import json import os import shutil import subprocess import asyncio schema = { "type": "object", "properties": { "file_path": {"type": "string"}, "success": {"type": "boolean"} }, "required": ["file_path", "success"] } php_bin = "D:/Softwares/phpstudy_pro/Extensions/php/php7.4.3nts/php.exe" out_dir = "D:/Softwares/phpstudy_pro/WWW/fenleixinxi/tmp_img/downloaded" os.makedirs(out_dir, exist_ok=True) # 1. 读取 prompt 映射字典 prompt_map_file = "D:/Softwares/phpstudy_pro/WWW/fenleixinxi/tmp_img/prompt_map.json" with open(prompt_map_file, "r", encoding="utf-8") as f: prompts_list = json.load(f) prompt_dict = {str(item["id"]): item["prompt"] for item in prompts_list} # 2. 从 PHP 脚本获取当前在数据库中真正尚未完成 OSS 入库的分类列表 proc = subprocess.run([php_bin, "D:/Softwares/phpstudy_pro/WWW/fenleixinxi/tmp_img/get_pending.php"], capture_output=True, text=True) pending_categories = json.loads(proc.stdout) if proc.stdout else [] log(f"Current pending in DB: {len(pending_categories)}") # 每次执行的总项数(默认 20 项,分为多小批平稳生成) total_to_process = args.get("count", 20) if "args" in globals() and isinstance(args, dict) else 20 items_to_run = pending_categories[:total_to_process] mini_batch_size = 3 # 每小批 3 个并发,稳健不触发 QPS 限流 log(f"Will process total {len(items_to_run)} items in chunks of {mini_batch_size}") async def generate_single(cat): cat_id = str(cat["id"]) name = cat["name"] prompt = prompt_dict.get(cat_id) if not prompt: log(f"[WARN] No prompt for #{cat_id} {name}") return False dest_path = os.path.join(out_dir, f"{cat_id}.jpg") agent_prompt = ( f"请生成分类『{name}』的配图,提示词:{prompt}\n" f"要求:1:1方形构图,写实摄影质感,绝对不要包含任何文字、水印、LOGO、AI标识或角标。\n" f"生成后通过 finish 工具返回保存的文件本地路径 file_path 和 success: true。" ) try: res = await agent( prompt=agent_prompt, schema=schema, type_name="image-generator", role=f"Gen-{name}", model="flash" ) if res and res.get("success") and res.get("file_path"): src_file = res["file_path"] if os.path.exists(src_file) and os.path.getsize(src_file) > 50000: shutil.copyfile(src_file, dest_path) log(f"[OK] #{cat_id} {name} -> {dest_path} (size: {os.path.getsize(dest_path)})") return True log(f"[FAIL] #{cat_id} {name} -> Invalid result: {res}") return False except Exception as e: log(f"[ERROR] #{cat_id} {name} -> {str(e)}") return False # 分小批逐步执行 total_succeeded = 0 for i in range(0, len(items_to_run), mini_batch_size): chunk = items_to_run[i:i + mini_batch_size] log(f"--- Running chunk {i // mini_batch_size + 1}: {[c['name'] for c in chunk]} ---") chunk_results = await parallel(chunk, generate_single) chunk_ok = sum(1 for r in chunk_results if r) total_succeeded += chunk_ok log(f"Chunk finished: {chunk_ok}/{len(chunk)} succeeded. (Cumulative: {total_succeeded}/{len(items_to_run)})") # 每一小批完成立即自动推送到 OSS 入库,确保成果实时落盘 if chunk_ok > 0: ingest_cmd = [php_bin, "think", "fc:category-images"] ingest_res = subprocess.run(ingest_cmd, cwd="D:/Softwares/phpstudy_pro/WWW/fenleixinxi", capture_output=True, text=True) log(f"Incremental Ingest: {ingest_res.stdout.strip().splitlines()[-2:] if ingest_res.stdout else 'None'}") await asyncio.sleep(2) # 平稳休眠 2 秒,防触发突发限速 log(f"All chunks in this run completed! Total success in this run: {total_succeeded}")