diff --git a/apps/worker/worker_app/tasks/viral_video.py b/apps/worker/worker_app/tasks/viral_video.py
index fb9892ae5..b02a6d4a7 100644
--- a/apps/worker/worker_app/tasks/viral_video.py
+++ b/apps/worker/worker_app/tasks/viral_video.py
@@ -1162,6 +1162,7 @@ def _step_render(job: ViralVideoJob, copy_result: dict, tts_audio_url: str | Non
     )
     logger.info("[爆款视频] Seedance prompt (前300字): %s", prompt[:300])
 
+    # 第一次调用：带参考图/首帧/音频/参考视频
     result = call_video_generation(
         prompt=prompt,
         image_url=first_image,
@@ -1175,8 +1176,12 @@ def _step_render(job: ViralVideoJob, copy_result: dict, tts_audio_url: str | Non
         reference_audios=ref_audios,
         reference_videos=ref_videos,
     )
-    if not result or not isinstance(result, dict):
-        # 读取详细错误信息（HTTP 状态码、响应 body、分类后的用户提示），不再笼统显示"返回为空"
+
+    # 肖像/内容审核拦截降级重试：去掉参考图/参考视频走纯文字 t2v 再试一次
+    # （不影响音频——去掉首帧和多参考图通常就能绕过真人照片审核，纯文字描述真人是允许的）
+    def _check_and_reraise(result):
+        if result and isinstance(result, dict):
+            return result
         from packages.shared.ai_service import get_last_video_error
 
         err = get_last_video_error() or {}
@@ -1196,6 +1201,33 @@ def _step_render(job: ViralVideoJob, copy_result: dict, tts_audio_url: str | Non
             (detail or "")[:500],
         )
         raise RuntimeError(err_msg)
+
+    if not result or not isinstance(result, dict):
+        from packages.shared.ai_service import get_last_video_error
+
+        err = get_last_video_error() or {}
+        err_code = err.get("error_code") or "unknown"
+        is_portrait = err_code == "portrait_intercept" or any(
+            k in (err.get("detail", "") or "").lower()
+            for k in ("real person", "privacy", "portrait", "sensitivecontent", "face", "人脸", "肖像", "真人")
+        )
+        if is_portrait and (first_image or rest_images or ref_videos):
+            logger.warning("[爆款视频] 参考素材触发真人/肖像审核拦截(error_code=%s)，降级为纯文字t2v重试一次", err_code)
+            # 去掉所有参考图/参考视频，纯 t2v 泛化生成；音频仍保留（不影响肖像审核）
+            result = call_video_generation(
+                prompt=prompt,
+                image_url=None,
+                duration=dur,
+                ratio=ratio,
+                resolution=resolution,
+                output_dir=str(tmpdir),
+                model=model,
+                generate_audio=gen_audio,
+                reference_images=[],
+                reference_audios=ref_audios,
+                reference_videos=[],
+            )
+        _check_and_reraise(result)
     video_path = result.get("video_path") or ""
     usage = result.get("usage")
     if not video_path or not Path(video_path).exists() or Path(video_path).stat().st_size == 0: