mirror of
https://github.com/mikf/gallery-dl.git
synced 2024-11-24 03:32:33 +01:00
258 lines
8.5 KiB
Python
258 lines
8.5 KiB
Python
# -*- coding: utf-8 -*-
|
|
|
|
# Copyright 2019-2020 Mike Fährmann
|
|
#
|
|
# This program is free software; you can redistribute it and/or modify
|
|
# it under the terms of the GNU General Public License version 2 as
|
|
# published by the Free Software Foundation.
|
|
|
|
"""Extractors for https://500px.com/"""
|
|
|
|
from .common import Extractor, Message
|
|
import json
|
|
|
|
|
|
BASE_PATTERN = r"(?:https?://)?(?:web\.)?500px\.com"
|
|
|
|
|
|
class _500pxExtractor(Extractor):
|
|
"""Base class for 500px extractors"""
|
|
category = "500px"
|
|
directory_fmt = ("{category}", "{user[username]}")
|
|
filename_fmt = "{id}_{name}.{extension}"
|
|
archive_fmt = "{id}"
|
|
root = "https://500px.com"
|
|
|
|
def __init__(self, match):
|
|
Extractor.__init__(self, match)
|
|
self.session.headers["Referer"] = self.root + "/"
|
|
|
|
def items(self):
|
|
first = True
|
|
data = self.metadata()
|
|
yield Message.Version, 1
|
|
|
|
for photo in self.photos():
|
|
url = photo["images"][-1]["url"]
|
|
photo["extension"] = photo["image_format"]
|
|
if data:
|
|
photo.update(data)
|
|
if first:
|
|
first = False
|
|
yield Message.Directory, photo
|
|
yield Message.Url, url, photo
|
|
|
|
def metadata(self):
|
|
"""Returns general metadata"""
|
|
|
|
def photos(self):
|
|
"""Returns an iterable containing all relevant photo IDs"""
|
|
|
|
def _extend(self, edges):
|
|
"""Extend photos with additional metadata and higher resolution URLs"""
|
|
url = "https://api.500px.com/v1/photos"
|
|
params = {
|
|
"expanded_user_info" : "true",
|
|
"include_tags" : "true",
|
|
"include_geo" : "true",
|
|
"include_equipment_info": "true",
|
|
"vendor_photos" : "true",
|
|
"include_licensing" : "true",
|
|
"include_releases" : "true",
|
|
"liked_by" : "1",
|
|
"following_sample" : "100",
|
|
"image_size" : "4096",
|
|
"ids" : ",".join(
|
|
str(edge["node"]["legacyId"]) for edge in edges),
|
|
}
|
|
|
|
data = self._request_api(url, params)["photos"]
|
|
return [
|
|
data[str(edge["node"]["legacyId"])]
|
|
for edge in edges
|
|
]
|
|
|
|
def _request_api(self, url, params, csrf_token=None):
|
|
headers = {"Origin": self.root, "X-CSRF-Token": csrf_token}
|
|
return self.request(url, headers=headers, params=params).json()
|
|
|
|
def _request_graphql(self, opname, variables, query_hash):
|
|
url = "https://api.500px.com/graphql"
|
|
params = {
|
|
"operationName": opname,
|
|
"variables" : json.dumps(variables),
|
|
"extensions" : '{"persistedQuery":{"version":1'
|
|
',"sha256Hash":"' + query_hash + '"}}',
|
|
}
|
|
return self.request(url, params=params).json()["data"]
|
|
|
|
|
|
class _500pxUserExtractor(_500pxExtractor):
|
|
"""Extractor for photos from a user's photostream on 500px.com"""
|
|
subcategory = "user"
|
|
pattern = BASE_PATTERN + r"/(?!photo/)(?:p/)?([^/?#]+)/?(?:$|[?#])"
|
|
test = (
|
|
("https://500px.com/p/light_expression_photography", {
|
|
"pattern": r"https?://drscdn.500px.org/photo/\d+/m%3D4096/v2",
|
|
"range": "1-99",
|
|
"count": 99,
|
|
}),
|
|
("https://500px.com/light_expression_photography"),
|
|
("https://web.500px.com/light_expression_photography"),
|
|
)
|
|
|
|
def __init__(self, match):
|
|
_500pxExtractor.__init__(self, match)
|
|
self.user = match.group(1)
|
|
|
|
def photos(self):
|
|
variables = {"username": self.user, "pageSize": 20}
|
|
photos = self._request_graphql(
|
|
"OtherPhotosQuery", variables,
|
|
"018a5e5117bd72bdf28066aad02c4f2d"
|
|
"8acdf7f6127215d231da60e24080eb1b",
|
|
)["user"]["photos"]
|
|
|
|
while True:
|
|
yield from self._extend(photos["edges"])
|
|
|
|
if not photos["pageInfo"]["hasNextPage"]:
|
|
return
|
|
|
|
variables["cursor"] = photos["pageInfo"]["endCursor"]
|
|
photos = self._request_graphql(
|
|
"OtherPhotosPaginationContainerQuery", variables,
|
|
"b4af70d42c71a5e43f0be36ce60dc81e"
|
|
"9742ebc117cde197350f2b86b5977d98",
|
|
)["userByUsername"]["photos"]
|
|
|
|
|
|
class _500pxGalleryExtractor(_500pxExtractor):
|
|
"""Extractor for photo galleries on 500px.com"""
|
|
subcategory = "gallery"
|
|
directory_fmt = ("{category}", "{user[username]}", "{gallery[name]}")
|
|
pattern = (BASE_PATTERN + r"/(?!photo/)(?:p/)?"
|
|
r"([^/?#]+)/galleries/([^/?#]+)")
|
|
test = (
|
|
("https://500px.com/p/fashvamp/galleries/lera", {
|
|
"url": "002dc81dee5b4a655f0e31ad8349e8903b296df6",
|
|
"count": 3,
|
|
"keyword": {
|
|
"gallery": dict,
|
|
"user": dict,
|
|
},
|
|
}),
|
|
("https://500px.com/fashvamp/galleries/lera"),
|
|
)
|
|
|
|
def __init__(self, match):
|
|
_500pxExtractor.__init__(self, match)
|
|
self.user_name, self.gallery_name = match.groups()
|
|
self.user_id = self._photos = None
|
|
|
|
def metadata(self):
|
|
user = self._request_graphql(
|
|
"ProfileRendererQuery", {"username": self.user_name},
|
|
"5a17a9af1830b58b94a912995b7947b24f27f1301c6ea8ab71a9eb1a6a86585b",
|
|
)["profile"]
|
|
self.user_id = str(user["legacyId"])
|
|
|
|
variables = {
|
|
"galleryOwnerLegacyId": self.user_id,
|
|
"ownerLegacyId" : self.user_id,
|
|
"slug" : self.gallery_name,
|
|
"token" : None,
|
|
"pageSize" : 20,
|
|
}
|
|
gallery = self._request_graphql(
|
|
"GalleriesDetailQueryRendererQuery", variables,
|
|
"fb8bb66d31b58903e2f01ebe66bbe7937b982753be3211855b7bce4e286c1a49",
|
|
)["gallery"]
|
|
|
|
self._photos = gallery["photos"]
|
|
del gallery["photos"]
|
|
return {
|
|
"gallery": gallery,
|
|
"user" : user,
|
|
}
|
|
|
|
def photos(self):
|
|
photos = self._photos
|
|
variables = {
|
|
"ownerLegacyId": self.user_id,
|
|
"slug" : self.gallery_name,
|
|
"token" : None,
|
|
"pageSize" : 20,
|
|
}
|
|
|
|
while True:
|
|
yield from self._extend(photos["edges"])
|
|
|
|
if not photos["pageInfo"]["hasNextPage"]:
|
|
return
|
|
|
|
variables["cursor"] = photos["pageInfo"]["endCursor"]
|
|
photos = self._request_graphql(
|
|
"GalleriesDetailPaginationContainerQuery", variables,
|
|
"457c66d976f56863c81795f03e98cb54"
|
|
"3c7c6cdae7abeab8fe9e8e8a67479fa9",
|
|
)["galleryByOwnerIdAndSlugOrToken"]["photos"]
|
|
|
|
|
|
class _500pxImageExtractor(_500pxExtractor):
|
|
"""Extractor for individual images from 500px.com"""
|
|
subcategory = "image"
|
|
pattern = BASE_PATTERN + r"/photo/(\d+)"
|
|
test = ("https://500px.com/photo/222049255/queen-of-coasts", {
|
|
"url": "fbdf7df39325cae02f5688e9f92935b0e7113315",
|
|
"count": 1,
|
|
"keyword": {
|
|
"camera": "Canon EOS 600D",
|
|
"camera_info": dict,
|
|
"comments": list,
|
|
"comments_count": int,
|
|
"created_at": "2017-08-01T08:40:05+00:00",
|
|
"description": str,
|
|
"editored_by": None,
|
|
"editors_choice": False,
|
|
"extension": "jpg",
|
|
"feature": "popular",
|
|
"feature_date": "2017-08-01T09:58:28+00:00",
|
|
"focal_length": "208",
|
|
"height": 3111,
|
|
"id": 222049255,
|
|
"image_format": "jpg",
|
|
"image_url": list,
|
|
"images": list,
|
|
"iso": "100",
|
|
"lens": "EF-S55-250mm f/4-5.6 IS II",
|
|
"lens_info": dict,
|
|
"liked": None,
|
|
"location": None,
|
|
"location_details": dict,
|
|
"name": "Queen Of Coasts",
|
|
"nsfw": False,
|
|
"privacy": False,
|
|
"profile": True,
|
|
"rating": float,
|
|
"status": 1,
|
|
"tags": list,
|
|
"taken_at": "2017-05-04T17:36:51+00:00",
|
|
"times_viewed": int,
|
|
"url": "/photo/222049255/Queen-Of-Coasts-by-Olesya-Nabieva",
|
|
"user": dict,
|
|
"user_id": 12847235,
|
|
"votes_count": int,
|
|
"watermark": True,
|
|
"width": 4637,
|
|
},
|
|
})
|
|
|
|
def __init__(self, match):
|
|
_500pxExtractor.__init__(self, match)
|
|
self.photo_id = match.group(1)
|
|
|
|
def photos(self):
|
|
edges = ({"node": {"legacyId": self.photo_id}},)
|
|
return self._extend(edges)
|