eKeerar
93 lines
3.0 KiB
Python
93 lines
3.0 KiB
Python
from __future__ import annotations
|
|
|
|
from pathlib import Path
|
|
|
|
import numpy as np
|
|
from PIL import Image, ImageFilter
|
|
|
|
|
|
ROOT = Path(r"D:\Work_AI\Dansori")
|
|
PUPPET = ROOT / "DansoriEQ" / "src" / "DansoriEQ.App" / "Assets" / "Characters" / "Puppets" / "LeeSoriV2"
|
|
SOURCE = PUPPET / "leesori_v2_source.png"
|
|
BACKUP = PUPPET / "leesori_v2_source_pre_waist_fix.png"
|
|
QA = PUPPET / "qa_source_waist_patch_black.png"
|
|
AI_PATCH = Path(
|
|
r"C:\Users\eKeerar\.codex\generated_images\019f277b-18ad-7800-8b99-e77ae92de4bd"
|
|
r"\ig_0f4c5e4720cfe26e016a47e80d6f9c8191ba3b82a1c165cf39.png"
|
|
)
|
|
|
|
|
|
def alpha_bbox(im: Image.Image) -> tuple[int, int, int, int]:
|
|
return im.getchannel("A").getbbox() or (0, 0, im.width, im.height)
|
|
|
|
|
|
def subject_bbox_on_checker(im: Image.Image) -> tuple[int, int, int, int]:
|
|
arr = np.array(im.convert("RGB"))
|
|
mask = arr.max(axis=2) < 225
|
|
ys, xs = np.where(mask)
|
|
return int(xs.min()), int(ys.min()), int(xs.max() + 1), int(ys.max() + 1)
|
|
|
|
|
|
def map_box(
|
|
src_box: tuple[int, int, int, int],
|
|
src_bbox: tuple[int, int, int, int],
|
|
ref_bbox: tuple[int, int, int, int],
|
|
) -> tuple[int, int, int, int]:
|
|
sx0, sy0, sx1, sy1 = src_bbox
|
|
rx0, ry0, rx1, ry1 = ref_bbox
|
|
sw, sh = sx1 - sx0, sy1 - sy0
|
|
rw, rh = rx1 - rx0, ry1 - ry0
|
|
x0, y0, x1, y1 = src_box
|
|
return (
|
|
int(rx0 + ((x0 - sx0) / sw) * rw),
|
|
int(ry0 + ((y0 - sy0) / sh) * rh),
|
|
int(rx0 + ((x1 - sx0) / sw) * rw),
|
|
int(ry0 + ((y1 - sy0) / sh) * rh),
|
|
)
|
|
|
|
|
|
def main() -> None:
|
|
base = Image.open(BACKUP if BACKUP.exists() else SOURCE).convert("RGBA")
|
|
ref = Image.open(AI_PATCH).convert("RGBA")
|
|
|
|
src_bbox = alpha_bbox(base)
|
|
ref_bbox = subject_bbox_on_checker(ref)
|
|
|
|
# Torso-only region: below crop-top chest, through waist and top waistband.
|
|
src_patch_box = (86, 306, 282, 505)
|
|
ref_patch_box = map_box(src_patch_box, src_bbox, ref_bbox)
|
|
|
|
patch = ref.crop(ref_patch_box).resize(
|
|
(src_patch_box[2] - src_patch_box[0], src_patch_box[3] - src_patch_box[1]),
|
|
Image.Resampling.LANCZOS,
|
|
)
|
|
|
|
mask = Image.new("L", patch.size, 0)
|
|
px = mask.load()
|
|
w, h = patch.size
|
|
for y in range(h):
|
|
for x in range(w):
|
|
nx = abs((x / max(1, w - 1)) * 2 - 1)
|
|
ny = abs((y / max(1, h - 1)) * 2 - 1)
|
|
edge = max(nx, ny)
|
|
value = int(max(0.0, min(1.0, (1.0 - edge) / 0.36)) * 210)
|
|
# Keep the upper chest and far lower pants transition conservative.
|
|
if y < 28:
|
|
value = int(value * (y / 28))
|
|
if y > h - 28:
|
|
value = int(value * ((h - y) / 28))
|
|
px[x, y] = value
|
|
mask = mask.filter(ImageFilter.GaussianBlur(5))
|
|
|
|
out = base.copy()
|
|
out.alpha_composite(Image.composite(patch, out.crop(src_patch_box), mask), (src_patch_box[0], src_patch_box[1]))
|
|
out.save(SOURCE)
|
|
|
|
black = Image.new("RGBA", out.size, (16, 16, 18, 255))
|
|
black.alpha_composite(out)
|
|
black.convert("RGB").save(QA)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|