 'pokemonlife22',
'creator': 'Pokemon',
'uploader_id': '6820838815978423302',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAA0tF1nBwQVVMyrGu3CqttkNgM68Do1OXUFuCY0CRQk8fEtSVDj89HqoqvbSTmUP2W',
'channel_id': 'MS4wLjABAAAA0tF1nBwQVVMyrGu3CqttkNgM68Do1OXUFuCY0CRQk8fEtSVDj89HqoqvbSTmUP2W',
'track': 'original sound',
'timestamp': 1643714123,
'duration': 6,
'thumbnail': r're:^https?://[\w\/\.\-]+(~[\w\-]+\.image)?',
'upload_date': '20220201',
'artist': 'Pokemon',
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
},
}, {
# hydration JSON is sent in a <script> element
'url': 'https://www.tiktok.com/@denidil6/video/7065799023130643713',
'info_dict': {
'id': '7065799023130643713',
'ext': 'mp4',
'title': '#denidil#денидил',
'description': '#denidil#денидил',
'uploader': 'denidil6',
'uploader_id': '7046664115636405250',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAAsvMSzFdQ4ikl3uR2TEJwMBbB2yZh2Zxwhx-WCo3rbDpAharE3GQCrFuJArI3C8QJ',
'artist': 'Holocron Music',
'album': 'Wolf Sounds (1 Hour) Enjoy the Company of the Animal That Is the Majestic King of the Night',
'track': 'Wolf Sounds (1 Hour) Enjoy the Company of the Animal That Is the Majestic King of the Night',
'timestamp': 1645134536,
'duration': 26,
'upload_date': '20220217',
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
},
'skip': 'This video is unavailable',
}, {
# slideshow audio-only mp3 format
'url': 'https://www.tiktok.com/@_le_cannibale_/video/7139980461132074283',
'info_dict': {
'id': '7139980461132074283',
'ext': 'mp3',
'title': 'TikTok video #7139980461132074283',
'description': '',
'creator': 'Antaura',
'uploader': '_le_cannibale_',
'uploader_id': '6604511138619654149',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAAoShJqaw_5gvy48y3azFeFcT4jeyKWbB0VVYasOCt2tTLwjNFIaDcHAM4D-QGXFOP',
'channel_id': 'MS4wLjABAAAAoShJqaw_5gvy48y3azFeFcT4jeyKWbB0VVYasOCt2tTLwjNFIaDcHAM4D-QGXFOP',
'artist': 'nathan !',
'track': 'grahamscott canon',
'upload_date': '20220905',
'timestamp': 1662406249,
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
'thumbnail': r're:^https://.+\.webp',
},
}, {
# only available via web
'url': 'https://www.tiktok.com/@moxypatch/video/7206382937372134662',
'md5': '6aba7fad816e8709ff2c149679ace165',
'info_dict': {
'id': '7206382937372134662',
'ext': 'mp4',
'title': 'md5:1d95c0b96560ca0e8a231af4172b2c0a',
'description': 'md5:1d95c0b96560ca0e8a231af4172b2c0a',
'creator': 'MoxyPatch',
'creators': ['MoxyPatch'],
'uploader': 'moxypatch',
'uploader_id': '7039142049363379205',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAAFhqKnngMHJSsifL0w1vFOP5kn3Ndo1ODp0XuIBkNMBCkALTvwILdpu12g3pTtL4V',
'channel_id': 'MS4wLjABAAAAFhqKnngMHJSsifL0w1vFOP5kn3Ndo1ODp0XuIBkNMBCkALTvwILdpu12g3pTtL4V',
'artists': ['your worst nightmare'],
'track': 'original sound',
'upload_date': '20230303',
'timestamp': 1677866781,
'duration': 10,
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
'thumbnail': r're:^https://.+',
'thumbnails': 'count:3',
},
'expected_warnings': ['Unable to find video in feed'],
}, {
# 1080p format
'url': 'https://www.tiktok.com/@tatemcrae/video/7107337212743830830',
'md5': '982512017a8a917124d5a08c8ae79621',
'info_dict': {
'id': '7107337212743830830',
'ext': 'mp4',
'title': 'new music video 4 dont come backkkk🧸🖤 i hope u enjoy !! @musicontiktok',
'description': 'new music video 4 dont come backkkk🧸🖤 i hope u enjoy !! @musicontiktok',
'uploader': 'tatemcrae',
'uploader_id': '86328792343818240',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAA-0bQT0CqebTRr6I4IkYvMDMKSRSJHLNPBo5HrSklJwyA2psXLSZG5FP-LMNpHnJd',
'channel_id': 'MS4wLjABAAAA-0bQT0CqebTRr6I4IkYvMDMKSRSJHLNPBo5HrSklJwyA2psXLSZG5FP-LMNpHnJd',
'creator': 'tate mcrae',
'artist': 'tate mcrae',
'track': 'original sound',
'upload_date': '20220609',
'timestamp': 1654805899,
'duration': 150,
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
'thumbnail': r're:^https://.+\.webp',
},
'skip': 'Unavailable via feed API, no formats available via web',
}, {
# Slideshow, audio-only m4a format
'url': 'https://www.tiktok.com/@hara_yoimiya/video/7253412088251534594',
'md5': '2ff8fe0174db2dbf49c597a7bef4e47d',
'info_dict': {
'id': '7253412088251534594',
'ext': 'm4a',
'title': 'я ред флаг простите #переписка #щитпост #тревожныйтиппривязанности #рекомендации ',
'description': 'я ред флаг простите #переписка #щитпост #тревожныйтиппривязанности #рекомендации ',
'uploader': 'hara_yoimiya',
'uploader_id': '6582536342634676230',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAAIAlDxriiPWLE-p8p1R_0Bx8qWKfi-7zwmGhzU8Mv25W8sNxjfIKrol31qTczzuLB',
'channel_id': 'MS4wLjABAAAAIAlDxriiPWLE-p8p1R_0Bx8qWKfi-7zwmGhzU8Mv25W8sNxjfIKrol31qTczzuLB',
'creator': 'лампочка',
'artist': 'Øneheart',
'album': 'watching the stars',
'track': 'watching the stars',
'upload_date': '20230708',
'timestamp': 1688816612,
'view_count': int,
'like_count': int,
'comment_count': int,
'repost_count': int,
'thumbnail': r're:^https://.+\.webp',
},
}, {
# Auto-captions available
'url': 'https://www.tiktok.com/@hankgreen1/video/7047596209028074758',
'only_matching': True
}]
def _real_extract(self, url):
video_id, user_id = self._match_valid_url(url).group('id', 'user_id')
try:
return self._extract_aweme_app(video_id)
except ExtractorError as e:
e.expected = True
self.report_warning(f'{e}; trying with webpage')
url = self._create_url(user_id, video_id)
webpage = self._download_webpage(url, video_id, headers={'User-Agent': 'Mozilla/5.0'})
if universal_data := self._get_universal_data(webpage, video_id):
self.write_debug('Found universal data for rehydration')
status = traverse_obj(universal_data, ('webapp.video-detail', 'statusCode', {int})) or 0
video_data = traverse_obj(universal_data, ('webapp.video-detail', 'itemInfo', 'itemStruct', {dict}))
elif sigi_data := self._get_sigi_state(webpage, video_id):
self.write_debug('Found sigi state data')
status = traverse_obj(sigi_data, ('VideoPage', 'statusCode', {int})) or 0
video_data = traverse_obj(sigi_data, ('ItemModule', video_id, {dict}))
elif next_data := self._search_nextjs_data(webpage, video_id, default='{}'):
self.write_debug('Found next.js data')
status = traverse_obj(next_data, ('props', 'pageProps', 'statusCode', {int})) or 0
video_data = traverse_obj(next_data, ('props', 'pageProps', 'itemInfo', 'itemStruct', {dict}))
else:
raise ExtractorError('Unable to extract webpage video data')
if video_data and status == 0:
return self._parse_aweme_video_web(video_data, url, video_id)
elif status == 10216:
raise ExtractorError('This video is private', expected=True)
raise ExtractorError(f'Video not available, status code {status}', video_id=video_id)
class TikTokUserIE(TikTokBaseIE):
IE_NAME = 'tiktok:user'
_VALID_URL = r'https?://(?:www\.)?tiktok\.com/@(?P<id>[\w\.-]+)/?(?:$|[#?])'
_WORKING = False
_TESTS = [{
'url': 'https://tiktok.com/@corgibobaa?lang=en',
'playlist_mincount': 45,
'info_dict': {
'id': '6935371178089399301',
'title': 'corgibobaa',
'thumbnail': r're:https://.+_1080x1080\.webp'
},
'expected_warnings': ['Retrying']
}, {
'url': 'https://www.tiktok.com/@6820838815978423302',
'playlist_mincount': 5,
'info_dict': {
'id': '6820838815978423302',
'title': '6820838815978423302',
'thumbnail': r're:https://.+_1080x1080\.webp'
},
'expected_warnings': ['Retrying']
}, {
'url': 'https://www.tiktok.com/@meme',
'playlist_mincount': 593,
'info_dict': {
'id': '79005827461758976',
'title': 'meme',
'thumbnail': r're:https://.+_1080x1080\.webp'
},
'expected_warnings': ['Retrying']
}]
r''' # TODO: Fix by adding _signature to api_url
def _entries(self, webpage, user_id, username):
secuid = self._search_regex(r'\"secUid\":\"(?P<secUid>[^\"]+)', webpage, username)
verifyfp_cookie = self._get_cookies('https://www.tiktok.com').get('s_v_web_id')
if not verifyfp_cookie:
raise ExtractorError('Improper cookies (missing s_v_web_id).', expected=True)
api_url = f'https://m.tiktok.com/api/post/item_list/?aid=1988&cookie_enabled=true&count=30&verifyFp={verifyfp_cookie.value}&secUid={secuid}&cursor='
cursor = '0'
for page in itertools.count():
data_json = self._download_json(api_url + cursor, username, note='Downloading Page %d' % page)
for video in data_json.get('itemList', []):
video_id = video['id']
video_url = f'https://www.tiktok.com/@{user_id}/video/{video_id}'
yield self._url_result(video_url, 'TikTok', video_id, str_or_none(video.get('desc')))
if not data_json.get('hasMore'):
break
cursor = data_json['cursor']
'''
def _video_entries_api(self, webpage, user_id, username):
query = {
'user_id': user_id,
'count': 21,
'max_cursor': 0,
'min_cursor': 0,
'retry_type': 'no_retry',
'device_id': ''.join(random.choices(string.digits, k=19)), # Some endpoints don't like randomized device_id, so it isn't directly set in _call_api.
}
for page in itertools.count(1):
for retry in self.RetryManager():
try:
post_list = self._call_api(
'aweme/post', query, username, note=f'Downloading user video list page {page}',
errnote='Unable to download user video list')
except ExtractorError as e:
if isinstance(e.cause, json.JSONDecodeError) and e.cause.pos == 0:
retry.error = e
continue
raise
yield from post_list.get('aweme_list', [])
if not post_list.get('has_more'):
break
query['max_cursor'] = post_list['max_cursor']
def _entries_api(self, user_id, videos):
for video in videos:
yield {
**self._parse_aweme_video_app(video),
'extractor_key': TikTokIE.ie_key(),
'extractor': 'TikTok',
'webpage_url': f'https://tiktok.com/@{user_id}/video/{video["aweme_id"]}',
}
def _real_extract(self, url):
user_name = self._match_id(url)
webpage = self._download_webpage(url, user_name, headers={
'User-Agent': 'facebookexternalhit/1.1 (+http://www.facebook.com/externalhit_uatext.php)'
})
user_id = self._html_search_regex(r'snssdk\d*://user/profile/(\d+)', webpage, 'user ID', default=None) or user_name
videos = LazyList(self._video_entries_api(webpage, user_id, user_name))
thumbnail = traverse_obj(videos, (0, 'author', 'avatar_larger', 'url_list', 0))
return self.playlist_result(self._entries_api(user_id, videos), user_id, user_name, thumbnail=thumbnail)
class TikTokBaseListIE(TikTokBaseIE): # XXX: Conventionally, base classes should end with BaseIE/InfoExtractor
def _entries(self, list_id, display_id):
query = {
self._QUERY_NAME: list_id,
'cursor': 0,
'count': 20,
'type': 5,
'device_id': ''.join(random.choices(string.digits, k=19))
}
for page in itertools.count(1):
for retry in self.RetryManager():
try:
post_list = self._call_api(
self._API_ENDPOINT, query, display_id, note=f'Downloading video list page {page}',
errnote='Unable to download video list')
except ExtractorError as e:
if isinstance(e.cause, json.JSONDecodeError) and e.cause.pos == 0:
retry.error = e
continue
raise
for video in post_list.get('aweme_list', []):
yield {
**self._parse_aweme_video_app(video),
'extractor_key': TikTokIE.ie_key(),
'extractor': 'TikTok',
'webpage_url': f'https://tiktok.com/@_/video/{video["aweme_id"]}',
}
if not post_list.get('has_more'):
break
query['cursor'] = post_list['cursor']
def _real_extract(self, url):
list_id = self._match_id(url)
return self.playlist_result(self._entries(list_id, list_id), list_id)
class TikTokSoundIE(TikTokBaseListIE):
IE_NAME = 'tiktok:sound'
_VALID_URL = r'https?://(?:www\.)?tiktok\.com/music/[\w\.-]+-(?P<id>[\d]+)[/?#&]?'
_WORKING = False
_QUERY_NAME = 'music_id'
_API_ENDPOINT = 'music/aweme'
_TESTS = [{
'url': 'https://www.tiktok.com/music/Build-a-Btch-6956990112127585029?lang=en',
'playlist_mincount': 100,
'info_dict': {
'id': '6956990112127585029'
},
'expected_warnings': ['Retrying']
}, {
# Actual entries are less than listed video count
'url': 'https://www.tiktok.com/music/jiefei-soap-remix-7036843036118469381',
'playlist_mincount': 2182,
'info_dict': {
'id': '7036843036118469381'
},
'expected_warnings': ['Retrying']
}]
class TikTokEffectIE(TikTokBaseListIE):
IE_NAME = 'tiktok:effect'
_VALID_URL = r'https?://(?:www\.)?tiktok\.com/sticker/[\w\.-]+-(?P<id>[\d]+)[/?#&]?'
_WORKING = False
_QUERY_NAME = 'sticker_id'
_API_ENDPOINT = 'sticker/aweme'
_TESTS = [{
'url': 'https://www.tiktok.com/sticker/MATERIAL-GWOOORL-1258156',
'playlist_mincount': 100,
'info_dict': {
'id': '1258156',
},
'expected_warnings': ['Retrying']
}, {
# Different entries between mobile and web, depending on region
'url': 'https://www.tiktok.com/sticker/Elf-Friend-479565',
'only_matching': True
}]
class TikTokTagIE(TikTokBaseListIE):
IE_NAME = 'tiktok:tag'
_VALID_URL = r'https?://(?:www\.)?tiktok\.com/tag/(?P<id>[^/?#&]+)'
_WORKING = False
_QUERY_NAME = 'ch_id'
_API_ENDPOINT = 'challenge/aweme'
_TESTS = [{
'url': 'https://tiktok.com/tag/hello2018',
'playlist_mincount': 39,
'info_dict': {
'id': '46294678',
'title': 'hello2018',
},
'expected_warnings': ['Retrying']
}, {
'url': 'https://tiktok.com/tag/fypシ?is_copy_url=0&is_from_webapp=v1',
'only_matching': True
}]
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id, headers={
'User-Agent': 'facebookexternalhit/1.1 (+http://www.facebook.com/externalhit_uatext.php)'
})
tag_id = self._html_search_regex(r'snssdk\d*://challenge/detail/(\d+)', webpage, 'tag ID')
return self.playlist_result(self._entries(tag_id, display_id), tag_id, display_id)
class DouyinIE(TikTokBaseIE):
_VALID_URL = r'https?://(?:www\.)?douyin\.com/video/(?P<id>[0-9]+)'
_TESTS = [{
'url': 'https://www.douyin.com/video/6961737553342991651',
'md5': '9ecce7bc5b302601018ecb2871c63a75',
'info_dict': {
'id': '6961737553342991651',
'ext': 'mp4',
'title': '#杨超越 小小水手带你去远航❤️',
'description': '#杨超越 小小水手带你去远航❤️',
'uploader': '6897520xka',
'uploader_id': '110403406559',
'uploader_url': 'https://www.douyin.com/user/MS4wLjABAAAAEKnfa654JAJ_N5lgZDQluwsxmY0lhfmEYNQBBkwGG98',
'channel_id': 'MS4wLjABAAAAEKnfa654JAJ_N5lgZDQluwsxmY0lhfmEYNQBBkwGG98',
'creator': '杨超越',
'creators': ['杨超越'],
'duration': 19,
'timestamp': 1620905839,
'upload_date': '20210513',
'track': '@杨超越创作的原声',
'artists': ['杨超越'],
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
'thumbnail': r're:https?://.+\.jpe?g',
},
}, {
'url': 'https://www.douyin.com/video/6982497745948921092',
'md5': '15c5e660b7048af3707304e3cc02bbb5',
'info_dict': {
'id': '6982497745948921092',
'ext': 'mp4',
'title': '这个夏日和小羊@杨超越 一起遇见白色幻想',
'description': '这个夏日和小羊@杨超越 一起遇见白色幻想',
'uploader': '0731chaoyue',
'uploader_id': '408654318141572',
'uploader_url': 'https://www.douyin.com/user/MS4wLjABAAAAZJpnglcjW2f_CMVcnqA_6oVBXKWMpH0F8LIHuUu8-lA',
'channel_id': 'MS4wLjABAAAAZJpnglcjW2f_CMVcnqA_6oVBXKWMpH0F8LIHuUu8-lA',
'creator': '杨超越工作室',
'creators': ['杨超越工作室'],
'duration': 42,
'timestamp': 1625739481,
'upload_date': '20210708',
'track': '@杨超越工作室创作的原声',
'artists': ['杨超越工作室'],
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
'thumbnail': r're:https?://.+\.jpe?g',
},
}, {
'url': 'https://www.douyin.com/video/6953975910773099811',
'md5': '0e6443758b8355db9a3c34864a4276be',
'info_dict': {
'id': '6953975910773099811',
'ext': 'mp4',
'title': '#一起看海 出现在你的夏日里',
'description': '#一起看海 出现在你的夏日里',
'uploader': '6897520xka',
'uploader_id': '110403406559',
'uploader_url': 'https://www.douyin.com/user/MS4wLjABAAAAEKnfa654JAJ_N5lgZDQluwsxmY0lhfmEYNQBBkwGG98',
'channel_id': 'MS4wLjABAAAAEKnfa654JAJ_N5lgZDQluwsxmY0lhfmEYNQBBkwGG98',
'creator': '杨超越',
'creators': ['杨超越'],
'duration': 17,
'timestamp': 1619098692,
'upload_date': '20210422',
'track': '@杨超越创作的原声',
'artists': ['杨超越'],
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
'thumbnail': r're:https?://.+\.jpe?g',
},
}, {
'url': 'https://www.douyin.com/video/6950251282489675042',
'md5': 'b4db86aec367ef810ddd38b1737d2fed',
'info_dict': {
'id': '6950251282489675042',
'ext': 'mp4',
'title': '哈哈哈,成功了哈哈哈哈哈哈',
'uploader': '杨超越',
'upload_date': '20210412',
'timestamp': 1618231483,
'uploader_id': '110403406559',
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
},
'skip': 'No longer available',
}, {
'url': 'https://www.douyin.com/video/6963263655114722595',
'md5': '1440bcf59d8700f8e014da073a4dfea8',
'info_dict': {
'id': '6963263655114722595',
'ext': 'mp4',
'title': '#哪个爱豆的105度最甜 换个角度看看我哈哈',
'description': '#哪个爱豆的105度最甜 换个角度看看我哈哈',
'uploader': '6897520xka',
'uploader_id': '110403406559',
'uploader_url': 'https://www.douyin.com/user/MS4wLjABAAAAEKnfa654JAJ_N5lgZDQluwsxmY0lhfmEYNQBBkwGG98',
'channel_id': 'MS4wLjABAAAAEKnfa654JAJ_N5lgZDQluwsxmY0lhfmEYNQBBkwGG98',
'creator': '杨超越',
'creators': ['杨超越'],
'duration': 15,
'timestamp': 1621261163,
'upload_date': '20210517',
'track': '@杨超越创作的原声',
'artists': ['杨超越'],
'view_count': int,
'like_count': int,
'repost_count': int,
'comment_count': int,
'thumbnail': r're:https?://.+\.jpe?g',
},
}]
_UPLOADER_URL_FORMAT = 'https://www.douyin.com/user/%s'
_WEBPAGE_HOST = 'https://www.douyin.com/'
def _real_extract(self, url):
video_id = self._match_id(url)
detail = traverse_obj(self._download_json(
'https://www.douyin.com/aweme/v1/web/aweme/detail/', video_id,
'Downloading web detail JSON', 'Failed to download web detail JSON',
query={'aweme_id': video_id}, fatal=False), ('aweme_detail', {dict}))
if not detail:
# TODO: Run verification challenge code to generate signature cookies
raise ExtractorError(
'Fresh cookies (not necessarily logged in) are needed',
expected=not self._get_cookies(self._WEBPAGE_HOST).get('s_v_web_id'))
return self._parse_aweme_video_app(detail)
class TikTokVMIE(InfoExtractor):
_VALID_URL = r'https?://(?:(?:vm|vt)\.tiktok\.com|(?:www\.)tiktok\.com/t)/(?P<id>\w+)'
IE_NAME = 'vm.tiktok'
_TESTS = [{
'url': 'https://www.tiktok.com/t/ZTRC5xgJp',
'info_dict': {
'id': '7170520270497680683',
'ext': 'mp4',
'title': 'md5:c64f6152330c2efe98093ccc8597871c',
'uploader_id': '6687535061741700102',
'upload_date': '20221127',
'view_count': int,
'like_count': int,
'comment_count': int,
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAAObqu3WCTXxmw2xwZ3iLEHnEecEIw7ks6rxWqOqOhaPja9BI7gqUQnjw8_5FSoDXX',
'album': 'Wave of Mutilation: Best of Pixies',
'thumbnail': r're:https://.+\.webp.*',
'duration': 5,
'timestamp': 1669516858,
'repost_count': int,
'artist': 'Pixies',
'track': 'Where Is My Mind?',
'description': 'md5:c64f6152330c2efe98093ccc8597871c',
'uploader': 'sigmachaddeus',
'creator': 'SigmaChad',
},
}, {
'url': 'https://vm.tiktok.com/ZTR45GpSF/',
'info_dict': {
'id': '7106798200794926362',
'ext': 'mp4',
'title': 'md5:edc3e7ea587847f8537468f2fe51d074',
'uploader_id': '6997695878846268418',
'upload_date': '20220608',
'view_count': int,
'like_count': int,
'comment_count': int,
'thumbnail': r're:https://.+\.webp.*',
'uploader_url': 'https://www.tiktok.com/@MS4wLjABAAAAdZ_NcPPgMneaGrW0hN8O_J_bwLshwNNERRF5DxOw2HKIzk0kdlLrR8RkVl1ksrMO',
'duration': 29,
'timestamp': 1654680400,
'repost_count': int,
'artist': 'Akihitoko',
'track': 'original sound',
'description': 'md5:edc3e7ea587847f8537468f2fe51d074',
'uploader': 'akihitoko1',
'creator': 'Akihitoko',
},
}, {
'url': 'https://vt.tiktok.com/ZSe4FqkKd',
'only_matching': True,
}]
def _real_extract(self, url):
new_url = self._request_webpage(
HEADRequest(url), self._match_id(url), headers={'User-Agent': 'facebookexternalhit/1.1'}).url
if self.suitable(new_url): # Prevent infinite loop in case redirect fails
raise UnsupportedError(new_url)
return self.url_result(new_url)
class TikTokLiveIE(TikTokBaseIE):
_VALID_URL = r'''(?x)https?://(?:
(?:www\.)?tiktok\.com/@(?P<uploader>[\w.-]+)/live|
m\.tiktok\.com/share/live/(?P<id>\d+)
)'''
IE_NAME = 'tiktok:live'
_TESTS = [{
'url': 'https://www.tiktok.com/@weathernewslive/live',
'info_dict': {
'id': '7210809319192726273',
'ext': 'mp4',
'title': r're:ウェザーニュースLiVE[\d\s:-]*',
'creator': 'ウェザーニュースLiVE',
'uploader': 'weathernewslive',
'uploader_id': '6621496731283095554',
'uploader_url': 'https://www.tiktok.com/@weathernewslive',
'live_status': 'is_live',
'concurrent_view_count': int,
},
'params': {'skip_download': 'm3u8'},
}, {
'url': 'https://www.tiktok.com/@pilarmagenta/live',
'info_dict': {
'id': '7209423610325322522',
'ext': 'mp4',
'title': str,
'creator': 'Pilarmagenta',
'uploader': 'pilarmagenta',
'uploader_id': '6624846890674683909',
'uploader_url': 'https://www.tiktok.com/@pilarmagenta',
'live_status': 'is_live',
'concurrent_view_count': int,
},
'skip': 'Livestream',
}, {
'url': 'https://m.tiktok.com/share/live/7209423610325322522/?language=en',
'only_matching': True,
}, {
'url': 'https://www.tiktok.com/@iris04201/live',
'only_matching': True,
}]
def _call_api(self, url, param, room_id, uploader, key=None):
response = traverse_obj(self._download_json(
url, room_id, fatal=False, query={
'aid': '1988',
param: room_id,
}), (key, {dict}), default={})
# status == 2 if live else 4
if int_or_none(response.get('status')) == 2:
return response
# If room_id is obtained via mobile share URL and cannot be refreshed, do not wait for live
elif not uploader:
raise ExtractorError('This livestream has ended', expected=True)
raise UserNotLive(video_id=uploader)
def _real_extract(self, url):
uploader, room_id = self._match_valid_url(url).group('uploader', 'id')
webpage = self._download_webpage(
url, uploader or room_id, headers={'User-Agent': 'Mozilla/5.0'}, fatal=not room_id)
if webpage:
data = self._get_sigi_state(webpage, uploader or room_id)
room_id = (traverse_obj(data, ('UserModule', 'users', ..., 'roomId', {str_or_none}), get_all=False)
or self._search_regex(r'snssdk\d*://live\?room_id=(\d+)', webpage, 'room ID', default=None)
or room_id)
uploader = uploader or traverse_obj(
data, ('LiveRoom', 'liveRoomUserInfo', 'user', 'uniqueId'),
('UserModule', 'users', ..., 'uniqueId'), get_all=False, expected_type=str)
if not room_id:
raise UserNotLive(video_id=uploader)
formats = []
live_info = self._call_api(
'https://webcast.tiktok.com/webcast/room/info', 'room_id', room_id, uploader, key='data')
get_quality = qualities(('SD1', 'ld', 'SD2', 'sd', 'HD1', 'hd', 'FULL_HD1', 'uhd', 'ORIGION', 'origin'))
parse_inner = lambda x: self._parse_json(x, None)
for quality, stream in traverse_obj(live_info, (
'stream_url', 'live_core_sdk_data', 'pull_data', 'stream_data',
{parse_inner}, 'data', {dict}), default={}).items():
sdk_params = traverse_obj(stream, ('main', 'sdk_params', {parse_inner}, {
'vcodec': ('VCodec', {str}),
'tbr': ('vbitrate', {lambda x: int_or_none(x, 1000)}),
'resolution': ('resolution', {lambda x: re.match(r'(?i)\d+x\d+|\d+p', x).group().lower()}),
}))
flv_url = traverse_obj(stream, ('main', 'flv', {url_or_none}))
if flv_url:
formats.append({
'url': flv_url,
'ext': 'flv',
'format_id': f'flv-{quality}',
'quality': get_quality(quality),
**sdk_params,
})
hls_url = traverse_obj(stream, ('main', 'hls', {url_or_none}))
if hls_url:
formats.append({
'url': hls_url,
'ext': 'mp4',
'protocol': 'm3u8_native',
'format_id': f'hls-{quality}',
'quality': get_quality(quality),
**sdk_params,
})
def get_vcodec(*keys):
return traverse_obj(live_info, (
'stream_url', *keys, {parse_inner}, 'VCodec', {str}))
for stream in ('hls', 'rtmp'):
stream_url = traverse_obj(live_info, ('stream_url', f'{stream}_pull_url', {url_or_none}))
if stream_url:
formats.append({
'url': stream_url,
'ext': 'mp4' if stream == 'hls' else 'flv',
'protocol': 'm3u8_native' if stream == 'hls' else 'https',
'format_id': f'{stream}-pull',
'vcodec': get_vcodec(f'{stream}_pull_url_params'),
'quality': get_quality('ORIGION'),
})
for f_id, f_url in traverse_obj(live_info, ('stream_url', 'flv_pull_url', {dict}), default={}).items():
if not url_or_none(f_url):
continue
formats.append({
'url': f_url,
'ext': 'flv',
'format_id': f'flv-{f_id}'.lower(),
'vcodec': get_vcodec('flv_pull_url_params', f_id),
'quality': get_quality(f_id),
})
# If uploader is a guest on another's livestream, primary endpoint will not have m3u8 URLs
if not traverse_obj(formats, lambda _, v: v['ext'] == 'mp4'):
live_info = merge_dicts(live_info, self._call_api(
'https://www.tiktok.com/api/live/detail/', 'roomID', room_id, uploader, key='LiveRoomInfo'))
if url_or_none(live_info.get('liveUrl')):
formats.append({
'url': live_info['liveUrl'],
'ext': 'mp4',
'protocol': 'm3u8_native',
'format_id': 'hls-fallback',
'vcodec': 'h264',
'quality': get_quality('origin'),
})
uploader = uploader or traverse_obj(live_info, ('ownerInfo', 'uniqueId'), ('owner', 'display_id'))
return {
'id': room_id,
'uploader': uploader,
'uploader_url': format_field(uploader, None, self._UPLOADER_URL_FORMAT) or None,
'is_live': True,
'formats': formats,
'_format_sort_fields': ('quality', 'ext'),
**traverse_obj(live_info, {
'title': 'title',
'uploader_id': (('ownerInfo', 'owner'), 'id', {str_or_none}),
'creator': (('ownerInfo', 'owner'), 'nickname'),
'concurrent_view_count': (('user_count', ('liveRoomStats', 'userCount')), {int_or_none}),
}, get_all=False),
}