diff --git a/apps/stories/management/commands/generate_default_covers.py b/apps/stories/management/commands/generate_default_covers.py
new file mode 100644
index 0000000..ad7c64d
--- /dev/null
+++ b/apps/stories/management/commands/generate_default_covers.py
@@ -0,0 +1,116 @@
+"""
+用新的 LLM 提炼逻辑重新生成默认故事封面并上传到 OSS。
+
+使用方法:
+    python manage.py generate_default_covers
+    python manage.py generate_default_covers --dry-run   # 仅打印提炼到的描述，不生成图片
+"""
+import uuid
+import logging
+
+import requests
+from django.conf import settings
+from django.core.management.base import BaseCommand
+
+from apps.stories.utils import DEFAULT_STORIES
+from apps.stories.services.llm_service import _extract_image_description
+
+logger = logging.getLogger(__name__)
+
+# 每个默认故事对应的 OSS key（与 utils.py 中的 cover_url 一致）
+DEFAULT_COVER_KEYS = {
+    "失控的魔法扫帚": "stories/defaults/失控的魔法扫帚_cover.png",
+}
+
+
+class Command(BaseCommand):
+    help = "用 LLM 提炼故事画面描述后调用 Seedream 4.5 重新生成默认故事封面"
+
+    def add_arguments(self, parser):
+        parser.add_argument(
+            "--dry-run",
+            action="store_true",
+            help="仅打印 LLM 提炼的画面描述，不实际生成图片",
+        )
+
+    def handle(self, *args, **options):
+        dry_run = options["dry_run"]
+        config = settings.LLM_CONFIG
+
+        if not config.get("API_KEY"):
+            self.stderr.write(self.style.ERROR("VOLCENGINE_API_KEY 未配置"))
+            return
+
+        try:
+            from volcenginesdkarkruntime import Ark
+        except ImportError:
+            self.stderr.write(self.style.ERROR("volcengine SDK 未安装"))
+            return
+
+        try:
+            from utils.oss import get_oss_client
+            import oss2
+        except ImportError:
+            self.stderr.write(self.style.ERROR("oss2 未安装"))
+            return
+
+        client = Ark(api_key=config["API_KEY"])
+        image_model = config.get("IMAGE_MODEL_NAME", "doubao-seedream-4-5-251128")
+        image_size = config.get("IMAGE_SIZE", "2560x1440")
+        oss_config = settings.ALIYUN_OSS
+        oss_base = f"https://{oss_config['BUCKET_NAME']}.{oss_config['ENDPOINT']}"
+
+        for story in DEFAULT_STORIES:
+            title = story["title"]
+            content = story["content"]
+            oss_key = DEFAULT_COVER_KEYS.get(title)
+
+            if not oss_key:
+                self.stdout.write(self.style.WARNING(f"[{title}] 未找到对应 OSS key，跳过"))
+                continue
+
+            self.stdout.write(f"\n[{title}]")
+
+            # Step 1: LLM 提炼画面描述
+            self.stdout.write("  正在用 LLM 提炼画面描述...")
+            scene_desc = _extract_image_description(
+                title, content, client, config["MODEL_NAME"]
+            )
+            self.stdout.write(self.style.SUCCESS(f"  画面描述（{len(scene_desc)} 字）：{scene_desc}"))
+
+            if dry_run:
+                self.stdout.write(self.style.NOTICE("  [dry-run] 跳过图片生成"))
+                continue
+
+            # Step 2: 文生图
+            image_prompt = (
+                f"儿童绘本封面插画，{scene_desc}，卡通可爱风格，色彩明亮鲜艳，高质量插画"
+            )
+            self.stdout.write(f"  正在生成封面图（{image_size}）...")
+            result = client.images.generate(
+                model=image_model,
+                prompt=image_prompt,
+                size=image_size,
+                response_format="url",
+                watermark=False,
+            )
+            image_url = result.data[0].url
+            self.stdout.write(f"  临时图片 URL: {image_url[:80]}...")
+
+            # Step 3: 下载图片
+            self.stdout.write("  正在下载图片...")
+            resp = requests.get(image_url, timeout=60)
+            resp.raise_for_status()
+
+            # Step 4: 覆盖上传到 OSS
+            self.stdout.write(f"  正在上传到 OSS: {oss_key}")
+            oss_client = get_oss_client()
+            oss_client.bucket.put_object(
+                oss_key,
+                resp.content,
+                headers={"Content-Type": "image/jpeg"},
+            )
+            final_url = f"{oss_base}/{oss_key}"
+            self.stdout.write(self.style.SUCCESS(f"  ✓ 封面已更新: {final_url}"))
+
+        self.stdout.write(self.style.SUCCESS("\n完成。"))