import re from os import sep import httpx import qrcode import string from pyrogram import ContinuePropagation from qrcode.image.pil import PilImage from io import BytesIO from PIL import Image, ImageDraw, ImageFont from defs.browser import get_browser from headers import headers def cut_text(old_str, cut): """ :说明: `get_cut_str` > 长文本自动换行切分 :参数: * `str: [type]`: 文本 * `cut: [type]`: 换行宽度,需要至少大于20 :返回: - `List`: 切分后的文本列表 """ punc = """,,、。.??)》】“"‘';;::!!·`~%^& """ si = 0 i = 0 next_str = old_str str_list = [] for s in next_str: if s in string.printable: si += 1 else: si += 2 i += 1 if next_str[0] == "\n": next_str = next_str[1:] elif s == "\n": str_list.append(next_str[: i - 1]) next_str = next_str[i - 1:] si = 0 i = 0 continue if si > cut: try: if next_str[i] in punc: i += 1 if next_str[i] in string.ascii_letters: for j in range(i, i - 18, -1): if next_str[j] == " ": i = j + 1 break except IndexError: str_list.append(next_str) return str_list str_list.append(next_str[:i]) next_str = next_str[i:] si = 0 i = 0 str_list.append(next_str) i = 0 non_wrap_str = [] for p in str_list: if not p: continue if p[-1] == "\n": p = p[:-1] non_wrap_str.append(p) i += 1 return non_wrap_str async def b23_extract(text): b23 = re.compile(r"b23.tv\\/(\w+)").search(text) if not b23: b23 = re.compile(r"b23.tv/(\w+)").search(text) try: assert b23 is not None url = f"https://b23.tv/{b23[1]}" except AssertionError: raise ContinuePropagation resp = httpx.head(url, follow_redirects=True) r = str(resp.url) return r async def video_info_get(cid): if cid[:2] == "av": video_info = httpx.get( f"http://api.bilibili.com/x/web-interface/view?aid={cid[2:]}" ) video_info = video_info.json() elif cid[:2] == "BV": video_info = httpx.get(f"http://api.bilibili.com/x/web-interface/view?bvid={cid}") video_info = video_info.json() else: return return video_info def numf(num: int): if num < 10000: view = str(num) elif num < 100000000: view = ("%.2f" % (num / 10000)) + "万" else: view = ("%.2f" % (num / 100000000)) + "亿" return view def binfo_image_create(video_info: dict): bg_y = 0 # 封面 pic_url = video_info["data"]["pic"] pic_get = httpx.get(pic_url).content pic_bio = BytesIO(pic_get) pic = Image.open(pic_bio) pic = pic.resize((560, 350)) pic_time_box = Image.new("RGBA", (560, 50), (0, 0, 0, 150)) pic.paste(pic_time_box, (0, 300), pic_time_box) bg_y += 350 + 20 # 时长 minutes, seconds = divmod(video_info["data"]["duration"], 60) hours, minutes = divmod(minutes, 60) video_time = f"{hours:02d}:{minutes:02d}:{seconds:02d}" tiem_font = ImageFont.truetype(f"resources{sep}font{sep}sarasa-mono-sc-bold.ttf", 30) draw = ImageDraw.Draw(pic) draw.text((10, 305), video_time, "white", tiem_font) # 分区 tname = video_info["data"]["tname"] tname_x, _ = tiem_font.getsize(tname) draw.text((560 - tname_x - 10, 305), tname, "white", tiem_font) # 标题 title = video_info["data"]["title"] title_font = ImageFont.truetype(f"resources{sep}font{sep}sarasa-mono-sc-bold.ttf", 25) title_cut_str = "\n".join(cut_text(title, 40)) _, title_text_y = title_font.getsize_multiline(title_cut_str) title_bg = Image.new("RGB", (560, title_text_y + 23), "#F5F5F7") draw = ImageDraw.Draw(title_bg) draw.text((15, 10), title_cut_str, "black", title_font) title_bg_y = title_bg.size[1] bg_y += title_bg_y # 简介 dynamic = ( "该视频没有简介" if video_info["data"]["desc"] == "" else video_info["data"]["desc"] ) dynamic_font = ImageFont.truetype(f"resources{sep}font{sep}sarasa-mono-sc-semibold.ttf", 18) dynamic_cut_str = "\n".join(cut_text(dynamic, 58)) _, dynamic_text_y = dynamic_font.getsize_multiline(dynamic_cut_str) dynamic_bg = Image.new("RGB", (560, dynamic_text_y + 24), "#F5F5F7") draw = ImageDraw.Draw(dynamic_bg) draw.rectangle((0, 0, 580, dynamic_text_y + 24), "#E1E1E5") draw.text((10, 10), dynamic_cut_str, "#474747", dynamic_font) dynamic_bg_y = dynamic_bg.size[1] bg_y += dynamic_bg_y # 视频数据 icon_font = ImageFont.truetype(f"resources{sep}font{sep}vanfont.ttf", 46) icon_color = (247, 145, 185) info_font = ImageFont.truetype(f"resources{sep}font{sep}sarasa-mono-sc-bold.ttf", 26) view = numf(video_info["data"]["stat"]["view"]) # 播放 \uE6E6 danmaku = numf(video_info["data"]["stat"]["danmaku"]) # 弹幕 \uE6E7 favorite = numf(video_info["data"]["stat"]["favorite"]) # 收藏 \uE6E1 coin = numf(video_info["data"]["stat"]["coin"]) # 投币 \uE6E4 like = numf(video_info["data"]["stat"]["like"]) # 点赞 \uE6E0 info_bg = Image.new("RGB", (560, 170), "#F5F5F7") draw = ImageDraw.Draw(info_bg) draw.text((5 + 10, 20), "\uE6E0", icon_color, icon_font) draw.text((5 + 64, 27), like, "#474747", info_font) draw.text((5 + 10 + 180, 20), "\uE6E4", icon_color, icon_font) draw.text((5 + 64 + 180, 27), coin, "#474747", info_font) draw.text((5 + 10 + 180 + 180, 20), "\uE6E1", icon_color, icon_font) draw.text((5 + 64 + 180 + 180, 27), favorite, "#474747", info_font) draw.text((5 + 100, 93), "\uE6E6", icon_color, icon_font) draw.text((5 + 154, 100), view, "#474747", info_font) draw.text((5 + 100 + 210, 93), "\uE6E7", icon_color, icon_font) draw.text((5 + 154 + 210, 100), danmaku, "#474747", info_font) info_bg_y = info_bg.size[1] bg_y += info_bg_y # UP主 # 等级 0-4 \uE6CB-F 5-6\uE6D0-1 # UP \uE723 if "staff" in video_info["data"]: up_list = [] for up in video_info["data"]["staff"]: up_mid = up["mid"] up_data = httpx.get( f"https://api.bilibili.com/x/space/acc/info?mid={up_mid}", headers=headers, ).json() up_list.append( { "name": up["name"], "up_title": up["title"], "face": up["face"], "color": up_data["data"]["vip"]["nickname_color"] if up_data["data"]["vip"]["nickname_color"] != "" else "black", "follower": up["follower"], "level": up_data["data"]["level"], } ) else: up_mid = video_info["data"]["owner"]["mid"] up_data = httpx.get( f"https://api.bilibili.com/x/space/acc/info?mid={up_mid}", headers=headers, ).json() up_stat = httpx.get( f"https://api.bilibili.com/x/relation/stat?vmid={up_mid}", headers=headers, ).json() up_list = [ { "name": up_data["data"]["name"], "up_title": "UP主", "face": up_data["data"]["face"], "color": up_data["data"]["vip"]["nickname_color"] if up_data["data"]["vip"]["nickname_color"] != "" else "black", "follower": up_stat["data"]["follower"], "level": up_data["data"]["level"], } ] up_num = len(up_list) up_bg = Image.new("RGB", (560, 20 + (up_num * 120) + 20), "#F5F5F7") draw = ImageDraw.Draw(up_bg) face_size = (80, 80) mask = Image.new("RGBA", face_size, color=(0, 0, 0, 0)) mask_draw = ImageDraw.Draw(mask) mask_draw.ellipse( (0, 0, face_size[0], face_size[1]), fill=(0, 0, 0, 255) # type: ignore ) name_font = ImageFont.truetype(f"resources{sep}font{sep}sarasa-mono-sc-bold.ttf", 24) up_title_font = ImageFont.truetype(f"resources{sep}font{sep}sarasa-mono-sc-bold.ttf", 20) follower_font = ImageFont.truetype(f"resources{sep}font{sep}sarasa-mono-sc-semibold.ttf", 22) i = 0 for up in up_list: if up["level"] == 0: up_level = "\uE6CB" level_color = (191, 191, 191) elif up["level"] == 1: up_level = "\uE6CC" level_color = (191, 191, 191) elif up["level"] == 2: up_level = "\uE6CD" level_color = (149, 221, 178) elif up["level"] == 3: up_level = "\uE6CE" level_color = (146, 209, 229) elif up["level"] == 4: up_level = "\uE6CF" level_color = (255, 179, 124) elif up["level"] == 5: up_level = "\uE6D0" level_color = (255, 108, 0) else: up_level = "\uE6D1" level_color = (255, 0, 0) # 头像 face_url = up["face"] face_get = httpx.get(face_url).content face_bio = BytesIO(face_get) face = Image.open(face_bio) face.convert("RGB") face = face.resize(face_size) up_bg.paste(face, (20, 20 + (i * 120)), mask) # 名字 draw.text((160, 25 + (i * 120)), up["name"], up["color"], name_font) name_size_x, _ = name_font.getsize(up["name"]) # 等级 draw.text( (160 + name_size_x + 10, 16 + (i * 120)), up_level, level_color, icon_font ) # 身份 up_title_size_x, up_title_size_y = up_title_font.getsize(up["up_title"]) draw.rectangle( ( 60, 10 + (i * 120), 73 + up_title_size_x, 18 + (i * 120) + up_title_size_y, ), "white", icon_color, 3, ) draw.text((67, 13 + (i * 120)), up["up_title"], icon_color, up_title_font) # 粉丝量 draw.text( (162, 66 + (i * 120)), "粉丝 " + numf(up["follower"]), "#474747", follower_font, ) i += 1 up_bg_y = up_bg.size[1] bg_y += up_bg_y # 底部栏 baner_bg = Image.new("RGB", (600, 170), icon_color) draw = ImageDraw.Draw(baner_bg) # 二维码 qr = qrcode.QRCode(border=1) qr.add_data("https://b23.tv/" + video_info["data"]["bvid"]) qr_image = qr.make_image(PilImage, fill_color=icon_color, back_color="#F5F5F7") qr_image = qr_image.resize((140, 140)) baner_bg.paste(qr_image, (50, 10)) # Logo # LOGO \uE725 logo_font = ImageFont.truetype(f"resources{sep}font{sep}vanfont.ttf", 100) draw.text((300, 28), "\uE725", "#F5F5F7", logo_font) bg_y += 170 video = Image.new("RGB", (600, bg_y + 40), "#F5F5F7") video.paste(pic, (20, 20)) video.paste(title_bg, (20, 390)) video.paste(dynamic_bg, (20, 390 + title_bg_y + 20)) video.paste(info_bg, (20, 390 + title_bg_y + 20 + dynamic_bg_y + 20)) video.paste(up_bg, (20, 390 + title_bg_y + 20 + dynamic_bg_y + 10 + info_bg_y)) video.paste( baner_bg, (0, 390 + title_bg_y + 20 + dynamic_bg_y + 10 + info_bg_y + up_bg_y) ) picture = BytesIO() video.save(picture, format="PNG") picture.name = "bilibili.png" return picture async def get_dynamic_screenshot_pc(dynamic_id): """电脑端动态截图""" url = f"https://t.bilibili.com/{dynamic_id}" browser = await get_browser() context = await browser.new_context( viewport={"width": 2560, "height": 1080}, device_scale_factor=2, ) await context.add_cookies( [ { "name": "hit-dyn-v2", "value": "1", "domain": ".bilibili.com", "path": "/", } ] ) page = await context.new_page() try: await page.goto(url, wait_until="networkidle", timeout=10000) # 动态被删除或者进审核了 if page.url == "https://www.bilibili.com/404": return None card = await page.query_selector(".card") assert card clip = await card.bounding_box() assert clip bar = await page.query_selector(".bili-dyn-action__icon") assert bar bar_bound = await bar.bounding_box() assert bar_bound clip["height"] = bar_bound["y"] - clip["y"] return await page.screenshot(clip=clip, full_page=True) except Exception: print(f"截取动态时发生错误:{url}") return await page.screenshot(full_page=True) finally: await context.close()