diff --git a/clip.gif b/clip.gif
new file mode 100644
index 0000000..e43c795
Binary files /dev/null and b/clip.gif differ
diff --git a/rdf.gif b/rdf.gif
deleted file mode 100644
index 6cb4ea6..0000000
Binary files a/rdf.gif and /dev/null differ
diff --git a/readme.md b/readme.md
index 09e0a7d..48b53dc 100644
--- a/readme.md
+++ b/readme.md
@@ -1,6 +1,6 @@
 # 3d-beats
 
-![Hand Classifier](rdf.gif)
+![3d-beats](clip.gif)
 
 At the core, this is an implementation of depth image pixel classification via randomized decision forests, as outlined in *Real-Time Human Pose Recognition in Parts from Single Depth Images* (look in docs). This is the technique used in the original microsoft kinect for realtime human pose estimation.