Peach: Local-first video metadata scraper and media manager.
"""演示库的人像素材:按钉死的清单取图,再排成封面与头像。
图片字节不进仓库(ADR-0026),进仓库的只有 `demo-portraits.json` 这份清单——去哪取、
取到的该是什么哈希。生成器运行时按清单取一次,缓存在输出目录旁边。
**清单是人工逐张看过后钉死的,不按图库自己的顺序自动取。** 图库把目录前缀当来源优先级
排序,排第一的答的是「先试哪一张」,不是「哪一张能放进公开文档」:实测排在前面的大图
多是片商宣传素材,内衣、泳装乃至露点都在里面。自动取必然翻车,所以这里只认清单。
"""
from __future__ import annotations
import hashlib
import json
import sys
import urllib.parse
import urllib.request
from dataclasses import dataclass
from pathlib import Path
SRC_DIR = Path(__file__).resolve().parent.parent / "src"
if str(SRC_DIR) not in sys.path:
sys.path.insert(0, str(SRC_DIR))
from peach.user_agent import USER_AGENT # noqa: E402
from peach.peach_proxy import read_url
MANIFEST_PATH = Path(__file__).resolve().parent / "demo-portraits.json"
GFRIENDS_RAW = "https://raw.githubusercontent.com/gfriends/gfriends/master/Content/"
#: 单张图的上限。清单里最大的一张 2880x1800 约 2.1 MB,留一倍余量;越界说明取错了东西。
MAX_IMAGE_BYTES = 4 * 1024 * 1024
COVER_LANDSCAPE = (800, 538)
COVER_PORTRAIT = (600, 900)
AVATAR_SIDE = 480
#: 封面与画面上的文字要落到 CJK 字形上。按平台找一份,找不到就不写字——
#: 缺字体的机器上宁可出一张没有标题的封面,也不要满屏豆腐块。
FONT_CANDIDATES = (
"C:/Windows/Fonts/msyh.ttc",
"C:/Windows/Fonts/simhei.ttf",
"/System/Library/Fonts/PingFang.ttc",
"/System/Library/Fonts/Hiragino Sans GB.ttc",
"/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",
"/usr/share/fonts/truetype/noto/NotoSansCJK-Regular.ttc",
)
@dataclass(frozen=True)
class Portrait:
"""清单里的一条:去哪取、取到的该是什么。"""
name: str
key: str
category: str
filename: str
sha256: str
width: int
height: int
@property
def url(self) -> str:
return (GFRIENDS_RAW + urllib.parse.quote(self.category)
+ "/" + urllib.parse.quote(self.filename))
@property
def cache_name(self) -> str:
"""缓存文件名带哈希前缀:清单换了图,旧缓存不会被当成新的。"""
return f"{self.sha256[:16]}-{self.category}.jpg"
@property
def landscape(self) -> bool:
return self.width > self.height
def load_manifest(path: Path | None = None) -> list[Portrait]:
"""读清单。顺序就是清单里的顺序,按种子取用由调用方决定。"""
document = json.loads((path or MANIFEST_PATH).read_text(encoding="utf-8"))
return [Portrait(**entry) for entry in document["entries"]]
def fetch(portrait: Portrait, cache_dir: Path, *, opener=None) -> Path:
"""取一张图到缓存,校验哈希。缓存命中就不再联网。
哈希不符是硬错误,不是可以将就的降级:清单的全部意义就是「取到的是我看过的那张」,
校验放过去等于回到自动取图。
"""
cache_dir.mkdir(parents=True, exist_ok=True)
target = cache_dir / portrait.cache_name
if target.is_file() and hashlib.sha256(target.read_bytes()).hexdigest() == portrait.sha256:
return target
request = urllib.request.Request(portrait.url, headers={"User-Agent": USER_AGENT})
if opener is None:
body = read_url(portrait.url, headers=dict(request.header_items()), max_bytes=MAX_IMAGE_BYTES)
else:
with opener(request, timeout=60) as response:
body = response.read(MAX_IMAGE_BYTES + 1)
if len(body) > MAX_IMAGE_BYTES:
raise RuntimeError(f"{portrait.name} 的图超过 {MAX_IMAGE_BYTES} 字节,取错了东西")
digest = hashlib.sha256(body).hexdigest()
if digest != portrait.sha256:
raise RuntimeError(
f"{portrait.name} 的图与清单对不上:清单 {portrait.sha256[:16]},取到 {digest[:16]}。"
"图库换了这一张,重新人工看过再更新清单。")
target.write_bytes(body)
return target
def load_font(size: int):
from PIL import ImageFont
for candidate in FONT_CANDIDATES:
path = Path(candidate)
if path.is_file():
try:
return ImageFont.truetype(str(path), size)
except OSError:
continue
return None
def _ellipsize(draw, text: str, font, room: int) -> str:
"""放不下就截断加省略号。长标题横着写出画框比截断难看得多。"""
if draw.textlength(text, font=font) <= room:
return text
trimmed = text
while trimmed and draw.textlength(trimmed + "…", font=font) > room:
trimmed = trimmed[:-1]
return (trimmed + "…") if trimmed else ""
def _cover_box(image, size: tuple[int, int], anchor: float = 0.15):
"""放大裁切填满画框,取景偏上以保住脸。"""
from PIL import Image
scale = max(size[0] / image.width, size[1] / image.height)
big = image.resize((round(image.width * scale), round(image.height * scale)),
Image.LANCZOS)
left = max(0, (big.width - size[0]) // 2)
top = max(0, min(int((big.height - size[1]) * anchor), big.height - size[1]))
return big.crop((left, top, left + size[0], top + size[1]))
def draw_cover(source: Path, destination: Path, *, code: str, title: str,
performer: str, studio: str, year: str, orientation: str) -> None:
"""封面:人像占一侧,另一侧排番号、标题与出演者。
横屏作品用横版封面,这也是番号作品封面的实际形状;竖屏作品用竖版,文字压在底部。
人像不拉伸、不靠模糊背景撑满——那两种做法出来都是「头像撑成海报」的样子。
"""
from PIL import Image, ImageDraw
image = Image.open(source).convert("RGB")
# 四行文字:番号、标题、出演者、年份与厂牌。字号与行距一起定,逐行累加落位,
# 不按固定偏移量硬写——竖版的文字块比横版矮,写死偏移会溢出画面底边。
rows = [row for row in ((code, 26, "#8a919c", 0), (title, 42, "#f3f5f8", 14),
(performer, 24, "#c5cad1", 18),
(f"{year} · {studio}", 21, "#8a919c", 26)) if row[0]]
fonts = [load_font(size) for _, size, _, _ in rows]
block = sum(size + lead for (_, size, _, lead) in rows)
if orientation == "竖屏":
canvas = _cover_box(image, COVER_PORTRAIT).convert("RGB")
width, height = COVER_PORTRAIT
# 上面一段做渐变过渡,文字全部落在下面的实心段——纯渐变带的顶部太淡,
# 第一行番号压在画面亮部时基本看不见。
fade, solid = 72, block + 56
overlay = Image.new("RGBA", (width, fade + solid), (10, 11, 14, 220))
painter = ImageDraw.Draw(overlay)
for row in range(fade):
painter.line([(0, row), (width, row)],
fill=(10, 11, 14, int(220 * (row / fade) ** 1.2)))
canvas.paste(overlay, (0, height - fade - solid), overlay)
left, cursor = 34, height - solid + 24
else:
canvas = Image.new("RGB", COVER_LANDSCAPE, "#14161a")
panel = round(COVER_LANDSCAPE[0] * 0.44)
canvas.paste(_cover_box(image, (panel, COVER_LANDSCAPE[1]), anchor=0.12), (0, 0))
left = panel + 44
cursor = (COVER_LANDSCAPE[1] - block) // 2
draw = ImageDraw.Draw(canvas, "RGBA")
if fonts[0] is None:
destination.parent.mkdir(parents=True, exist_ok=True)
canvas.save(destination,
format="PNG" if destination.suffix.lower() == ".png" else "JPEG",
quality=90)
return
room = canvas.width - left - 30
for (text, size, colour, lead), font in zip(rows, fonts):
cursor += lead
draw.text((left, cursor), _ellipsize(draw, text, font, room), font=font, fill=colour)
cursor += size
destination.parent.mkdir(parents=True, exist_ok=True)
canvas.save(destination, format="PNG" if destination.suffix.lower() == ".png" else "JPEG",
quality=90)
def draw_avatar(source: Path, destination: Path) -> None:
"""头像:按短边取正方形,偏上取景。"""
from PIL import Image
image = Image.open(source).convert("RGB")
side = min(image.width, image.height)
left = (image.width - side) // 2
top = int((image.height - side) * 0.12)
square = image.crop((left, top, left + side, top + side))
destination.parent.mkdir(parents=True, exist_ok=True)
square.resize((AVATAR_SIDE, AVATAR_SIDE), Image.LANCZOS).save(destination, quality=90)