diff --git a/hyvideo/modules/models.py b/hyvideo/modules/models.py
index 0e76321..766eacf 100644
--- a/hyvideo/modules/models.py
+++ b/hyvideo/modules/models.py
@@ -1166,6 +1166,12 @@ class HYVideoDiffusionTransformer(ModelMixin, ConfigMixin):
                     self.teacache_skipped_steps_uncond += 1
                 else:
                     self.teacache_skipped_steps_cond += 1
+                
+                # Verify tensor dimensions match before adding
+                if img.shape == previous_residual.shape:
+                    img = img + previous_residual.to(img.device)
+                else:
+                    should_calc = True # Force recalculation if dimensions don't match
 
             if should_calc:
                 ori_img = img.clone()