diff --git a/examples/dreambooth/train_dreambooth_lora_flux2_img2img.py b/examples/dreambooth/train_dreambooth_lora_flux2_img2img.py
index 32bce9531b71..e5c092cb16ea 100644
--- a/examples/dreambooth/train_dreambooth_lora_flux2_img2img.py
+++ b/examples/dreambooth/train_dreambooth_lora_flux2_img2img.py
@@ -741,6 +741,9 @@ def __init__(
 
         self.buckets = buckets
 
+        # Initialize image processor for condition image preprocessing
+        self.image_processor = Flux2ImageProcessor()
+
         # if --dataset_name is provided or a metadata jsonl file is provided in the local --instance_data directory,
         # we load the training data using load_dataset
         if args.dataset_name is not None:
@@ -827,13 +830,13 @@ def __init__(
                 dest_image = self.cond_images[i]
                 image_width, image_height = dest_image.size
                 if image_width * image_height > 1024 * 1024:
-                    dest_image = Flux2ImageProcessor.image_processor._resize_to_target_area(dest_image, 1024 * 1024)
+                    dest_image = self.image_processor._resize_to_target_area(dest_image, 1024 * 1024)
                     image_width, image_height = dest_image.size
 
                 multiple_of = 2 ** (4 - 1)  # 2 ** (len(vae.config.block_out_channels) - 1), temp!
                 image_width = (image_width // multiple_of) * multiple_of
                 image_height = (image_height // multiple_of) * multiple_of
-                dest_image = Flux2ImageProcessor.image_processor.preprocess(
+                dest_image = self.image_processor.preprocess(
                     dest_image, height=image_height, width=image_width, resize_mode="crop"
                 )