diff --git a/.gitignore b/.gitignore
index e9c99a8ae..7f5b5ceb1 100644
--- a/.gitignore
+++ b/.gitignore
@@ -14,7 +14,9 @@ store/
 unittest_prediction
 unittest_prediction_fast
 unittest_prediction_roi_subset
+unittest_prediction.nii.gz
 unittest_prediction_fast.nii.gz
+unittest_prediction_fast_body_seg.nii.gz
 
 *.tfstate
 *.tfstate.backup
diff --git a/bin/crop_to_body b/bin/crop_to_body
index a010cc42d..9bfe5ae49 100644
--- a/bin/crop_to_body
+++ b/bin/crop_to_body
@@ -34,6 +34,9 @@ def main():
 
     parser.add_argument("-ns", "--nr_thr_saving", type=int, help="Nr of threads for saving segmentations", 
                         default=6)
+    
+    parser.add_argument("-d", "--device", choices=["gpu", "cpu"], help="Device to run on (default: gpu).",
+                        default="gpu")
 
     parser.add_argument("-q", "--quiet", action="store_true", help="Print no intermediate outputs",
                         default=False)
@@ -45,23 +48,25 @@ def main():
 
     quiet, verbose = args.quiet, args.verbose
 
-    if not torch.cuda.is_available():
-        print("No GPU detected. Running on CPU. This can be very slow. The '--fast' option can help to some extend.")
+    device = "cuda" if args.device == "gpu" else "cpu"
+    if device == "cuda" and not torch.cuda.is_available():
+        print("No GPU detected. Running on CPU.")
+        device = "cpu"
 
     setup_nnunet()
 
     from totalsegmentator.nnunet import nnUNet_predict_image  # this has to be after setting new env vars
 
     crop_addon = [3, 3, 3]  # default value
-    download_pretrained_weights(269)
+    download_pretrained_weights(300)
 
     st = time.time()
     if not quiet: print("Generating rough body segmentation...")
-    body_seg = nnUNet_predict_image(args.input, None, 269, model="3d_fullres", folds=[0],
-                        trainer="nnUNetTrainerV2", tta=False, multilabel_image=True, resample=6.0,
+    body_seg = nnUNet_predict_image(args.input, None, 300, model="3d_fullres", folds=[0],
+                        trainer="nnUNetTrainer", tta=False, multilabel_image=True, resample=6.0,
                         crop=None, crop_path=None, task_name="body", nora_tag="None", preview=False, 
                         save_binary=False, nr_threads_resampling=args.nr_thr_resamp, nr_threads_saving=1, 
-                        crop_addon=crop_addon, quiet=quiet, verbose=verbose, test=0)
+                        crop_addon=crop_addon, quiet=quiet, verbose=verbose, test=0, device=device)
     if verbose: print(f"Rough body segmentation generated in {time.time()-st:.2f}s")
 
     body_seg_data = body_seg.get_fdata()
diff --git a/tests/unittest_prediction.nii.gz b/tests/unittest_prediction.nii.gz
deleted file mode 100644
index 1f518a4b5..000000000
Binary files a/tests/unittest_prediction.nii.gz and /dev/null differ
diff --git a/tests/unittest_prediction_fast_body_seg.nii.gz b/tests/unittest_prediction_fast_body_seg.nii.gz
deleted file mode 100644
index c8ae81e0c..000000000
Binary files a/tests/unittest_prediction_fast_body_seg.nii.gz and /dev/null differ
diff --git a/totalsegmentator/python_api.py b/totalsegmentator/python_api.py
index f7ffd2315..4884ea7fd 100644
--- a/totalsegmentator/python_api.py
+++ b/totalsegmentator/python_api.py
@@ -51,16 +51,16 @@ def totalsegmentator(input, output, ml=False, nr_thr_resamp=1, nr_thr_saving=6,
     input = Path(input)
     output = Path(output)
 
-    # available devices: gpu | cpu | mps
-    if device == "gpu": device = "cuda"
-
     nora_tag = "None" if nora_tag is None else nora_tag
 
     if not quiet: 
         print("\nIf you use this tool please cite: https://pubs.rsna.org/doi/10.1148/ryai.230024\n")
 
-    if not torch.cuda.is_available():
-        print("No GPU detected. Running on CPU. This can be very slow. The '--fast' option can help to some extend.")
+    # available devices: gpu | cpu | mps
+    if device == "gpu": device = "cuda"
+    if device == "cuda" and not torch.cuda.is_available():
+        print("No GPU detected. Running on CPU. This can be very slow. The '--fast' or the `--roi_subset` option can help to reduce runtime.")
+        device = "cpu"
 
     setup_nnunet()
     setup_totalseg()