diff --git a/README.md b/README.md
index 16cfc00..199a53b 100644
--- a/README.md
+++ b/README.md
@@ -131,6 +131,10 @@ and end with an image number starting from 0, such as `random_image_0.jpg`,
 `--size` flag displays information on the images, such as the size of the first
 image and the size of the overall directory.
 
+Note that for creating a very large number of images, systems can easily run out
+of memory. In this case, increase the `--chunksize` to reduce the amount of
+memory allocated by each multiprocessing pool.
+
 ### TFRecords
 TFRecords can also be easily generated using the application. This command
 expects images to be pre-loaded to be used as the basis for the TFRecord files.
diff --git a/imagine/imagine.py b/imagine/imagine.py
index 2543239..e226b2c 100755
--- a/imagine/imagine.py
+++ b/imagine/imagine.py
@@ -72,7 +72,9 @@ def check_directory_exists(directory):
 @click.option('--image_format', default='png', required=True)
 @click.option('--seed', default=0)
 @click.option('--size', is_flag=True, default=False)
-def create_images(path, name, width, height, count, image_format, seed, size):
+@click.option('--chunksize', default=64)
+def create_images(path, name, width, height, count, image_format, seed, size,
+                  chunksize):
     click.echo("Creating {} {} files located at {} of {}x{} resolution with a "
                "base filename of {}".format(count, image_format, path, width,
                                             height, name))
@@ -84,9 +86,13 @@ def create_images(path, name, width, height, count, image_format, seed, size):
     pool = Pool()
     try:
         start_time = perf_counter()
+        # NOTE: For very large image counts on memory-constrained systems, this
+        # can stall-out. Either reduce the image count request, or increase the
+        # chunk size.
         pool.starmap(image_creation,
                      ((combined_path, width, height, seed, image_format, n)
-                      for n in range(count)))
+                      for n in range(count)),
+                     chunksize=chunksize)
     finally:
         pool.close()
         pool.join()