【全网独家】AIGC 最佳实践：从文本描述生成高质量图像<广泛用于视觉创意和广告设计>（代码+测试部署）

from transformers import BertTokenizer, BertModel
import torch

# 加载预训练的 BERT 模型和 tokenizer
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
text_encoder = BertModel.from_pretrained('bert-base-uncased')

def encode_text(text):
    inputs = tokenizer(text, return_tensors='pt', padding=True, truncation=True, max_length=512)
    with torch.no_grad():
        outputs = text_encoder(**inputs)
    return outputs.last_hidden_state.mean(dim=1)  # 使用平均池化来获得文本嵌入

# 示例文本
text = "A futuristic cityscape at sunset."
text_vector = encode_text(text)
print(text_vector.shape)  # 输出为 (batch_size, hidden_size)

import torch.nn as nn

class ImageDecoder(nn.Module):
    def __init__(self, input_dim, output_channels, img_size):
        super(ImageDecoder, self).__init__()
        self.fc = nn.Linear(input_dim, 8192)
        self.deconv = nn.Sequential(
            nn.ConvTranspose2d(512, 256, kernel_size=4, stride=2, padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(),
            nn.ConvTranspose2d(256, 128, kernel_size=4, stride=2, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(),
            nn.ConvTranspose2d(128, 64, kernel_size=4, stride=2, padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(),
            nn.ConvTranspose2d(64, output_channels, kernel_size=4, stride=2, padding=1),
            nn.Tanh()
        )
        self.img_size = img_size

    def forward(self, x):
        x = self.fc(x)
        x = x.view(-1, 512, self.img_size // 8, self.img_size // 8)
        x = self.deconv(x)
        return x

# 初始化图像解码器
img_size = 64
image_decoder = ImageDecoder(input_dim=text_vector.size(1), output_channels=3, img_size=img_size)

# 解码文本向量
generated_image = image_decoder(text_vector)
print(generated_image.shape)  # 输出为 (batch_size, 3, img_size, img_size)

import numpy as np

def diffusion_process(image, steps=10, beta=0.1):
    noisy_images = [image]
    for step in range(steps):
        noise = torch.randn_like(image) * np.sqrt(beta)
        noisy_image = noisy_images[-1] + noise
        noisy_images.append(noisy_image)
    return noisy_images

def reverse_diffusion_process(noisy_images, steps=10, alpha=0.9):
    denoised_image = noisy_images[-1]
    denoised_images = [denoised_image]
    for step in range(steps - 1, 0, -1):
        denoised_image = denoised_images[-1] * alpha + noisy_images[step] * (1 - alpha)
        denoised_images.append(denoised_image)
    denoised_images.reverse()
    return denoised_images

# 生成初始图像
initial_image = generated_image

# 扩散过程
noisy_images = diffusion_process(initial_image, steps=10)

# 反向扩散过程
denoised_images = reverse_diffusion_process(noisy_images, steps=10)

# 最终生成的高质量图像
final_image = denoised_images[-1]
print(final_image.shape)  # 输出为 (batch_size, 3, img_size, img_size)

+--------------------+
| Input Text Prompt  |
+---------+----------+
          |
          v
    +-----+------+
    | Text       |
    | Encoder    |
    +-----+------+
          |
          v
  +-------+--------+
  | Text Embedding |
  +-------+--------+
          |
          v
    +-----+------+
    | Image      |
    | Decoder    |
    +-----+------+
          |
          v
  +-------+--------+
  | Generated Image|
  +-----------------+

pip install openai

import openai

# 设置 API 密钥
openai.api_key = 'YOUR_API_KEY'

def generate_image(prompt):
    response = openai.Image.create(
        prompt=prompt,
        n=1,
        size="512x512"
    )
    return response['data'][0]['url']

if __name__ == "__main__":
    prompt = "A futuristic cityscape at sunset."
    image_url = generate_image(prompt)
    print(f"Generated Image URL: {image_url}")

pip install Flask

from flask import Flask, request, jsonify
import openai

app = Flask(__name__)

# 设置 API 密钥
openai.api_key = 'YOUR_API_KEY'

@app.route('/generate-image', methods=['POST'])
def generate_image():
    data = request.json
    prompt = data.get('prompt', '')
    response = openai.Image.create(
        prompt=prompt,
        n=1,
        size="512x512"
    )
    return jsonify({'image_url': response['data'][0]['url']})

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=5000)

curl -X POST http://localhost:5000/generate-image -H "Content-Type: application/json" -d '{"prompt": "A cat sitting on a sofa."}'