X-Git-Url: https://jfr.im/git/yt-dlp.git/blobdiff_plain/e897bd8292a41999cf51dba91b390db5643c72db..5dbac313ae4e3e8521dfe2e1a6a048a98ff4b4fe:/yt_dlp/extractor/nuvid.py diff --git a/yt_dlp/extractor/nuvid.py b/yt_dlp/extractor/nuvid.py index 0ef0ec70b..99a426b25 100644 --- a/yt_dlp/extractor/nuvid.py +++ b/yt_dlp/extractor/nuvid.py @@ -22,7 +22,7 @@ class NuvidIE(InfoExtractor): 'duration': 321.0, 'age_limit': 18, 'thumbnail': r're:https?://.+\.jpg', - } + }, }, { 'url': 'https://m.nuvid.com/video/6523263', 'md5': 'ebd22ce8e47e1d9a4d0756a15c67da52', @@ -34,7 +34,7 @@ class NuvidIE(InfoExtractor): 'age_limit': 18, 'thumbnail': r're:https?://.+\.jpg', 'thumbnails': list, - } + }, }, { 'url': 'http://m.nuvid.com/video/6415801/', 'md5': '638d5ececb138d5753593f751ae3f697', @@ -45,7 +45,7 @@ class NuvidIE(InfoExtractor): 'duration': 1882, 'age_limit': 18, 'thumbnail': r're:https?://.+\.jpg', - } + }, }] def _real_extract(self, url): @@ -64,7 +64,7 @@ def _real_extract(self, url): }) webpage = self._download_webpage( - 'http://m.nuvid.com/video/%s' % (video_id, ), + f'http://m.nuvid.com/video/{video_id}', video_id, 'Downloading video page', fatal=False) or '' title = strip_or_none(video_data.get('title') or self._html_search_regex(