CLOTH_WEB_V1 / clothGen.py
SIUBIU's picture
Update clothGen.py
cf9150b verified
raw
history blame
8.52 kB
import fal_client
import pandas as pd
from prompt_gen import prompt_gen
import requests
import os
from openai import OpenAI
import random
import shutil
from pathlib import Path
from PIL import Image
nv_prompt_file = pd.read_excel('汉服-女词库.xlsx')
na_prompt_file = pd.read_excel('汉服-男词库.xlsx')
nv_prompt = nv_prompt_file.to_string(index=False)
na_prompt = na_prompt_file.to_string(index=False)
save_directory = "downloads"
def prompt_nan(prompt):
client = OpenAI()
completion = client.chat.completions.create(
model="gpt-4o",
messages=[
{"role": "system",
"content": "You are a helpful assistant.", },
{"role": "user",
"content": "Please showcase the overall appearance of this Hanfu robe against a contrasting white "
"background. Highlight its intricate details and unique design elements, including" + prompt,
}
]
)
print("change prompt: ")
print(completion.choices[0].message.content)
return completion.choices[0].message.content
def pro_gen(advice, gender, index):
prompt = prompt_gen(advice, gender)
start_index = prompt.find("Begin")
if start_index == -1:
start_index = prompt.find("begin")
prompt__gen = ""
if start_index != -1:
start_index += len("Begin\n")
end_index = prompt.find("End")
if end_index != -1:
prompt__gen = prompt[start_index:end_index]
# if gender == "男":
# prompt__gen = prompt_nan(prompt__gen)
filename = os.path.join(save_directory, f"prompt_{index}.txt")
with open(filename, "w") as file:
file.write(prompt__gen)
# print(prompt__gen)
else:
print("No 'promptEnd' found after 'prompt'.")
else:
print("No 'prompt' found in the text.")
return prompt__gen
def generate(lora_path, prompt__gen, index):
# print(prompt__gen)
handler = fal_client.submit(
"fal-ai/fast-sdxl",
arguments={
"prompt": prompt__gen,
"negative_prompt": "human, people, person, man, woman, child, model, face, head, eyes, hands, arms, legs, "
"feet, hair, portrait, worst quality, low quality, normal quality, lowres, signature, "
"watermark, jpeg artifacts, logo, monochrome, grayscale, ugly",
"image_size": "portrait_4_3",
"num_inference_steps": 28,
"guidance_scale": 7.5,
"num_images": 1,
"loras": [{"path": lora_path, "scale": 0.7}],
"embeddings": [],
"safety_checker_version": "v1",
"format": "jpeg"
},
)
request_id = handler.request_id
result = fal_client.result("fal-ai/fast-sdxl", request_id)
image_index = index * 2 - 1
for image in result['images']:
response = requests.get(image['url'])
if response.status_code == 200:
filename = os.path.join(save_directory, f"cloth_{image_index}.jpeg")
with open(filename, 'wb') as f:
f.write(response.content)
image_index += 1
else:
print(f"Failed to download image from {image['url']}")
client = OpenAI()
completion = client.chat.completions.create(
model="gpt-4o",
messages=[
{"role": "system",
"content": "You are a helpful assistant.", },
{"role": "user",
"content": prompt__gen + "以上是一段对于一套汉服的描述,请根据描述内容对该套汉服进行介绍。要求以介绍的口吻输出内容",
}
]
)
cloth_intro = ("汉服,是汉民族的传统服饰。又称衣冠、衣裳、汉装。汉服是中国“衣冠上国”“礼仪之邦”“锦绣中华”的体现,承载了中国的染织绣等杰出"
"工艺和美学,传承了30多项中国非物质文化遗产以及受保护的中国工艺美术。\n") + completion.choices[
0].message.content
filename = os.path.join(save_directory, f"cloth_intro_{index * 2 - 1}.txt")
with open(filename, "w") as file:
file.write(cloth_intro)
def convert_image_to_jpeg(input_path, output_path):
try:
image = Image.open(input_path)
if image.mode in ('RGBA', 'LA'):
image = image.convert('RGB')
image.save(output_path, 'JPEG')
except Exception as e:
print(f"转换图像时出错: {e}")
def pic_match(prompt__gen, cates, folder_path, intro_path, index):
client = OpenAI()
completion = client.chat.completions.create(
model="gpt-4o",
messages=[
{"role": "system",
"content": "You are a helpful assistant.", },
{"role": "user",
"content": prompt__gen + "以上是关于一套汉族服饰的描述,请根据描述内容从以下几种颜色中选择最符合描述的一种,可选颜色包括:" + cates
+ ". 仅需输出一种颜色名称,不要带任何符号",
}
]
)
print(f"Selected color: {completion.choices[0].message.content}")
folder_path = os.path.join(folder_path, completion.choices[0].message.content)
files = os.listdir(folder_path)
random_file = random.choice(files)
source_file_path = os.path.join(folder_path, random_file)
file_prefix, file_ext = os.path.splitext(random_file)
target_file_path = os.path.join(save_directory, f"cloth_{index * 2}.jpeg")
convert_image_to_jpeg(source_file_path, target_file_path)
file_extension = ".txt"
search_path = Path(intro_path)
for file in search_path.glob(f"{file_prefix}*{file_extension}"):
if file.is_file():
with open(file, "r") as f:
content = f.read()
client = OpenAI()
completion = client.chat.completions.create(
model="gpt-4o",
messages=[
{"role": "system",
"content": "You are a helpful assistant.", },
{"role": "user",
"content": content + "以上是一段对于一套汉服的描述,请根据描述内容对该套汉服进行介绍。要求以介绍的口吻输出内容",
}
]
)
cloth_intro = ("汉服,是汉民族的传统服饰。又称衣冠、衣裳、汉装。汉服是中国“衣冠上国”“礼仪之邦”“锦绣中华”的体现,承载了中国的染织绣等杰出"
"工艺和美学,传承了30多项中国非物质文化遗产以及受保护的中国工艺美术。\n") + \
completion.choices[0].message.content
filename = os.path.join(save_directory, f"cloth_intro_{index * 2}.txt")
with open(filename, "w") as file:
file.write(cloth_intro)
return target_file_path
def cloth_gen(gender):
cates = "Black, Blue, Green, Orange, Pink, Red, Violet, White, Yellow"
lora_path = "https://huggingface.co/PPSharks/PPSharksModels/resolve/main/NV.safetensors"
folder_path = "database/female"
intro_path = "database/female_intro"
if gender == "男":
lora_path = "https://huggingface.co/PPSharks/PPSharksModels/resolve/main/NA.safetensors"
cates = "Black, Blue, Green, Brown, Red, Violet"
folder_path = "database/male"
intro_path = "database/male_intro"
elif gender == "女":
lora_path = "https://huggingface.co/PPSharks/PPSharksModels/resolve/main/NV.safetensors"
cates = "Black, Blue, Green, Orange, Pink, Red, Violet, White, Yellow"
folder_path = "database/female"
intro_path = "database/female_intro"
cloth_image = []
for i in range(1, 4):
with open(os.path.join(save_directory, f"prompt_{i}.txt"), "r") as file:
prompt__gen = file.read()
generate(lora_path, prompt__gen, i)
cloth_image.append(os.path.join(save_directory, f"cloth_{i*2-1}.jpeg"))
pic_path = pic_match(prompt__gen, cates, folder_path, intro_path, i)
cloth_image.append(pic_path)
with open(os.path.join(save_directory, f"cloth_intro_1.txt"), "r") as file:
cloth_intro = file.read()
return cloth_image, cloth_image[0], cloth_intro