| 1 | import os |
| 2 | import sys |
| 3 | |
| 4 | models_dir = os.path.dirname(os.path.abspath(__file__)) |
| 5 | backend_dir = os.path.dirname(models_dir) |
| 6 | if backend_dir not in sys.path: |
| 7 | sys.path.insert(0, backend_dir) |
| 8 | |
| 9 | import time |
| 10 | import uuid |
| 11 | import base64 |
| 12 | import httpx |
| 13 | import logging |
| 14 | from openai import OpenAI |
| 15 | from config import Config |
| 16 | try: |
| 17 | from models.image_processor import ImageProcessor |
| 18 | except ImportError: |
| 19 | from image_processor import ImageProcessor |
| 20 | |
| 21 | logger = logging.getLogger(__name__) |
| 22 | |
| 23 | |
| 24 | class ImageGPT: |
| 25 | """ |
| 26 | OpenAI 图片生成客户端 |
| 27 | 支持模型： |
| 28 | - sora_image → Images API |
| 29 | - gpt-image-2 → Responses API |
| 30 | """ |
| 31 | def __init__(self, |
| 32 | api_key: str = None, |
| 33 | base_url: str = None, |
| 34 | proxy: str = None, |
| 35 | timeout: float = 300.0): |
| 36 | """ |
| 37 | OpenAI 图片生成客户端 |
| 38 | :param api_key: API Key |
| 39 | :param base_url: 自定义 Base URL |
| 40 | :param proxy: 当前 provider 显式启用时使用的代理 |
| 41 | :param timeout: 超时时间 |
| 42 | """ |
| 43 | self.api_key = api_key or Config.OPENAI_API_KEY |
| 44 | self.timeout = timeout |
| 45 | |
| 46 | kwargs = {"api_key": self.api_key, "timeout": self.timeout} |
| 47 | |
| 48 | self.base_url = base_url |
| 49 | if proxy is None: |
| 50 | proxy = Config.provider_proxy("openai") |
| 51 | if proxy: |
| 52 | kwargs["http_client"] = httpx.Client( |
| 53 | proxy=proxy, |
| 54 | timeout=self.timeout, |
| 55 | ) |
| 56 | if self.base_url: |
| 57 | kwargs["base_url"] = self.base_url |
| 58 | |
| 59 | self.client = OpenAI(**kwargs) |
| 60 | self.max_attempts = 10 |
| 61 | self.image_processor = ImageProcessor() |
| 62 | |
| 63 | def _encode_image_to_base64(self, image_path: str) -> str: |
| 64 | """将本地图片转换为 Base64 编码""" |
| 65 | if not image_path or not os.path.exists(image_path): |
| 66 | return image_path |
| 67 | |
| 68 | try: |
| 69 | with open(image_path, "rb") as f: |
| 70 | img_data = base64.b64encode(f.read()).decode("utf-8") |
| 71 | ext = os.path.splitext(image_path)[1].lower().replace(".", "") |
| 72 | if ext not in ["png", "jpg", "jpeg", "webp"]: |
| 73 | ext = "png" |
| 74 | return f"data:image/{ext};base64,{img_data}" |
| 75 | except Exception as e: |
| 76 | logger.warning("Failed to encode image %s: %s", image_path, e) |
| 77 | return image_path |
| 78 | |
| 79 | def generate_image(self, prompt, size="1024x1024", quality="high", model="gpt-image-2", |
| 80 | save_dir=None, image_urls=None): |
| 81 | """Generate a single image, download it, and return the local file path. |
| 82 | |
| 83 | Args: |
| 84 | prompt: 图片描述提示词 |
| 85 | size: 图片尺寸 |
| 86 | quality: 图片质量 |
| 87 | model: 模型名称 (sora_image / gpt-image-2) |
| 88 | save_dir: 保存目录（不传则返回 URL 或 base64） |
| 89 | image_urls: 参考图片 URL 列表（仅 gpt-image-2 支持） |
| 90 | """ |
| 91 | |
| 92 | attempts = 0 |
| 93 | last_error = None |
| 94 | |
| 95 | # 处理参考图片 |
| 96 | extra_body = {} |
| 97 | if image_urls and isinstance(image_urls, list) and len(image_urls) > 0: |
| 98 | # 中转站通常支持通过 extra_body 传递 image_url 或 ref_image |
| 99 | # 这里我们将第一张图作为参考图 |
| 100 | ref_images = [self._encode_image_to_base64(image_urls[i]) for i in range(min(len(image_urls), 6))] |
| 101 | extra_body = {"image_url": ref_images} |
| 102 | |
| 103 | while attempts < self.max_attempts: |
| 104 | try: |
| 105 | response = self.client.images.generate( |
| 106 | model=model, |
| 107 | prompt=prompt, |
| 108 | size=size, |
| 109 | quality=quality, |
| 110 | n=1, |
| 111 | extra_body=extra_body |
| 112 | ) |
| 113 | |
| 114 | if not response or not response.data: |
| 115 | raise RuntimeError("OpenAI API 返回数据为空") |
| 116 | |
| 117 | img_data = response.data[0] |
| 118 | file_path = None |
| 119 | |
| 120 | # 1. 处理 Base64 格式 (中转站常用) |
| 121 | if hasattr(img_data, 'b64_json') and img_data.b64_json: |
| 122 | if save_dir: |
| 123 | os.makedirs(save_dir, exist_ok=True) |
| 124 | file_name = f"gpt_{int(time.time())}_{uuid.uuid4().hex[:6]}.png" |
| 125 | file_path = os.path.join(save_dir, file_name) |
| 126 | with open(file_path, "wb") as f: |
| 127 | f.write(base64.b64decode(img_data.b64_json)) |
| 128 | return file_path |
| 129 | return img_data.b64_json |
| 130 | |
| 131 | # 2. 处理 URL 格式 |
| 132 | elif hasattr(img_data, 'url') and img_data.url: |
| 133 | url = img_data.url |
| 134 | if save_dir: |
| 135 | os.makedirs(save_dir, exist_ok=True) |
| 136 | file_name = f"gpt_{int(time.time())}_{uuid.uuid4().hex[:6]}.png" |
| 137 | file_path = os.path.join(save_dir, file_name) |
| 138 | if self.image_processor.download_image(url, file_path, proxies=Config.requests_proxies("openai")): |
| 139 | return file_path |
| 140 | return url |
| 141 | |
| 142 | raise RuntimeError("未在响应中找到 url 或 b64_json") |
| 143 | except Exception as e: |
| 144 | last_error = e |
| 145 | # Other errors: wait before retry |
| 146 | logger.warning("OpenAI image generation failed; retrying in 10 seconds: %s", e) |
| 147 | time.sleep(10) |
| 148 | break # Break inner loop to retry all models |
| 149 | attempts += 1 |
| 150 | raise Exception(f"Max attempts reached, failed to generate image. Last error: {last_error}") |
| 151 | |
| 152 | def generate_images(self, prompt, count=4, size="1024x1024", quality="standard", model=None): |
| 153 | """Generate multiple image URLs by calling Images API 'count' times.""" |
| 154 | urls = [] |
| 155 | for _ in range(count): |
| 156 | url = self.generate_image(prompt=prompt, size=size, quality=quality, model=model) |
| 157 | urls.append(url) |
| 158 | return urls |
| 159 | |
| 160 | |
| 161 | if __name__ == "__main__": |
| 162 | import sys |
| 163 | import tempfile |
| 164 | sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) |
| 165 | from config import Config |
| 166 | |
| 167 | MODELS = ["gpt-image-2"] |
| 168 | save_dir = "code/result/image/test_avail" |
| 169 | api_key = Config.OPENAI_API_KEY |
| 170 | base_url = Config.OPENAI_BASE_URL |
| 171 | if not api_key: |
| 172 | print("✗ OPENAI_API_KEY 未设置，跳过") |
| 173 | sys.exit(1) |
| 174 | print("=== GPT 图片生成测试 ===") |
| 175 | print(f" API Key: {api_key[:6]}***") |
| 176 | print(f" Base URL: {base_url}") |
| 177 | |
| 178 | |
| 179 | # 文生图 |
| 180 | print("\n=== GPT 文生图可用性测试 ===") |
| 181 | img_prompt = "A cute orange cat lying on a sunny windowsill, watercolor style" |
| 182 | img_path = "" |
| 183 | client = ImageGPT(api_key=api_key, base_url=Config.OPENAI_BASE_URL, proxy=Config.provider_proxy("openai")) |
| 184 | for model in MODELS: |
| 185 | print(f"\nTesting model: {model}") |
| 186 | print(f"Prompt: {img_prompt}") |
| 187 | print(f"Image path: {img_path}") |
| 188 | client.max_attempts = 1 |
| 189 | t0 = time.time() |
| 190 | os.makedirs(save_dir, exist_ok=True) |
| 191 | try: |
| 192 | path = client.generate_image(prompt=img_prompt, size="1024x1024", |
| 193 | model=model, save_dir=save_dir) |
| 194 | elapsed = time.time() - t0 |
| 195 | print(f"✓ 生成成功 ({elapsed:.1f}s): {path}") |
| 196 | except Exception as e: |
| 197 | elapsed = time.time() - t0 |
| 198 | print(f"✗ 失败 ({elapsed:.1f}s): {e}") |
| 199 | |
| 200 | # 图生图 |
| 201 | print("\n=== GPT 图生图可用性测试 ===") |
| 202 | img_prompt = "Turn this cat into a cute cartoon character with big eyes and a playful expression" |
| 203 | img_path = "code/result/image/test_avail/test_input.jpg" |
| 204 | for model in MODELS: |
| 205 | print(f"\nTesting model: {model}") |
| 206 | print(f"Prompt: {img_prompt}") |
| 207 | print(f"Image path: {img_path}") |
| 208 | client.max_attempts = 1 |
| 209 | t0 = time.time() |
| 210 | os.makedirs(save_dir, exist_ok=True) |
| 211 | try: |
| 212 | path = client.generate_image(prompt=img_prompt, size="1024x1024", |
| 213 | model=model, save_dir=save_dir, image_urls=[img_path]) |
| 214 | elapsed = time.time() - t0 |
| 215 | print(f"✓ 生成成功 ({elapsed:.1f}s): {path}") |
| 216 | except Exception as e: |
| 217 | elapsed = time.time() - t0 |
| 218 | print(f"✗ 失败 ({elapsed:.1f}s): {e}") |
| 219 |