mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-25 00:31:26 +01:00
Compare commits
7 Commits
57c1b9e1c3
...
c6fd2b3aca
Author | SHA1 | Date | |
---|---|---|---|
|
c6fd2b3aca | ||
|
477b64b153 | ||
|
e60917c379 | ||
|
5ce0b127b2 | ||
|
49a3edd276 | ||
|
ec1d3c97d0 | ||
|
1564c71921 |
|
@ -747,6 +747,14 @@ from .gettr import (
|
|||
GettrStreamingIE,
|
||||
)
|
||||
from .giantbomb import GiantBombIE
|
||||
from .giphy import (
|
||||
GiphyChannelIE,
|
||||
GiphyChannelPageIE,
|
||||
GiphyIE,
|
||||
GiphySearchIE,
|
||||
GiphySearchURLIE,
|
||||
GiphyStoriesIE,
|
||||
)
|
||||
from .glide import GlideIE
|
||||
from .globalplayer import (
|
||||
GlobalPlayerAudioEpisodeIE,
|
||||
|
|
529
yt_dlp/extractor/giphy.py
Normal file
529
yt_dlp/extractor/giphy.py
Normal file
|
@ -0,0 +1,529 @@
|
|||
import itertools
|
||||
import re
|
||||
|
||||
from .common import InfoExtractor, SearchInfoExtractor
|
||||
from ..utils import (
|
||||
determine_ext,
|
||||
int_or_none,
|
||||
merge_dicts,
|
||||
str_or_none,
|
||||
traverse_obj,
|
||||
url_basename,
|
||||
url_or_none,
|
||||
)
|
||||
|
||||
|
||||
class GiphyBaseIE(InfoExtractor):
|
||||
_GIPHY_FE_WEB_API_KEY = 'Gc7131jiJuvI7IdN0HZ1D7nh0ow5BU6g'
|
||||
_GIPHY_FE_STORIES_AND_GIPHY_TV_API_KEY = '3eFQvabDx69SMoOemSPiYfh9FY0nzO9x'
|
||||
_GIPHY_MAX = 5000
|
||||
|
||||
def _extract_formats(self, media, is_still=False):
|
||||
f = []
|
||||
for format_id in media:
|
||||
if ('_still' in format_id) == is_still:
|
||||
for key in [k for k, v in media[format_id].items() if str(v)[:4] == 'http']:
|
||||
i = traverse_obj(media[format_id], {
|
||||
'width': ('width', {int_or_none}),
|
||||
'height': ('height', {int_or_none}),
|
||||
'url': (key, {url_or_none}),
|
||||
})
|
||||
f.append({
|
||||
'format_id': format_id,
|
||||
**i,
|
||||
})
|
||||
return f
|
||||
|
||||
def _extract_info(self, gif_data, video_id):
|
||||
formats, thumbnails, subtitles, uploader = [], [], {}, {}
|
||||
# formats, thumbnails, subtitles
|
||||
if data := gif_data.get('video'):
|
||||
for lang in data.get('captions', {}):
|
||||
for key in data['captions'][lang]:
|
||||
subtitles.setdefault(lang, []).append({'url': data['captions'][lang][key]})
|
||||
for category in ['assets', 'previews']:
|
||||
formats.extend(self._extract_formats(data.get(category, {})))
|
||||
if data.get('hls_manifest_url'):
|
||||
hls_fmts, hls_subs = self._extract_m3u8_formats_and_subtitles(
|
||||
data['hls_manifest_url'], video_id, 'mp4', m3u8_id='hls', fatal=False)
|
||||
formats.extend(hls_fmts)
|
||||
self._merge_subtitles(hls_subs, target=subtitles)
|
||||
if data.get('dash_manifest_url'):
|
||||
dash_fmts, dash_subs = self._extract_mpd_formats_and_subtitles(
|
||||
data['dash_manifest_url'], video_id, mpd_id='dash', fatal=False)
|
||||
formats.extend(dash_fmts)
|
||||
self._merge_subtitles(dash_subs, target=subtitles)
|
||||
if data := gif_data.get('images'):
|
||||
if data.get('looping'):
|
||||
data['looping']['height'] = traverse_obj(data, ('original_mp4', 'height', {int_or_none}))
|
||||
data['looping']['width'] = traverse_obj(data, ('original_mp4', 'width', {int_or_none}))
|
||||
sorted_data = dict(sorted(data.items(), reverse=True))
|
||||
formats.extend(self._extract_formats(sorted_data))
|
||||
thumbnails.extend(self._extract_formats(data, is_still=True))
|
||||
self._remove_duplicate_formats(formats)
|
||||
for f in formats:
|
||||
f.setdefault('http_headers', {})['Accept'] = 'video/*,image/*'
|
||||
for l in subtitles:
|
||||
for s in subtitles[l]:
|
||||
s.setdefault('http_headers', {})['Accept'] = 'text/*'
|
||||
for t in thumbnails:
|
||||
t.setdefault('http_headers', {})['Accept'] = 'image/*'
|
||||
# uploader
|
||||
if data := gif_data.get('user'):
|
||||
if isinstance(data, dict):
|
||||
uploader = traverse_obj(data, {
|
||||
'uploader': (('display_name', 'name', 'attribution_display_name', 'username'),
|
||||
{lambda x: x or gif_data.get('username')}),
|
||||
'uploader_id': ('username', {lambda x: x or gif_data.get('username')}),
|
||||
'uploader_url': (('profile_url', 'website_url'),
|
||||
{lambda x: f'https://giphy.com{x}' if x and x[0] == '/' else url_or_none(x)}),
|
||||
}, get_all=False)
|
||||
# basic info
|
||||
info = {
|
||||
**traverse_obj(gif_data, {
|
||||
'id': ('id', {lambda x: x or video_id}),
|
||||
'title': ('title', {lambda x: x.strip() if x else ''}),
|
||||
'description': ((None, 'video'), ('alt_text', 'description'),
|
||||
{lambda x: x.strip() if x and not x.startswith('Discover & share') else None}),
|
||||
'tags': ('tags', {list}),
|
||||
'age_limit': ('rating', {lambda x: 18 if x in ['r', 'nc-17'] else None}),
|
||||
'upload_date': (('import_datetime', 'create_datetime'),
|
||||
{lambda x: x[:10].replace('-', '') if x else None}),
|
||||
}, get_all=False),
|
||||
}
|
||||
return {
|
||||
**info,
|
||||
**{k: v for k, v in uploader.items() if v is not None},
|
||||
'thumbnails': thumbnails,
|
||||
'formats': formats,
|
||||
'subtitles': subtitles,
|
||||
}
|
||||
|
||||
def _api_channel_feed(self, channel_id):
|
||||
count, offset = 0, 0
|
||||
query_url = f'https://giphy.com/api/v4/channels/{channel_id}/feed/?offset={offset}'
|
||||
for _ in itertools.count(1):
|
||||
search_results = self._download_json(query_url, channel_id, fatal=False,
|
||||
note=f'Downloading feed data {offset + 1}-{offset + 25}')
|
||||
if not search_results or not search_results.get('results'):
|
||||
return
|
||||
for video in search_results['results']:
|
||||
yield {
|
||||
**self._extract_info(video, video['id']),
|
||||
'webpage_url': video['url'],
|
||||
'webpage_url_basename': url_basename(video['url']),
|
||||
}
|
||||
count += len(search_results.get('results'))
|
||||
if count >= (int_or_none(self.get_param('playlistend')) or (self._GIPHY_MAX + 1)):
|
||||
return
|
||||
query_url = url_or_none(search_results.get('next')) or ''
|
||||
offset = int(self._search_regex(r'offset=(\d+)', query_url, 'offset', default=0))
|
||||
# offset cannot exceed 5000
|
||||
if not query_url or offset > self._GIPHY_MAX:
|
||||
return
|
||||
|
||||
|
||||
class GiphyIE(GiphyBaseIE):
|
||||
_VALID_URL = r'https?://giphy\.com/(?:clips|gifs|stickers|embed)/(?:.+[/-])?(?P<id>[^/]+)/?$'
|
||||
_TESTS = [{
|
||||
'url': 'https://giphy.com/gifs/l2JIcQ4UH5SoPtMJi',
|
||||
'info_dict': {
|
||||
'id': 'l2JIcQ4UH5SoPtMJi',
|
||||
'ext': 'mp4',
|
||||
'title': 'excited slow motion GIF by Cats the Musical',
|
||||
'tags': ['giphyupload', 'excited', 'cats', 'musical', 'flip', 'slow motion', 'somersault', 'cats musical'],
|
||||
'thumbnail': r're:^https?://.*',
|
||||
'upload_date': '20160125',
|
||||
'uploader': 'Cats the Musical',
|
||||
'uploader_id': 'catsmusical',
|
||||
'uploader_url': 'https://giphy.com/catsmusical/',
|
||||
},
|
||||
}, {
|
||||
'url': 'http://giphy.com/gifs/l3vR8BKU0m8uX2mAg',
|
||||
'info_dict': {
|
||||
'id': 'l3vR8BKU0m8uX2mAg',
|
||||
'ext': 'mp4',
|
||||
'title': 'Giphy video #l3vR8BKU0m8uX2mAg',
|
||||
'tags': ['giphyupload'],
|
||||
'thumbnail': r're:^https?://.*',
|
||||
'upload_date': '20161022',
|
||||
'uploader': 'gus123',
|
||||
'uploader_id': 'gus123',
|
||||
'uploader_url': 'https://giphy.com/channel/gus123/',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://giphy.com/gifs/digitalpratik-digital-pratik-happy-fathers-day-dad-E1trcBzr59SGvmRDPY',
|
||||
'info_dict': {
|
||||
'id': 'E1trcBzr59SGvmRDPY',
|
||||
'ext': 'mp4',
|
||||
'title': 'Happy Fathers Day GIF by Digital Pratik',
|
||||
'tags': 'count:14',
|
||||
'thumbnail': r're:^https?://.*',
|
||||
'upload_date': '20210619',
|
||||
'uploader': 'Digital Pratik',
|
||||
'uploader_id': 'digitalpratik',
|
||||
'uploader_url': 'https://giphy.com/digitalpratik/',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://giphy.com/clips/southpark-south-park-episode-4-season-20-YyOPrvilA8FdiuSiQi',
|
||||
'info_dict': {
|
||||
'id': 'YyOPrvilA8FdiuSiQi',
|
||||
'ext': 'mp4',
|
||||
'title': 'You Can\'t Break Up With Me',
|
||||
'description': 'South Park, Season 20, Episode 4, Wieners Out',
|
||||
'tags': 'count:17',
|
||||
'thumbnail': r're:^https?://.*',
|
||||
'upload_date': '20220516',
|
||||
'uploader': 'South Park',
|
||||
'uploader_id': 'southpark',
|
||||
'uploader_url': 'https://giphy.com/southpark/',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://giphy.com/stickers/mario-PFxFYEZNUavG8',
|
||||
'info_dict': {
|
||||
'id': 'PFxFYEZNUavG8',
|
||||
'ext': 'mp4',
|
||||
'title': 'nintendo mario STICKER',
|
||||
'tags': ['transparent', 'gaming', 'nintendo', 'mario', 'giphynintendos'],
|
||||
'thumbnail': r're:^https?://.*',
|
||||
'upload_date': '20160908',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://giphy.com/embed/00xGP4zv8xENZ2tc3Y',
|
||||
'info_dict': {
|
||||
'id': '00xGP4zv8xENZ2tc3Y',
|
||||
'ext': 'mp4',
|
||||
'title': 'Love Is Blind Wow GIF by NETFLIX',
|
||||
'description': 'md5:89445e21c848eef12af249faef4fcf9f',
|
||||
'tags': 'count:24',
|
||||
'thumbnail': r're:^https?://.*',
|
||||
'upload_date': '20220214',
|
||||
'uploader': 'NETFLIX',
|
||||
'uploader_id': 'netflix',
|
||||
'uploader_url': 'https://giphy.com/netflix/',
|
||||
},
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
webpage = self._download_webpage(url.replace('/embed/', '/gifs/'), video_id)
|
||||
|
||||
title = (self._html_search_meta('twitter:title', webpage, default=None)
|
||||
or self._og_search_title(webpage).replace(' - Find & Share on GIPHY', '').strip())
|
||||
description = (self._html_search_meta('twitter:description', webpage, default=None)
|
||||
or self._og_search_description(webpage))
|
||||
description = description if not description.startswith('Discover & share') else None
|
||||
|
||||
gif_data = {}
|
||||
# search for: \"gif\":{\"type\":\"...},
|
||||
if json_str := self._html_search_regex(r'\\"\w+\\":({\\"type\\":\\"(?!emoji).*?is_dynamic\\":\w+}),',
|
||||
webpage, 'video_data', default=None):
|
||||
gif_data = self._parse_json(json_str.replace(r'\"', '"'), video_id)
|
||||
# search for: gif: {"...},
|
||||
elif json_str := self._html_search_regex(r'\s+\w+:\s*({".*?}),\n\s+', webpage, 'video_data', default='{}'):
|
||||
gif_data = self._parse_json(json_str, video_id)
|
||||
|
||||
info = self._extract_info(gif_data, video_id)
|
||||
|
||||
if not info.get('formats'):
|
||||
formats = []
|
||||
if url := self._og_search_video_url(webpage, default=None):
|
||||
formats.append({
|
||||
'format_id': determine_ext(url),
|
||||
'width': int_or_none(self._og_search_property('video:width', webpage)),
|
||||
'height': int_or_none(self._og_search_property('video:height', webpage)),
|
||||
'url': url,
|
||||
})
|
||||
if url := self._og_search_thumbnail(webpage, default=None):
|
||||
formats.append({
|
||||
'format_id': determine_ext(url),
|
||||
'width': int_or_none(self._og_search_property('image:width', webpage)),
|
||||
'height': int_or_none(self._og_search_property('image:height', webpage)),
|
||||
'url': url,
|
||||
})
|
||||
if url := self._html_search_meta('twitter:image', webpage, default=None):
|
||||
thumbnails = [{
|
||||
'width': int_or_none(self._html_search_meta('twitter:image:width', webpage, default=None)),
|
||||
'height': int_or_none(self._html_search_meta('twitter:image:height', webpage, default=None)),
|
||||
'url': url,
|
||||
}]
|
||||
info['formats'] = formats
|
||||
if not info.get('thumbnails'):
|
||||
info['thumbnails'] = thumbnails
|
||||
|
||||
if not info.get('uploader'):
|
||||
uploader = {}
|
||||
if data := gif_data.get('user'):
|
||||
if isinstance(data, str):
|
||||
idx = data.replace('$', '')
|
||||
if json_str := self._html_search_regex(rf'\\n{idx}:({{.*?}})\\n\w+:',
|
||||
webpage, 'uploader_data', default=None):
|
||||
json_str = re.sub(r'"\]\)self\.__next_f\.push\(\[\d+,"', '', json_str).replace(r'\"', '"')
|
||||
data = self._parse_json(json_str, video_id, fatal=False)
|
||||
if isinstance(data, dict):
|
||||
uploader = traverse_obj(data, {
|
||||
'uploader': (('display_name', 'name', 'attribution_display_name', 'username'),
|
||||
{lambda x: x or gif_data.get('username')}),
|
||||
'uploader_id': ('username', {str_or_none}),
|
||||
'uploader_url': (('profile_url', 'website_url'),
|
||||
{lambda x: f'https://giphy.com{x}' if x and x[0] == '/' else url_or_none(x)}),
|
||||
}, get_all=False)
|
||||
if not uploader:
|
||||
up_id = (gif_data.get('username')
|
||||
or self._html_search_regex(r'<div>@(\w+)</div>', webpage, 'uploader_id', default=None)
|
||||
or self._html_search_regex(r'"woff2"/><link[^>]+\.giphy\.com/(?:channel_assets|avatars)/(.+?)/',
|
||||
webpage, 'uploader_id', default=None))
|
||||
up_name = (title[(title.rfind(' by ') + 4):] if title.rfind(' by ') > 0 else None
|
||||
or self._html_search_regex(r'(?s)<h2\b[^>]*>([^<]+)</h2>', webpage, 'uploader', default=None)
|
||||
or self._html_search_regex(r'twitter:creator"[^>]+="((?!@giphy").*?)"', webpage, 'uploader', default=None)
|
||||
or up_id)
|
||||
uploader = {
|
||||
'uploader': up_name,
|
||||
'uploader_id': up_id,
|
||||
'uploader_url': (f'https://giphy.com/channel/{up_id}/' if up_id else None),
|
||||
}
|
||||
info = merge_dicts(info, {**{k: v for k, v in uploader.items() if v is not None}})
|
||||
|
||||
return {
|
||||
**merge_dicts(info, {
|
||||
'title': title,
|
||||
'description': description,
|
||||
}),
|
||||
}
|
||||
|
||||
|
||||
class GiphyChannelPageIE(GiphyBaseIE):
|
||||
_VALID_URL = r'https?://giphy\.com/(?!(?:clips|gifs|stickers|stories|search|embed)/)(?:.+/)?(?P<id>[^/]+)/?$'
|
||||
_TESTS = [{
|
||||
'url': 'https://giphy.com/catsmusical/',
|
||||
'playlist_count': 10,
|
||||
'info_dict': {
|
||||
'id': '8707',
|
||||
'title': 'Cats the Musical',
|
||||
'uploader_id': 'catsmusical',
|
||||
'uploader_url': 'https://giphy.com/channel/catsmusical',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://giphy.com/channel/catsmusical',
|
||||
'playlist_count': 10,
|
||||
'info_dict': {
|
||||
'id': '8707',
|
||||
'title': 'Cats the Musical',
|
||||
'uploader_id': 'catsmusical',
|
||||
'uploader_url': 'https://giphy.com/channel/catsmusical',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://giphy.com/southpark/reactions/lol',
|
||||
'playlist_count': 42,
|
||||
'info_dict': {
|
||||
'id': '1044',
|
||||
'title': 'LOL',
|
||||
'uploader_id': 'southpark',
|
||||
'uploader_url': 'https://giphy.com/channel/southpark',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://giphy.com/corgiyolk/cute-and-wholesome-corgi',
|
||||
'playlist_count': 14,
|
||||
'info_dict': {
|
||||
'id': '34458076',
|
||||
'title': 'Cute and Wholesome corgi',
|
||||
'uploader_id': 'corgiyolk',
|
||||
'uploader_url': 'https://giphy.com/channel/corgiyolk',
|
||||
},
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
|
||||
# search for: {"channelId": ...} or {..., "channel_id": ...,
|
||||
if channel_id := self._html_search_regex(r'\{[^\}\n]*"channel_?[iI]d":\s*"?([^",\}]+)[",\}]',
|
||||
webpage, 'channel_id', default=None):
|
||||
uploader_id = self._html_search_meta('twitter:creator', webpage).replace('@', '').lower()
|
||||
entries = []
|
||||
for i in self._api_channel_feed(channel_id):
|
||||
entries.append(i)
|
||||
|
||||
return {
|
||||
'id': channel_id,
|
||||
'title': (self._html_search_meta('twitter:title', webpage)
|
||||
or self._og_search_title(webpage)
|
||||
).replace(' GIFs on GIPHY - Be Animated', '').strip(),
|
||||
'uploader_id': uploader_id,
|
||||
'uploader_url': f'https://giphy.com/channel/{uploader_id}' if uploader_id != 'giphy' else None,
|
||||
'_type': 'playlist',
|
||||
'entries': entries,
|
||||
}
|
||||
|
||||
|
||||
class GiphyChannelIE(GiphyBaseIE, SearchInfoExtractor):
|
||||
IE_NAME = 'giphy:channel'
|
||||
IE_DESC = 'Giphy Channel'
|
||||
_SEARCH_KEY = 'giphychannel'
|
||||
_TESTS = [{
|
||||
'url': 'giphychannel30:pbsnature',
|
||||
'playlist_count': 30,
|
||||
'info_dict': {
|
||||
'id': 'pbsnature',
|
||||
'title': 'pbsnature',
|
||||
},
|
||||
}]
|
||||
|
||||
def _search_results(self, query):
|
||||
if webpage := self._download_webpage(f'https://giphy.com/channel/{query}', query):
|
||||
if channel_id := self._html_search_regex(r'\{[^\}\n]*"channel_?[iI]d":\s*"?([^",\}]+)[",\}]',
|
||||
webpage, 'channel_id', default=None):
|
||||
return self._api_channel_feed(channel_id)
|
||||
|
||||
|
||||
class GiphySearchIE(GiphyBaseIE, SearchInfoExtractor):
|
||||
IE_NAME = 'giphy:search'
|
||||
IE_DESC = 'Giphy Search'
|
||||
_SEARCH_KEY = 'giphysearch'
|
||||
_TESTS = [{
|
||||
'url': 'giphysearch20:super mario',
|
||||
'playlist_count': 20,
|
||||
'info_dict': {
|
||||
'id': 'super mario',
|
||||
'title': 'super mario',
|
||||
},
|
||||
}, {
|
||||
'url': 'giphysearch40:mickey&type=clips,stickers',
|
||||
'playlist_count': 40,
|
||||
'info_dict': {
|
||||
'id': 'mickey&type=clips,stickers',
|
||||
'title': 'mickey&type=clips,stickers',
|
||||
},
|
||||
}]
|
||||
|
||||
def _search_results(self, query):
|
||||
def search_query(query, offset, limit, category):
|
||||
# search api:
|
||||
# https://api.giphy.com/v1/gifs/search?rating=pg-13&offset=40&limit=15&type=gifs&q={query}&excludeDynamicResults=undefined&api_key=Gc7131jiJuvI7IdN0HZ1D7nh0ow5BU6g&pingback_id=1904d6e524cee33d
|
||||
return self._download_json(
|
||||
f'https://api.giphy.com/v1/{category}/search', query,
|
||||
note=f'Downloading {category} result {offset + 1}-{offset + limit}', query={
|
||||
'rating': 'r', # MPA film rating
|
||||
'offset': offset,
|
||||
'limit': limit,
|
||||
'type': category, # known types: 'clips', 'gifs', 'stickers', 'text', 'videos'
|
||||
'q': query,
|
||||
'excludeDynamicResults': 'undefined',
|
||||
'api_key': self._GIPHY_FE_WEB_API_KEY,
|
||||
})
|
||||
|
||||
# type: comma delimited list
|
||||
types = self._search_regex(r'&type=([^&]+)', query, 'type', default='gifs,stickers,videos')
|
||||
types = [(f'{x}s' if x[-1] != 's' and any(x in t for t in ['clips', 'gifs', 'stickers', 'videos']) else x)
|
||||
for x in [x.strip() for x in types.lower().split(',')]]
|
||||
query = query.split('&type=')[0]
|
||||
|
||||
offset, limit = 0, 50
|
||||
types_done = []
|
||||
for _ in itertools.count(1):
|
||||
for t in types:
|
||||
if t not in types_done:
|
||||
search_type = 'videos' if t == 'clips' else t # clips use 'videos' type
|
||||
search_results = search_query(query, offset, limit, search_type)
|
||||
if not search_results.get('data'):
|
||||
self.to_screen(f'{query}: {offset} {t} found')
|
||||
types_done.append(t)
|
||||
else:
|
||||
for video in search_results['data']:
|
||||
yield {
|
||||
**self._extract_info(video, video['id']),
|
||||
'webpage_url': video['url'],
|
||||
'webpage_url_basename': url_basename(video['url']),
|
||||
}
|
||||
if len(search_results['data']) < limit:
|
||||
total = offset + len(search_results['data'])
|
||||
self.to_screen(f'{query}: {total} {t} found')
|
||||
types_done.append(t)
|
||||
if len(types) > len(types_done):
|
||||
offset += limit
|
||||
if offset >= self._GIPHY_MAX:
|
||||
return
|
||||
else:
|
||||
return
|
||||
|
||||
|
||||
class GiphySearchURLIE(GiphyBaseIE):
|
||||
_VALID_URL = r'https?://giphy\.com/search/(?P<id>[^/]+)'
|
||||
_TESTS = [{
|
||||
'url': 'https://giphy.com/search/carcinoma',
|
||||
'playlist_count': 13,
|
||||
'info_dict': {
|
||||
'id': 'carcinoma',
|
||||
'title': 'carcinoma',
|
||||
},
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
if query := self._match_id(url):
|
||||
query = query.replace('-', ' ')
|
||||
return self.url_result(url=f'giphysearchall:{query}', url_transparent=True)
|
||||
|
||||
|
||||
class GiphyStoriesIE(GiphyBaseIE):
|
||||
_VALID_URL = r'https?://giphy\.com/stories/(?P<id>[^/]+)'
|
||||
_TESTS = [{
|
||||
'url': 'https://giphy.com/stories/indian-garyvee-de56ee78-adb5',
|
||||
'playlist_count': 6,
|
||||
'info_dict': {
|
||||
'id': 'de56ee78-adb5-428e-afb6-765bb489c4fc',
|
||||
'title': 'Indian Garyvee',
|
||||
'description': 'md5:468a8e2c898225baac05e32dcafe313b',
|
||||
'tags': [],
|
||||
'upload_date': '20190710',
|
||||
'uploader': 'Digital Pratik',
|
||||
'uploader_id': 'digitalpratik',
|
||||
'uploader_url': 'https://giphy.com/digitalpratik',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://giphy.com/stories/a-2021-oscars-wrap-up-161cda9a-4517',
|
||||
'playlist_count': 10,
|
||||
'info_dict': {
|
||||
'id': '161cda9a-4517-473b-9fea-6731af6c0d49',
|
||||
'title': 'A 2021 Oscars Wrap-Up',
|
||||
'description': 'What were the best moments in GIFs? Glad you asked!',
|
||||
'tags': [],
|
||||
'upload_date': '20210426',
|
||||
'uploader': 'Entertainment GIFs',
|
||||
'uploader_id': 'entertainment',
|
||||
'uploader_url': 'https://giphy.com/entertainment',
|
||||
},
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
slug = self._match_id(url)
|
||||
# https://x.giphy.com/v1/stories/slug/{slug}?api_key=3eFQvabDx69SMoOemSPiYfh9FY0nzO9x
|
||||
story = self._download_json(f'https://x.giphy.com/v1/stories/slug/{slug}?api_key={self._GIPHY_FE_STORIES_AND_GIPHY_TV_API_KEY}', slug)
|
||||
|
||||
if data := story.get('data'):
|
||||
entries = []
|
||||
for video in data.get('gifs'):
|
||||
entries.append({
|
||||
**self._extract_info(video['gif'], video['gif']['id']),
|
||||
'description': video['caption'],
|
||||
'webpage_url': video['gif']['url'],
|
||||
'webpage_url_basename': url_basename(video['gif']['url']),
|
||||
})
|
||||
info = traverse_obj(data, {
|
||||
'id': ('story_id', {lambda x: x or slug}),
|
||||
'title': ('title', {str_or_none}),
|
||||
'description': ('description', {str_or_none}),
|
||||
'tags': ('tags', {list}),
|
||||
'thumbnails': ('cover_gif', 'gif', 'images', {dict}, {lambda x: self._extract_formats(x, is_still=True)}),
|
||||
'upload_date': (('create_datetime', 'publish_datetime'),
|
||||
{lambda x: x[:10].replace('-', '') if x else None}),
|
||||
'uploader': ('user', ('display_name', 'username'), {str_or_none}),
|
||||
'uploader_id': ('user', 'username', {str_or_none}),
|
||||
'uploader_url': ('user', ('profile_url', 'website_url'),
|
||||
{lambda x: f'https://giphy.com{x}' if x and x[0] == '/' else url_or_none(x)}),
|
||||
}, get_all=False)
|
||||
|
||||
return {
|
||||
**info,
|
||||
'_type': 'playlist',
|
||||
'entries': entries,
|
||||
}
|
Loading…
Reference in New Issue
Block a user