diff --git a/btrfs-send-receive.sh b/btrfs-send-receive.sh
index b93e47d..2e21803 100755
--- a/btrfs-send-receive.sh
+++ b/btrfs-send-receive.sh
@@ -57,6 +57,9 @@ btrfs filesystem usage .
 duperemove -dr . > /dev/null
 btrfs subvolume sync .
 btrfs filesystem sync .
+btrfs balance start --full-balance --enqueue .
+btrfs subvolume sync .
+btrfs filesystem sync .
 ## And to finish things off we shrink the filesystem to the minimum size.
 $OUTPUT_DIR/btrfs-shrink.py
 ## Sync changes to disk.
diff --git a/btrfs-shrink.py b/btrfs-shrink.py
index 6b4f32e..b8b8085 100755
--- a/btrfs-shrink.py
+++ b/btrfs-shrink.py
@@ -8,6 +8,11 @@
 # portion but leave a bit of a buffer behind. Then we keep resizing until the
 # resize starts failing.
 
+# TODO we can produce an even more squeezed image by deriving a complete partition
+# from the initial partition. https://btrfs.readthedocs.io/en/latest/Seeding-device.html
+# It essentially transfers the data from the seed device with nary a fragmentation.
+# Requires juggling loop devices though, and the gains are in the sub 500MiB range.
+
 import json
 import os
 import math