diff --git a/XHS.py b/XHS.py
index 63c490a..118f1fa 100644
--- a/XHS.py
+++ b/XHS.py
@@ -454,43 +454,71 @@ def collect_videos_from_explore_cards(
     seen_urls: set[str],
     seen_files: set[Path],
 ) -> int:
-    note_urls = collect_note_urls_from_page(page, limit=max_videos * 8)
     downloaded = 0
-    for note_url in note_urls:
-        if downloaded >= max_videos:
-            break
-        page.get(note_url)
+    visited_note_ids: set[str] = set()
+    empty_rounds = 0
+    max_empty_rounds = 4
+
+    while downloaded < max_videos and empty_rounds < max_empty_rounds:
+        page.get(start_url)
         time.sleep(2)
-        candidates = group_video_candidates(
-            extract_video_candidates_from_html(
-                page.run_js("return document.documentElement.outerHTML"),
-                video_id=extract_note_id_from_url(note_url),
-            )
+        note_urls = filter_unvisited_note_urls(
+            collect_note_urls_from_page(page, limit=max_videos * 12),
+            visited_note_ids,
         )
-        if not candidates:
+        if not note_urls:
+            empty_rounds += 1
+            scroll_feed(page)
             continue
-        for candidate in candidates:
+
+        round_downloaded = 0
+        for note_url in note_urls:
             if downloaded >= max_videos:
                 break
-            if candidate.video_url in seen_urls:
-                continue
-            seen_urls.add(candidate.video_url)
-            output_path = build_output_path(candidate, output_dir=output_dir)
-            if output_path in seen_files or output_path.exists():
-                continue
-            try:
-                download_video(
-                    requests_module=requests_module,
-                    headers=build_headers(start_url),
-                    video_url=candidate.video_url,
-                    output_path=output_path,
+            note_id = extract_note_id_from_url(note_url)
+            visited_note_ids.add(note_id)
+            page.get(note_url)
+            time.sleep(2)
+            candidates = group_video_candidates(
+                extract_video_candidates_from_html(
+                    page.run_js("return document.documentElement.outerHTML"),
+                    video_id=note_id,
                 )
-            except Exception as exc:
-                print(f"[WARN] 下载失败 {candidate.video_id}: {exc}")
+            )
+            if not candidates:
                 continue
-            downloaded += 1
-            seen_files.add(output_path)
-            print(f"[OK] 已保存: {output_path}")
+            for candidate in candidates:
+                if downloaded >= max_videos:
+                    break
+                if candidate.video_url in seen_urls:
+                    continue
+                seen_urls.add(candidate.video_url)
+                output_path = build_output_path(candidate, output_dir=output_dir)
+                if output_path in seen_files or output_path.exists():
+                    continue
+                try:
+                    download_video(
+                        requests_module=requests_module,
+                        headers=build_headers(start_url),
+                        video_url=candidate.video_url,
+                        output_path=output_path,
+                    )
+                except Exception as exc:
+                    print(f"[WARN] 下载失败 {candidate.video_id}: {exc}")
+                    continue
+                downloaded += 1
+                round_downloaded += 1
+                seen_files.add(output_path)
+                print(f"[OK] 已保存: {output_path}")
+
+        if round_downloaded == 0:
+            empty_rounds += 1
+        else:
+            empty_rounds = 0
+        if downloaded < max_videos:
+            page.get(start_url)
+            time.sleep(1)
+            scroll_feed(page)
     return downloaded
 
 
@@ -529,6 +557,10 @@ return Array.from(document.querySelectorAll('a[href*="/explore/"]'))
     return normalize_note_urls([str(url) for url in raw_urls])[:limit]
 
 
+def filter_unvisited_note_urls(urls: list[str], visited_note_ids: set[str]) -> list[str]:
+    return [url for url in urls if extract_note_id_from_url(url) not in visited_note_ids]
+
+
 def build_parser() -> argparse.ArgumentParser:
     parser = argparse.ArgumentParser(description="附着到已登录小红书 Chrome，监听 feed 响应并下载视频")
     parser.add_argument("--max-videos", type=int, default=10, help="最多下载视频数量，默认 10")
diff --git a/test_xhs.py b/test_xhs.py
index cd1736c..34a3275 100644
--- a/test_xhs.py
+++ b/test_xhs.py
@@ -273,6 +273,17 @@ class XhsModuleTests(unittest.TestCase):
         )
         self.assertEqual(urls, ["https://www.xiaohongshu.com/explore/abc?xsec_token=token&xsec_source="])
 
+    def test_filter_unvisited_note_urls_skips_seen_note_ids(self) -> None:
+        module = importlib.import_module("XHS")
+        urls = [
+            "https://www.xiaohongshu.com/explore/abc?xsec_token=token",
+            "https://www.xiaohongshu.com/explore/def?xsec_token=token",
+        ]
+        self.assertEqual(
+            module.filter_unvisited_note_urls(urls, {"abc"}),
+            ["https://www.xiaohongshu.com/explore/def?xsec_token=token"],
+        )
+
 
 if __name__ == "__main__":
     unittest.main()