diff --git a/wan_4_frame_ultimate.py b/wan_4_frame_ultimate.py
index b508004..a5833e2 100644
--- a/wan_4_frame_ultimate.py
+++ b/wan_4_frame_ultimate.py
@@ -210,7 +210,11 @@ class WanFourFrameReferenceUltimate:
             "concat_mask": mask_low_reshaped
         })
         
-        negative_out = negative
+        # 修复：negative也设置图像条件（使用high noise的条件）
+        negative_out = node_helpers.conditioning_set_values(negative, {
+            "concat_latent_image": concat_latent_image_high,
+            "concat_mask": mask_high_reshaped
+        })
         
         clip_vision_output = self._merge_clip_vision_outputs(
             clip_vision_frame_1, clip_vision_frame_2, 
@@ -220,6 +224,9 @@ class WanFourFrameReferenceUltimate:
         if clip_vision_output is not None:
             positive_low_noise = node_helpers.conditioning_set_values(positive_low_noise, 
                                                                    {"clip_vision_output": clip_vision_output})
+            # 修复：negative也应用clip_vision_output
+            negative_out = node_helpers.conditioning_set_values(negative_out, 
+                                                             {"clip_vision_output": clip_vision_output})
         
         return (positive_high_noise, positive_low_noise, negative_out, {"samples": latent})