| 1 | import os |
| 2 | import json |
| 3 | import logging |
| 4 | import time |
| 5 | import uuid |
| 6 | import threading |
| 7 | from contextlib import contextmanager |
| 8 | from typing import Any |
| 9 | try: |
| 10 | import dashscope |
| 11 | from dashscope import MultiModalConversation |
| 12 | from dashscope.aigc.image_generation import ImageGeneration |
| 13 | except ImportError: |
| 14 | dashscope = None |
| 15 | MultiModalConversation = None |
| 16 | ImageGeneration = None |
| 17 | try: |
| 18 | from .image_processor import ImageProcessor |
| 19 | except ImportError: |
| 20 | from image_processor import ImageProcessor |
| 21 | |
| 22 | class DashScopeClient: |
| 23 | _proxy_env_lock = threading.Lock() |
| 24 | |
| 25 | def __init__(self, api_key=None, base_url=None, local_proxy=None): |
| 26 | self.api_key = api_key or os.getenv("DASHSCOPE_API_KEY") |
| 27 | # 默认使用中国（北京）地域 API，如果环境变量或参数未设置则使用默认地址 |
| 28 | self.base_url = base_url or os.getenv("DASHSCOPE_BASE_URL") |
| 29 | self.local_proxy = local_proxy |
| 30 | if dashscope: |
| 31 | dashscope.api_key = self.api_key |
| 32 | dashscope.base_http_api_url = self.base_url |
| 33 | self.image_processor = ImageProcessor(local_proxy=local_proxy) |
| 34 | |
| 35 | @contextmanager |
| 36 | def _proxy_env(self): |
| 37 | if not self.local_proxy: |
| 38 | yield |
| 39 | return |
| 40 | |
| 41 | with self._proxy_env_lock: |
| 42 | keys = ("HTTP_PROXY", "HTTPS_PROXY", "http_proxy", "https_proxy") |
| 43 | old_values = {key: os.environ.get(key) for key in keys} |
| 44 | try: |
| 45 | for key in keys: |
| 46 | os.environ[key] = self.local_proxy |
| 47 | yield |
| 48 | finally: |
| 49 | for key, value in old_values.items(): |
| 50 | if value is None: |
| 51 | os.environ.pop(key, None) |
| 52 | else: |
| 53 | os.environ[key] = value |
| 54 | |
| 55 | def _extract_image_urls(self, payload: Any) -> list[str]: |
| 56 | """Extract image URLs from the different DashScope response shapes.""" |
| 57 | results = [] |
| 58 | |
| 59 | def to_plain(value): |
| 60 | if value is None: |
| 61 | return None |
| 62 | if isinstance(value, (str, int, float, bool)): |
| 63 | return value |
| 64 | if isinstance(value, list): |
| 65 | return [to_plain(item) for item in value] |
| 66 | if isinstance(value, tuple): |
| 67 | return [to_plain(item) for item in value] |
| 68 | if isinstance(value, dict): |
| 69 | return {key: to_plain(val) for key, val in value.items()} |
| 70 | if hasattr(value, "to_dict"): |
| 71 | return to_plain(value.to_dict()) |
| 72 | if hasattr(value, "__dict__"): |
| 73 | return { |
| 74 | key: to_plain(val) |
| 75 | for key, val in value.__dict__.items() |
| 76 | if not key.startswith("_") |
| 77 | } |
| 78 | return value |
| 79 | |
| 80 | def walk(value): |
| 81 | if isinstance(value, dict): |
| 82 | for key, val in value.items(): |
| 83 | if key in {"image", "url", "image_url"} and isinstance(val, str): |
| 84 | results.append(val) |
| 85 | else: |
| 86 | walk(val) |
| 87 | elif isinstance(value, list): |
| 88 | for item in value: |
| 89 | walk(item) |
| 90 | |
| 91 | walk(to_plain(payload)) |
| 92 | return [url for url in results if isinstance(url, str)] |
| 93 | |
| 94 | def generate_image(self, prompt, model="wan2.7-image", size="1024*1024", n=1, session_id=None, save_dir=None): |
| 95 | """ |
| 96 | Text to Image generation using DashScope |
| 97 | """ |
| 98 | if ImageGeneration is None: |
| 99 | raise RuntimeError("dashscope package not installed. Run: pip install dashscope") |
| 100 | |
| 101 | try: |
| 102 | messages = [{"role": "user", "content": [{"text": prompt}]}] |
| 103 | with self._proxy_env(): |
| 104 | response = ImageGeneration.call( |
| 105 | model=model, |
| 106 | api_key=self.api_key, |
| 107 | messages=messages, |
| 108 | n=n, |
| 109 | size=size, |
| 110 | watermark=False, |
| 111 | ) |
| 112 | |
| 113 | if response.status_code == 200: |
| 114 | results = self._extract_image_urls(getattr(response, "output", None)) |
| 115 | if not results: |
| 116 | raise RuntimeError(f"DashScope image generation returned no image URLs. output={getattr(response, 'output', None)}") |
| 117 | |
| 118 | # Check if we should download |
| 119 | if save_dir: |
| 120 | os.makedirs(save_dir, exist_ok=True) |
| 121 | local_files = [] |
| 122 | for i, url in enumerate(results): |
| 123 | file_name = f"ds_{session_id if session_id else 'nosess'}_{int(time.time())}_{i}_{uuid.uuid4().hex[:6]}.png" |
| 124 | file_path = os.path.join(save_dir, file_name) |
| 125 | if self.image_processor.download_image(url, file_path): |
| 126 | local_files.append(file_path) |
| 127 | return local_files |
| 128 | |
| 129 | return results |
| 130 | else: |
| 131 | raise RuntimeError(f"Image generation failed: {response.code}, {response.message}, status={response.status_code}") |
| 132 | except Exception as e: |
| 133 | logging.error(f"Error in generate_image (DashScope): {e}") |
| 134 | raise |
| 135 | |
| 136 | def edit_image(self, prompt, image_urls, model="wan2.7-image", size="1920*1080", n=1, session_id=None, save_dir=None): |
| 137 | """ |
| 138 | Image editing/compositing using DashScope ImageGeneration |
| 139 | """ |
| 140 | if ImageGeneration is None: |
| 141 | raise RuntimeError("dashscope package not installed. Run: pip install dashscope") |
| 142 | |
| 143 | # Prepare content |
| 144 | content_list = [] |
| 145 | for img_url in image_urls: |
| 146 | content_list.append({"image": img_url}) |
| 147 | content_list.append({"text": prompt}) |
| 148 | |
| 149 | messages = [ |
| 150 | { |
| 151 | "role": "user", |
| 152 | "content": content_list |
| 153 | } |
| 154 | ] |
| 155 | |
| 156 | try: |
| 157 | # Use ImageGeneration.call with messages, same as generate_image |
| 158 | with self._proxy_env(): |
| 159 | response = ImageGeneration.call( |
| 160 | model=model, |
| 161 | api_key=self.api_key, |
| 162 | messages=messages, |
| 163 | n=n, |
| 164 | size=size, |
| 165 | watermark=False, |
| 166 | ) |
| 167 | |
| 168 | if response.status_code == 200: |
| 169 | results = self._extract_image_urls(getattr(response, "output", None)) |
| 170 | if not results: |
| 171 | raise RuntimeError(f"DashScope image edit returned no image URLs. output={getattr(response, 'output', None)}") |
| 172 | |
| 173 | # Check if we should download |
| 174 | if save_dir: |
| 175 | os.makedirs(save_dir, exist_ok=True) |
| 176 | local_files = [] |
| 177 | for i, url in enumerate(results): |
| 178 | file_name = f"ds_{session_id if session_id else 'nosess'}_{int(time.time())}_{i}_{uuid.uuid4().hex[:6]}.png" |
| 179 | file_path = os.path.join(save_dir, file_name) |
| 180 | if self.image_processor.download_image(url, file_path): |
| 181 | local_files.append(file_path) |
| 182 | return local_files |
| 183 | |
| 184 | return results |
| 185 | else: |
| 186 | raise RuntimeError(f"Image edit failed: {response.code}, {response.message}, status={response.status_code}") |
| 187 | except Exception as e: |
| 188 | logging.error(f"Error in edit_image: {e}") |
| 189 | raise |
| 190 | |
| 191 | |
| 192 | if __name__ == "__main__": |
| 193 | import sys |
| 194 | sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) |
| 195 | from config import Config |
| 196 | |
| 197 | print("=== DashScope 图片生成可用性测试 ===") |
| 198 | MODELS=["wan2.6-t2i", "wan2.7-image", "wan2.7-image-pro"] |
| 199 | save_dir = "code/result/image/test_avail" |
| 200 | api_key = Config.DASHSCOPE_API_KEY |
| 201 | base_url = Config.DASHSCOPE_BASE_URL |
| 202 | if not api_key: |
| 203 | print("✗ DASHSCOPE_API_KEY 未设置，跳过") |
| 204 | sys.exit(1) |
| 205 | print(f" API Key: {api_key[:6]}***{api_key[-4:]}") |
| 206 | print(f" Base URL: {base_url}") |
| 207 | client = DashScopeClient(api_key=api_key, base_url=base_url) |
| 208 | |
| 209 | # 文生图 |
| 210 | print("\n=== 文生图测试 ===") |
| 211 | prompt = "一只橘猫躺在阳光下的窗台上，水彩画风格" |
| 212 | for model in MODELS: |
| 213 | print(f"\nPrompt: {prompt}") |
| 214 | print(f"model: {model}") |
| 215 | os.makedirs(save_dir, exist_ok=True) |
| 216 | t0 = time.time() |
| 217 | try: |
| 218 | paths = client.generate_image( |
| 219 | prompt=prompt, model=model, |
| 220 | size="1024*1024", save_dir=save_dir, |
| 221 | ) |
| 222 | elapsed = time.time() - t0 |
| 223 | if paths: |
| 224 | print(f"✓ 生成 {len(paths)} 张图片 ({elapsed:.1f}s): {paths}") |
| 225 | else: |
| 226 | print(f"✗ 返回空列表 ({elapsed:.1f}s)") |
| 227 | except Exception as e: |
| 228 | print(f"✗ 失败: {e}") |
| 229 | sys.exit(1) |
| 230 | |
| 231 | # 图生图 |
| 232 | print("\n=== 图生图测试 ===") |
| 233 | img_path = "code/result/image/test_avail/test_input.png" |
| 234 | prompt = "在这张图片的基础上，添加一些飞舞的樱花花瓣，绘制为水彩画风格" |
| 235 | for model in MODELS: |
| 236 | print(f"\nPrompt: {prompt}") |
| 237 | print(f"model: {model}") |
| 238 | os.makedirs(save_dir, exist_ok=True) |
| 239 | t0 = time.time() |
| 240 | try: |
| 241 | paths = client.edit_image( |
| 242 | prompt=prompt, image_urls=[img_path], model=model, |
| 243 | size="1024*1024", save_dir=save_dir, |
| 244 | ) |
| 245 | elapsed = time.time() - t0 |
| 246 | if paths: |
| 247 | print(f"✓ 生成 {len(paths)} 张图片 ({elapsed:.1f}s): {paths}") |
| 248 | else: |
| 249 | print(f"✗ 返回空列表 ({elapsed:.1f}s)") |
| 250 | except Exception as e: |
| 251 | print(f"✗ 失败: {e}") |
| 252 | sys.exit(1) |
| 253 |