summaryrefslogtreecommitdiffstats
path: root/yt_dlp/extractor/projectveritas.py
diff options
context:
space:
mode:
Diffstat (limited to 'yt_dlp/extractor/projectveritas.py')
-rw-r--r--yt_dlp/extractor/projectveritas.py10
1 files changed, 5 insertions, 5 deletions
diff --git a/yt_dlp/extractor/projectveritas.py b/yt_dlp/extractor/projectveritas.py
index daf1405..6f42485 100644
--- a/yt_dlp/extractor/projectveritas.py
+++ b/yt_dlp/extractor/projectveritas.py
@@ -17,7 +17,7 @@ class ProjectVeritasIE(InfoExtractor):
'title': 'Exclusive: Inside The New York and New Jersey Hospitals Battling Coronavirus',
'upload_date': '20200327',
'thumbnail': 'md5:6076477fe50b03eb8708be9415e18e1c',
- }
+ },
}, {
'url': 'https://www.projectveritas.com/video/ilhan-omar-connected-ballot-harvester-in-cash-for-ballots-scheme-car-is-full/',
'info_dict': {
@@ -26,13 +26,13 @@ class ProjectVeritasIE(InfoExtractor):
'title': 'Ilhan Omar connected Ballot Harvester in cash-for-ballots scheme: "Car is full" of absentee ballots',
'upload_date': '20200927',
'thumbnail': 'md5:194b8edf0e2ba64f25500ff4378369a4',
- }
+ },
}]
def _real_extract(self, url):
- id, type = self._match_valid_url(url).group('id', 'type')
- api_url = f'https://www.projectveritas.com/page-data/{type}/{id}/page-data.json'
- data_json = self._download_json(api_url, id)['result']['data']
+ video_id, video_type = self._match_valid_url(url).group('id', 'type')
+ api_url = f'https://www.projectveritas.com/page-data/{video_type}/{video_id}/page-data.json'
+ data_json = self._download_json(api_url, video_id)['result']['data']
main_data = traverse_obj(data_json, 'video', 'post')
video_id = main_data['id']
thumbnail = traverse_obj(main_data, ('image', 'ogImage', 'src'))