python生成psd文件实例
作者：AI算法网奇
文章介绍了如何使用Python生成包含多个图层的PSD文件,通过脚本`gen_psd.py`实现关键点分割人脸并生成多图层效果,作者分享了自己的经验,希望能为读者提供参考并支持脚本之家
python生成psd文件
多个图层，方便ps打开编辑
gen_psd.py
from PIL import Image
from psd_tools import PSDImage
from psd_tools.api.layers import PixelLayer
def image_to_psd(image_obj: Image, save_path):
# 确保图像模式为 RGBA
if image_obj.mode != "RGBA":
image_obj = image_obj.convert("RGBA")
# 将PIL图像转换为PSD格式
psd = PSDImage.frompil(image_obj)
# 创建一个新图层
pixel_layer = PixelLayer.frompil(image_obj, psd)
pixel_layer.visible = True # 设置图层为可见
psd.append(pixel_layer) # 将图层添加到PSD中
psd.save(save_path) # 保存为PSD文件
if __name__ == "__main__":
image_obj = Image.open(r"D:\project_2025\live2d\talking-head-anime-4-demo-main\demo\character_model\character.png")
save_path = 'demo.psd'
image_to_psd(image_obj, save_path)创建多个图层
from PIL import Image
from psd_tools import PSDImage
from psd_tools.api.layers import PixelLayer
def image_to_psd(image_paths, save_path):
# 读取第一张图，作为 PSD 画布
base_img = Image.open(image_paths[0]).convert("RGBA")
psd = PSDImage.frompil(base_img)
# 第一个图层
layer0 = PixelLayer.frompil(base_img, psd)
layer0.name = "Base"
layer0.visible = True
psd.append(layer0)
# 后续图片作为新图层
for i, img_path in enumerate(image_paths[1:], start=1):
img = Image.open(img_path).convert("RGBA")
layer = PixelLayer.frompil(img, psd)
layer.name = f"Layer_{i}"
layer.visible = True
psd.append(layer)
# 保存 PSD
psd.save(save_path)
if __name__ == "__main__":
image_paths = [
r"D:\project_2025\live2d\talking-head-anime-4-demo-main\demo\data\images\lambda_02_face_mask.png",
r"D:\project_2025\live2d\talking-head-anime-4-demo-main\demo\data\images\lambda_02.png",
]
image_to_psd(image_paths, "demo.psd")
关键点分割人脸，生成多图层
import os
from PIL import Image
from psd_tools import PSDImage
import cv2
import numpy as np
import os
from psd_tools.api.layers import PixelLayer
from Skps import FaceAna
def generate_eye_ellipse_mask(image_shape, landmarks, indices, scale_x=2, scale_y=1.25):
"""
使用最小外接椭圆生成眼睛 mask
"""
mask = np.zeros(image_shape[:2], dtype=np.uint8)
pts = landmarks[indices].astype(np.int32)
if pts.shape[0] < 5:
return mask
ellipse = cv2.fitEllipse(pts)
(cx, cy), (w, h), angle = ellipse
w *= scale_x
h *= scale_y
cv2.ellipse(
mask,
((int(cx), int(cy)), (int(w), int(h)), angle),
255,
-1
)
return mask
def generate_part_mask(image_shape, landmarks, indices):
"""
根据关键点索引生成对应部位的二进制遮罩。
Args:
image_shape: 原图尺寸 (H, W)
landmarks: 人脸关键点坐标数组
indices: 特定部位的关键点索引列表
Returns:
mask: 二值化遮罩 (0/255)
"""
mask = np.zeros(image_shape[:2], dtype=np.uint8)
pts = landmarks[indices].astype(np.int32)
# 使用凸包或最小矩形来定义区域
if len(indices) > 2: # 对于眼睛、嘴巴等轮廓点
hull = cv2.convexHull(pts)
cv2.fillConvexPoly(mask, hull, 255)
else: # 对于可能需要矩形定义的部位
x, y, w, h = cv2.boundingRect(pts)
cv2.rectangle(mask, (x, y), (x + w, y + h), 255, -1)
return mask
def masks_to_psd(base_image_path, masks_dict, output_psd_path):
# 1. 读取基础图像作为背景层
base_img = Image.open(base_image_path).convert("RGBA")
# 2. 创建一个以基础图像为画布的PSD对象[citation:4][citation:9]
psd = PSDImage.frompil(base_img)
# 3. 为每个部位创建图层[citation:4][citation:9]
for layer_name, mask in masks_dict.items():
# 将二值mask (0/255) 转换为RGBA图像
rgba_array = np.array(base_img).copy() # 形状为 (H, W, 4)
rgba_array[mask == 0, 3] = 0 # 索引3代表RGBA中的A（Alpha）通道
part_img = Image.fromarray(rgba_array, mode='RGBA')
# 第一个图层
layer0 = PixelLayer.frompil(part_img, psd)
layer0.name = layer_name
layer0.visible = True
psd.append(layer0)
# 4. 保存PSD文件[citation:4]
psd.save(output_psd_path)
print(f"PSD文件已生成: {output_psd_path}")
def generate_face_outer_mask(image_shape, landmarks, indices):
"""
生成头部轮廓以外的 mask
"""
h, w = image_shape[:2]
mask_face = np.zeros((h, w), np.uint8)
pts = landmarks[indices].astype(np.int32)
hull = cv2.convexHull(pts)
cv2.fillConvexPoly(mask_face, hull, 255)
# 反转：脸外 = 255
mask_outer = cv2.bitwise_not(mask_face)
return mask_outer
def generate_brow_mask(image_shape, landmarks, indices, scale_x=1.2, scale_y=1.5):
"""
生成眉毛 mask（扁椭圆 / 拉长）
"""
mask = np.zeros(image_shape[:2], dtype=np.uint8)
pts = landmarks[indices].astype(np.int32)
if pts.shape[0] < 3:
return mask
hull = cv2.convexHull(pts)
# 计算中心
cx = np.mean(hull[:, 0, 0])
cy = np.mean(hull[:, 0, 1])
# 缩放 hull（手动仿射）
scaled = []
for p in hull[:, 0, :]:
x = cx + (p[0] - cx) * scale_x
y = cy + (p[1] - cy) * scale_y
scaled.append([int(x), int(y)])
scaled = np.array(scaled, np.int32)
cv2.fillConvexPoly(mask, scaled, 255)
return mask
def process_single_image(image_path, facer, output_dir="output"):
"""
处理单张图片的主流程。
"""
# 1. 读取图片并运行关键点检测
image = cv2.imread(image_path)
result = facer.run(image)
# 假设只处理检测到的第一张脸
if len(result) == 0:
print(f"未检测到人脸: {image_path}")
return
landmarks = result[0]['kps'] # 形状应为 (98, 2)
# 2. 定义各部位的关键点索引 (需根据你的98点模型调整)
# 以下索引为示例，请务必根据你的模型定义进行核对和修改
PARTS_INDEX = {
"Face_Outline": list(range(0, 32)), # 脸部轮廓示例索引
"Left_Eye": list(range(60, 68)), # 左眼
"Right_Eye": list(range(68, 76)), # 右眼
"Nose": list(range(51, 60)), # 鼻子
"Mouth": list(range(76, 96)), # 嘴巴
"Left_Brow" : list(range(33, 41)),
"Right_Brow" : list(range(42, 50))
# 你可以根据需要添加更多部位，如眉毛: list(range(33, 51))
}
# 3. 为每个部位生成遮罩
masks = {}
for part_name, indices in PARTS_INDEX.items():
if part_name == "Face_Outline":
mask = generate_face_outer_mask(image.shape, landmarks, indices)
elif part_name in ["Left_Eye", "Right_Eye"]:
mask = generate_eye_ellipse_mask(image.shape, landmarks, indices)
elif part_name in ["Left_Brow", "Right_Brow"]:
mask = generate_brow_mask(image.shape, landmarks, indices)
else:
mask = generate_part_mask(image.shape, landmarks, indices)
masks[part_name] = mask
# 可选：保存每个部位的遮罩为PNG以供检查
# cv2.imwrite(os.path.join(output_dir, f"{part_name}.png"), mask)
# 4. 生成PSD
os.makedirs(output_dir, exist_ok=True)
base_name = os.path.splitext(os.path.basename(image_path))[0]
psd_path = os.path.join(output_dir, f"{base_name}_layers.psd")
masks_to_psd(image_path, masks, psd_path)
if __name__ == "__main__":
# 初始化你的关键点检测器
facer = FaceAna()
image_path = r"D:\project_2025\live2d\talking-head-anime-4-demo-main\demo\data\images\lambda_02.png"
process_single_image(image_path, facer, output_dir="psd_output")
总结
以上为个人经验，希望能给大家一个参考，也希望大家多多支持脚本之家。