From 4b6d03ed873b04f027b4c7b734b7f8bd3c89eae6 Mon Sep 17 00:00:00 2001 From: Aakash Gajjar Date: Fri, 25 Oct 2019 13:35:54 +0530 Subject: [PATCH] [tiktok] fix extraction --- youtube_dl/extractor/extractors.py | 5 +- youtube_dl/extractor/tiktok.py | 316 ++++++++++++++++++++--------- 2 files changed, 216 insertions(+), 105 deletions(-) diff --git a/youtube_dl/extractor/extractors.py b/youtube_dl/extractor/extractors.py index 1db21529f..abf5bb48d 100644 --- a/youtube_dl/extractor/extractors.py +++ b/youtube_dl/extractor/extractors.py @@ -1151,10 +1151,7 @@ from .thisav import ThisAVIE from .thisoldhouse import ThisOldHouseIE from .threeqsdn import ThreeQSDNIE -from .tiktok import ( - TikTokIE, - TikTokUserIE, -) +from .tiktok import TikTokIE from .tinypic import TinyPicIE from .tmz import ( TMZIE, diff --git a/youtube_dl/extractor/tiktok.py b/youtube_dl/extractor/tiktok.py index 66088b9ab..539fc9ecd 100644 --- a/youtube_dl/extractor/tiktok.py +++ b/youtube_dl/extractor/tiktok.py @@ -1,138 +1,252 @@ # coding: utf-8 from __future__ import unicode_literals +from datetime import datetime +import re from .common import InfoExtractor from ..utils import ( - compat_str, ExtractorError, int_or_none, str_or_none, - try_get, - url_or_none, + try_get ) class TikTokBaseIE(InfoExtractor): - def _extract_aweme(self, data): - video = data['video'] - description = str_or_none(try_get(data, lambda x: x['desc'])) - width = int_or_none(try_get(data, lambda x: video['width'])) - height = int_or_none(try_get(data, lambda x: video['height'])) + def _video_info(self, video_info): + return { + 'id': str_or_none(video_info.get('id')), + 'thumbnail': try_get(video_info, lambda x: x['covers'][0], str) or try_get(video_info, lambda x: x['video']['videoMeta']['cover'][0], str), + 'video_url': try_get(video_info, lambda x: x['video']['urls'][0], str) or video_info.get('video', {}).get('urls', [None])[0], + 'width': try_get(video_info, lambda x: x['video']['videoMeta']['width'], int) or try_get(video_info, lambda x: x['width'], int), + 'height': try_get(video_info, lambda x: x['video']['videoMeta']['height'], int) or try_get(video_info, lambda x: x['height'], int), + 'duration': try_get(video_info, lambda x: x['video']['videoMeta']['duration'], int), + 'description': str_or_none(video_info.get('text')), + 'comment_count': int_or_none(video_info.get('commentCount')), + 'like_count': int_or_none(video_info.get('diggCount')), + 'repost_count': int_or_none(video_info.get('shareCount')), + 'timestamp': str_or_none(video_info.get('createTime')), + 'track_id': str_or_none(video_info.get('musicId')) + } + + def _author_info(self, author_info): + return { + 'uploader': str_or_none(author_info.get('uniqueId')), + 'creator': str_or_none(author_info.get('nickName')), + 'uploader_id': str_or_none(author_info.get('userId')), + 'channel_id': str_or_none(author_info.get('userId')) + } + + def _track_info(self, track_info): + return { + 'track': str_or_none(track_info.get('musicName')), + 'track_id': str_or_none(track_info.get('musicId')), + 'artist': str_or_none(track_info.get('authorName')) + } + + def _share_info(self, share_info): + return { + 'title': str_or_none(share_info.get('title')), + 'description': str_or_none(share_info.get('desc')), + 'image': try_get(share_info, lambda x: x['image'], dict), + 'width': try_get(share_info, lambda x: x['image']['width'], int), + 'height': try_get(share_info, lambda x: x['image']['height'], int), + } + + def _extract_aweme(self, video_data, webpage): + video_info_data = try_get( + video_data, lambda x: x['videoData']['itemInfos'], dict) + author_info_data = try_get( + video_data, lambda x: x['videoData']['authorInfos'], dict) + track_info_data = try_get( + video_data, lambda x: x['videoData']['musicInfos'], dict) + share_info_data = try_get(video_data, lambda x: x['shareMeta'], dict) + + video_info = self._video_info(video_info_data) + author_info = self._author_info(author_info_data) + track_info = self._track_info(track_info_data) + share_info = self._share_info(share_info_data) + + timestamp = int(video_info.get('timestamp')) or 0 + date = str_or_none(datetime.fromtimestamp( + timestamp).strftime('%Y%m%d')) + + thumbnails = [] + thumbnails.append({ + 'url': video_info.get('thumbnail') or self._og_search_thumbnail(webpage), + 'width': video_info.get('width'), + 'height': video_info.get('height') + }) + + description = video_info.get( + 'description') or share_info.get('description') + if description is None: + tags = [] + else: + tags = re.findall(r"#(\w+)", description) - format_urls = set() formats = [] - for format_id in ( - 'play_addr_lowbr', 'play_addr', 'play_addr_h264', - 'download_addr'): - for format in try_get( - video, lambda x: x[format_id]['url_list'], list) or []: - format_url = url_or_none(format) - if not format_url: - continue - if format_url in format_urls: - continue - format_urls.add(format_url) - formats.append({ - 'url': format_url, - 'ext': 'mp4', - 'height': height, - 'width': width, - }) - self._sort_formats(formats) - - thumbnail = url_or_none(try_get( - video, lambda x: x['cover']['url_list'][0], compat_str)) - uploader = try_get(data, lambda x: x['author']['nickname'], compat_str) - timestamp = int_or_none(data.get('create_time')) - comment_count = int_or_none(data.get('comment_count')) or int_or_none( - try_get(data, lambda x: x['statistics']['comment_count'])) - repost_count = int_or_none(try_get( - data, lambda x: x['statistics']['share_count'])) - - aweme_id = data['aweme_id'] + formats.append({ + 'url': video_info.get('video_url') or self._og_search_video_url(webpage), + 'ext': 'mp4', + 'height': video_info.get('height'), + 'width': video_info.get('width'), + }) return { - 'id': aweme_id, - 'title': uploader or aweme_id, + 'artist': track_info.get('artist'), + 'channel_id': author_info.get('channel_id'), + 'channel_url': 'https://www.tiktok.com/@{}'.format(author_info.get('uploader')), + 'comment_count': video_info.get('comment_count'), + 'creator': author_info.get('creator'), 'description': description, - 'thumbnail': thumbnail, - 'uploader': uploader, - 'timestamp': timestamp, - 'comment_count': comment_count, - 'repost_count': repost_count, + 'duration': video_info.get('duration'), 'formats': formats, + 'height': video_info.get('height'), + 'id': video_info.get('id'), + 'like_count': video_info.get('like_count'), + 'playlist_title': share_info.get('title'), + 'playlist_uploader': author_info.get('uploader'), + 'playlist_uploader_id': author_info.get('uploader_id'), + 'repost_count': video_info.get('repost_count'), + 'release_date': date, + 'tags': tags, + 'thumbnail': video_info.get('thumbnail'), + 'thumbnails': thumbnails, + 'timestamp': int(video_info.get('timestamp')), + 'title': share_info.get('title') or self._og_search_title(webpage), + 'track': track_info.get('track'), + 'track_id': track_info.get('track_id'), + 'upload_date': date, + 'uploader': author_info.get('uploader'), + 'uploader_id': author_info.get('uploader_id'), + 'uploader_url': 'https://www.tiktok.com/@{}'.format(author_info.get('uploader')), + 'webpage_url': self._og_search_url(webpage), + 'width': video_info.get('width') } class TikTokIE(TikTokBaseIE): _VALID_URL = r'''(?x) - https?:// - (?: - (?:m\.)?tiktok\.com/v| - (?:www\.)?tiktok\.com/share/video - ) - /(?P\d+) - ''' + https?:// + (?: + (?:www|m)\. + (?:tiktok.com)\/ + (@(?P[\w\.]+))? + (?:v|video|embed|trending)?(?:\/)? + (?:video)?(?:\/)? + (?:\?shareId=)? + ) + (?P[\d]{6,}) + (?:\.html)? + (?:\?.*)? + $ + ''' + _TESTS = [{ - 'url': 'https://m.tiktok.com/v/6606727368545406213.html', - 'md5': 'd584b572e92fcd48888051f238022420', + 'url': 'https://www.tiktok.com/@leenabhushan/video/6748451240264420610', + 'md5': '34a7543afd5a151b0840ba6736fb633b', 'info_dict': { - 'id': '6606727368545406213', + 'id': '6748451240264420610', 'ext': 'mp4', - 'title': 'Zureeal', - 'description': '#bowsette#mario#cosplay#uk#lgbt#gaming#asian#bowsettecosplay', - 'thumbnail': r're:^https?://.*~noop.image', - 'uploader': 'Zureeal', - 'timestamp': 1538248586, - 'upload_date': '20180929', + 'title': 'facestoriesbyleenabh on TikTok', + 'description': 'md5:a9f6c0c44a1ff2249cae610372d0ae95', + 'thumbnail': r're:^https?://[\w\/\.\-]+(~[\w\-]+\.image)?', + 'uploader': 'leenabhushan', + 'timestamp': 1571246252, + 'upload_date': '20191016', 'comment_count': int, 'repost_count': int, + 'like_count': int, + 'playlist_title': 'facestoriesbyleenabh on TikTok', + 'playlist_uploader': 'leenabhushan', + 'playlist_uploader_id': '6691488002098119685', + 'artist': 'Jass Manak', + 'channel_id': '6691488002098119685', + 'channel_url': 'https://www.tiktok.com/@leenabhushan', + 'creator': 'facestoriesbyleenabh', + 'duration': 13, + 'formats': list, + 'height': 1280, + 'release_date': '20191016', + 'tags': list, + 'thumbnails': list, + 'track': 'Lehanga', + 'track_id': '6716465478027447045', + 'uploader_id': '6691488002098119685', + 'uploader_url': r're:https://www.tiktok.com/@leenabhushan', + 'webpage_url': r're:https://www.tiktok.com/@leenabhushan/(video/)?6748451240264420610', + 'width': 720, } }, { - 'url': 'https://www.tiktok.com/share/video/6606727368545406213', - 'only_matching': True, + 'url': 'https://www.tiktok.com/@patroxofficial/video/6742501081818877190?langCountry=en', + 'md5': '06b9800d47d5fe51a19e322dd86e61c9', + 'info_dict': { + 'artist': 'Evan Todd, Jessica Keenan Wynn, Alice Lee, Barrett Wilbert Weed & Jon Eidson', + 'channel_id': '18702747', + 'channel_url': 'https://www.tiktok.com/@patroxofficial', + 'comment_count': int, + 'creator': 'patroX', + 'description': 'md5:5e2a23877420bb85ce6521dbee39ba94', + 'duration': 27, + 'ext': 'mp4', + 'formats': list, + 'height': 960, + 'id': '6742501081818877190', + 'like_count': int, + 'playlist_title': 'patroX on TikTok', + 'playlist_uploader_id': '18702747', + 'playlist_uploader': 'patroxofficial', + 'release_date': '20190930', + 'repost_count': int, + 'tags': list, + 'thumbnail': r're:^https?://[\w\/\.\-]+(~[\w\-]+\.image)?', + 'thumbnails': list, + 'timestamp': 1569860870, + 'title': 'patroX on TikTok', + 'track_id': '209649576000286720', + 'track': 'Big Fun', + 'upload_date': '20190930', + 'uploader_id': '18702747', + 'uploader_url': r're:https://www.tiktok.com/@patroxofficial', + 'uploader': 'patroxofficial', + 'webpage_url': r're:https://www.tiktok.com/@patroxofficial/(video/)?6742501081818877190', + 'width': 540, + } + }, { + 'url': 'https://m.tiktok.com/v/6749869095467945218.html', + 'only_matching': True + }, { + 'url': 'https://www.tiktok.com/@cchelseameow/video/6751181801206729990', + 'only_matching': True + }, { + 'url': 'https://www.tiktok.com/embed/6567659045795758085', + 'only_matching': True + }, { + 'url': 'https://www.tiktok.com/trending?shareId=6744531482393545985', + 'only_matching': True + }, { + 'url': 'https://www.tiktok.com/@leenabhushan/video/6748451240264420610?enter_from=h5_m', + 'only_matching': True }] def _real_extract(self, url): video_id = self._match_id(url) - webpage = self._download_webpage( - 'https://m.tiktok.com/v/%s.html' % video_id, video_id) - data = self._parse_json(self._search_regex( - r'\bdata\s*=\s*({.+?})\s*;', webpage, 'data'), video_id) - return self._extract_aweme(data) + headers = { + 'User-Agent': 'Mozilla/5.0 (X11; Linux x86_64; rv:72.0) Gecko/20100101 Firefox/72.0', + 'Referer': url + } + webpage = self._download_webpage(url, video_id, headers=headers, note='Downloading video webpage') + json_string = self._search_regex( + r'id=\"__NEXT_DATA__\"\s+type=\"application\/json\">\s*(?P[^<]+)', + webpage, 'json_string', group='json_string') + json_data = self._parse_json(json_string, video_id) + video_data = try_get(json_data, lambda x: x['props']['pageProps'], expected_type=dict) -class TikTokUserIE(TikTokBaseIE): - _VALID_URL = r'''(?x) - https?:// - (?: - (?:m\.)?tiktok\.com/h5/share/usr| - (?:www\.)?tiktok\.com/share/user - ) - /(?P\d+) - ''' - _TESTS = [{ - 'url': 'https://m.tiktok.com/h5/share/usr/188294915489964032.html', - 'info_dict': { - 'id': '188294915489964032', - }, - 'playlist_mincount': 24, - }, { - 'url': 'https://www.tiktok.com/share/user/188294915489964032', - 'only_matching': True, - }] + # Chech statusCode for success + if video_data.get('statusCode') == 0: + return self._extract_aweme(video_data, webpage) + + raise ExtractorError("Video not available", video_id=video_id) - def _real_extract(self, url): - user_id = self._match_id(url) - data = self._download_json( - 'https://m.tiktok.com/h5/share/usr/list/%s/' % user_id, user_id, - query={'_signature': '_'}) - entries = [] - for aweme in data['aweme_list']: - try: - entry = self._extract_aweme(aweme) - except ExtractorError: - continue - entry['extractor_key'] = TikTokIE.ie_key() - entries.append(entry) - return self.playlist_result(entries, user_id)