diff --git a/README.md b/README.md
index 11354cc..87c14da 100644
--- a/README.md
+++ b/README.md
@@ -22,7 +22,7 @@ python3 main.py --device=cuda --port=8080 --model=lama
 
 - `--crop-trigger-size`: If image size large then crop-trigger-size, crop each area from original image to do inference.
   Mainly for performance and memory reasons on **very** large image.Default is 2042,2042
-- `--crop-size`: Crop size for `--crop-trigger-size`. Default is 512,512.
+- `--crop-margin`: Margin around bounding box of painted stroke when crop mode triggered. Default is 256.
 
 ### Start server with LDM model
 
diff --git a/lama_cleaner/lama/__init__.py b/lama_cleaner/lama/__init__.py
index 49f2596..9efc6c2 100644
--- a/lama_cleaner/lama/__init__.py
+++ b/lama_cleaner/lama/__init__.py
@@ -14,16 +14,16 @@ LAMA_MODEL_URL = os.environ.get(
 
 
 class LaMa:
-    def __init__(self, crop_trigger_size: List[int], crop_size: List[int], device):
+    def __init__(self, crop_trigger_size: List[int], crop_margin: int, device):
         """
 
         Args:
             crop_trigger_size: h, w
-            crop_size: h, w
+            crop_margin:
             device:
         """
         self.crop_trigger_size = crop_trigger_size
-        self.crop_size = crop_size
+        self.crop_margin = crop_margin
         self.device = device
 
         if os.environ.get("LAMA_MODEL"):
@@ -79,12 +79,10 @@ class LaMa:
         box_w = box[2] - box[0]
         cx = (box[0] + box[2]) // 2
         cy = (box[1] + box[3]) // 2
-        crop_h, crop_w = self.crop_size
         img_h, img_w = image.shape[1:]
 
-        # TODO: when box_w > crop_w, add some margin around?
-        w = max(crop_w, box_w)
-        h = max(crop_h, box_h)
+        w = box_w + self.crop_margin * 2
+        h = box_h + self.crop_margin * 2
 
         l = max(cx - w // 2, 0)
         t = max(cy - h // 2, 0)
@@ -94,7 +92,7 @@ class LaMa:
         crop_img = image[:, t:b, l:r]
         crop_mask = mask[:, t:b, l:r]
 
-        print(f"Apply zoom in size width x height: {crop_img.shape}")
+        print(f"box size: ({box_h},{box_w}) crop size: {crop_img.shape}")
 
         return self._run(crop_img, crop_mask), [l, t, r, b]
 
diff --git a/lama_cleaner/ldm/__init__.py b/lama_cleaner/ldm/__init__.py
index 092cab9..fa077ec 100644
--- a/lama_cleaner/ldm/__init__.py
+++ b/lama_cleaner/ldm/__init__.py
@@ -318,8 +318,10 @@ class LDM:
                 cx = box_x + box_w // 2
                 cy = box_y + box_h // 2
 
-                w = max(512, box_w)
-                h = max(512, box_h)
+                # w = max(512, box_w)
+                # h = max(512, box_h)
+                w = box_w + 512
+                h = box_h + 512
 
                 left = max(cx - w // 2, 0)
                 top = max(cy - h // 2, 0)
diff --git a/main.py b/main.py
index 8ca074c..25f4836 100644
--- a/main.py
+++ b/main.py
@@ -107,7 +107,8 @@ def get_args_parser():
                              "crop each area from original image to do inference."
                              "Mainly for performance and memory reasons"
                              "Only for lama")
-    parser.add_argument("--crop-size", default="512,512")
+    parser.add_argument("--crop-margin", type=int, default=256,
+                        help="Margin around bounding box of painted stroke when crop mode triggered")
     parser.add_argument(
         "--ldm-steps",
         default=50,
@@ -131,11 +132,9 @@ def main():
     device = torch.device(args.device)
 
     crop_trigger_size = [int(it) for it in args.crop_trigger_size.split(",")]
-    crop_size = [int(it) for it in args.crop_size.split(",")]
 
     if args.model == "lama":
-        model = LaMa(crop_trigger_size=crop_trigger_size,
-                     crop_size=crop_size, device=device)
+        model = LaMa(crop_trigger_size=crop_trigger_size, crop_margin=args.crop_margin, device=device)
     elif args.model == "ldm":
         model = LDM(device, steps=args.ldm_steps)
     else: