#!/usr/bin/env python3 """ 重新生成第 13-14 页 - 使用优化后的角色一致性提示词 图片质量：1k (512x512) """ import os import requests import json import base64 import time from pathlib import Path from datetime import datetime # API 配置 OPENROUTER_API_KEY = os.environ.get('OPENROUTER_API_KEY') API_URL = "https://openrouter.ai/api/v1/chat/completions" MODEL = "google/gemini-3.1-flash-image-preview" # 输出目录 OUTPUT_DIR = Path("/root/.openclaw/workspace/trilingual-picturebook/output") OUTPUT_DIR.mkdir(exist_ok=True) # 角色参考图路径 CHARACTER_SHEET_DIR = Path("/root/.openclaw/workspace/trilingual-picturebook") YANGYANG_REF = CHARACTER_SHEET_DIR / "_charsheet_yangyang.png" ZHUAZHUA_REF = CHARACTER_SHEET_DIR / "_charsheet_zhuazhua.png" # 图像分辨率 - 1k 质量 IMAGE_SIZE = "512x512" def load_image_as_base64(image_path: Path) -> str: """加载图片为 base64 data URL""" with open(image_path, "rb") as f: image_data = f.read() base64_data = base64.b64encode(image_data).decode('utf-8') return f"data:image/png;base64,{base64_data}" def generate_image(prompt_text: str, filename: str, use_references: bool = True): """生成单张图片""" print(f"\n📸 生成：{filename}") headers = { "Authorization": f"Bearer {OPENROUTER_API_KEY}", "Content-Type": "application/json", "HTTP-Referer": "https://openclaw.ai", "X-Title": "Picturebook Regenerator" } # 构建消息内容 content_parts = [] # 添加参考图（如果使用） if use_references: if YANGYANG_REF.exists(): print(f" 📎 参考图：yangyang") content_parts.append({ "type": "image_url", "image_url": load_image_as_base64(YANGYANG_REF) }) if ZHUAZHUA_REF.exists(): print(f" 📎 参考图：zhuazhua") content_parts.append({ "type": "image_url", "image_url": load_image_as_base64(ZHUAZHUA_REF) }) # 添加文字提示词 - 强调角色一致性 content_parts.append({ "type": "text", "text": f"""IMPORTANT: Use the provided character reference images to maintain consistent character appearance. {prompt_text} Style: children's book illustration, watercolor style, soft pastel colors, warm lighting Characters: Keep consistent with reference images (2 year old Chinese twins) Composition: square, no text, no watermark, high quality Resolution: {IMAGE_SIZE}""" }) payload = { "model": MODEL, "messages": [ { "role": "user", "content": content_parts } ], "max_tokens": 1000 } # 发送请求（带重试） max_retries = 5 base_delay = 10 for attempt in range(max_retries): try: response = requests.post(API_URL, headers=headers, json=payload, timeout=120) if response.status_code == 429: wait_time = base_delay * (2 ** attempt) print(f" ⏳ 限流，等待 {wait_time} 秒...") time.sleep(wait_time) continue response.raise_for_status() result = response.json() # 提取图片 if "choices" in result and len(result["choices"]) > 0: message = result["choices"][0].get("message", {}) if "images" in message and len(message["images"]) > 0: image_url = message["images"][0].get("image_url", "") if isinstance(image_url, dict): image_url = image_url.get("url", "") if image_url.startswith("data:image"): image_data = base64.b64decode(image_url.split(",")[1]) else: image_resp = requests.get(image_url, timeout=30) image_data = image_resp.content # 保存图片 output_path = OUTPUT_DIR / f"{filename}.png" with open(output_path, "wb") as f: f.write(image_data) file_size_kb = len(image_data) / 1024 print(f" ✅ 已保存：{filename} ({file_size_kb:.1f} KB)") return {"success": True, "path": str(output_path), "size_kb": file_size_kb} print(f" ❌ 未返回图片数据") return {"success": False, "error": "No image data"} except Exception as e: print(f" ❌ 错误：{e}") if attempt < max_retries - 1: wait_time = base_delay * (2 ** attempt) time.sleep(wait_time) else: return {"success": False, "error": str(e)} return {"success": False, "error": "Max retries exceeded"} def main(): print("=" * 60) print("📖 重新生成第 13-14 页（优化角色一致性）") print("=" * 60) print(f"📁 输出目录：{OUTPUT_DIR}") print(f"📐 图片尺寸：{IMAGE_SIZE}") print(f"📎 角色参考：yangyang + zhuazhua") # 第 13 页提示词 - 优化版 prompt_13 = """ Scene: Twin toddlers holding hands, walking down wooden stairs together Character details (MUST match reference images): - Boy (yangyang): 2 year old Chinese boy, short black hair, round face, wearing light blue outfit, leading the way - Girl (zhuazhua): 2 year old Chinese girl, black hair with two small pigtails, round face, wearing pink dress, following brother Emotion: Warm sibling bond, brother protecting sister, gentle morning light Action: Boy holds girl's hand, carefully guiding her down the stairs, both looking down at steps Background: Wooden staircase in home, soft morning sunlight from window, warm cozy atmosphere Style notes: Watercolor illustration for toddlers, soft edges, pastel colors, square composition """ # 第 14 页提示词 - 优化版 prompt_14 = """ Scene: Golden retriever puppy greeting the twins at bottom of stairs Character details (MUST match reference images): - Boy (yangyang): 2 year old Chinese boy, short black hair, blue outfit, visible on stairs above - Girl (zhuazhua): 2 year old Chinese girl, black hair with pigtails, pink dress, visible on stairs above - Puppy: Cute golden retriever puppy, sitting at bottom of stairs, looking up excitedly Emotion: Joyful greeting, puppy happy to see the children, tail wagging Action: Puppy looking up at the twins with happy expression, twins about to reach bottom of stairs Background: Bottom of wooden staircase, home interior, warm morning light Style notes: Watercolor illustration for toddlers, soft edges, pastel colors, square composition """ results = [] # 生成第 13 页 result_13 = generate_image(prompt_13, "gemini-page-13-retry", use_references=True) results.append(("gemini-page-13-retry", result_13)) # 等待一下 time.sleep(3) # 生成第 14 页 result_14 = generate_image(prompt_14, "gemini-page-14-retry", use_references=True) results.append(("gemini-page-14-retry", result_14)) # 保存结果日志 log_path = OUTPUT_DIR / f"regenerate-log-{datetime.now().strftime('%Y%m%d-%H%M%S')}.json" with open(log_path, "w", encoding="utf-8") as f: json.dump(results, f, ensure_ascii=False, indent=2) print(f"\n📋 结果日志：{log_path}") print("\n" + "=" * 60) print("✅ 生成完成！") print("=" * 60) if __name__ == "__main__": main()