diff --git a/examples/instruct_pix2pix/train_instruct_pix2pix.py b/examples/instruct_pix2pix/train_instruct_pix2pix.py
index d7f1288f3804..d1caf281a2c5 100644
--- a/examples/instruct_pix2pix/train_instruct_pix2pix.py
+++ b/examples/instruct_pix2pix/train_instruct_pix2pix.py
@@ -695,7 +695,7 @@ def preprocess_images(examples):
         )
         # We need to ensure that the original and the edited images undergo the same
         # augmentation transforms.
-        images = np.concatenate([original_images, edited_images])
+        images = np.stack([original_images, edited_images])
         images = torch.tensor(images)
         images = 2 * (images / 255) - 1
         return train_transforms(images)
@@ -706,7 +706,7 @@ def preprocess_train(examples):
         # Since the original and edited images were concatenated before
         # applying the transformations, we need to separate them and reshape
         # them accordingly.
-        original_images, edited_images = preprocessed_images.chunk(2)
+        original_images, edited_images = preprocessed_images
         original_images = original_images.reshape(-1, 3, args.resolution, args.resolution)
         edited_images = edited_images.reshape(-1, 3, args.resolution, args.resolution)
 
diff --git a/examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py b/examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py
index fafc50d092fb..5f01e2f2bb09 100644
--- a/examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py
+++ b/examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py
@@ -766,7 +766,7 @@ def preprocess_images(examples):
         )
         # We need to ensure that the original and the edited images undergo the same
         # augmentation transforms.
-        images = np.concatenate([original_images, edited_images])
+        images = np.stack([original_images, edited_images])
         images = torch.tensor(images)
         images = 2 * (images / 255) - 1
         return train_transforms(images)
@@ -906,7 +906,7 @@ def preprocess_train(examples):
         # Since the original and edited images were concatenated before
         # applying the transformations, we need to separate them and reshape
         # them accordingly.
-        original_images, edited_images = preprocessed_images.chunk(2)
+        original_images, edited_images = preprocessed_images
         original_images = original_images.reshape(-1, 3, args.resolution, args.resolution)
         edited_images = edited_images.reshape(-1, 3, args.resolution, args.resolution)