diff --git a/guild.yml b/guild.yml
index f3488dd..800a292 100644
--- a/guild.yml
+++ b/guild.yml
@@ -1,4 +1,6 @@
 - package: nvae-package
+  data-files:
+    - vae/splits/*.json
 
 - model: nvae-mixture-logistic
   sourcecode:
diff --git a/prepare.py b/prepare.py
index 22efe06..c90c1a9 100644
--- a/prepare.py
+++ b/prepare.py
@@ -1,7 +1,7 @@
 import argparse
+from joblib import Parallel, delayed
 from pathlib import Path
 import shutil
-import pandas as pd
 import torchvision
 from tqdm import tqdm
 
@@ -15,22 +15,31 @@ def image_path(directory, index):
     return directory / f'{index}.png'
 
 
-def save_images(dataset, directory):
+def prepare_and_save(image, directory, index, left, top, crop_size):
+    return (
+        image.crop((left, top, left + crop_size, top + crop_size))
+        .resize((problem.settings.WIDTH, problem.settings.HEIGHT))
+        .save(image_path(directory, index))
+    )
+
+
+def save_images(dataset, directory, n_jobs):
     original_size = (178, 218)
     crop_size = 148
     left = (original_size[0] - crop_size) // 2
     top = (original_size[1] - crop_size) // 2
 
-    for index, (image, _) in enumerate(tqdm(dataset)):
-        (
-            image.crop((left, top, left + crop_size, top + crop_size))
-            .resize((problem.settings.WIDTH, problem.settings.HEIGHT))
-            .save(image_path(directory, index))
+    Parallel(n_jobs)(
+        delayed(prepare_and_save)(
+            image, directory, index, left, top, crop_size
         )
+        for index, (image, _) in enumerate(tqdm(dataset))
+    )
 
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
+    parser.add_argument('--n_jobs', default=4, type=int)
     args = parser.parse_args()
 
     dataset = torchvision.datasets.CelebA(
@@ -39,5 +48,5 @@ def save_images(dataset, directory):
 
     directory = Path('prepared')
     directory.mkdir(parents=True)
-    save_images(dataset, directory)
+    save_images(dataset, directory, args.n_jobs)
     shutil.rmtree(CACHE_ROOT)