diff --git a/Images/3DDFA_V2.jpg b/Images/3DDFA_V2.jpg
new file mode 100644
index 0000000..4a2f5fe
Binary files /dev/null and b/Images/3DDFA_V2.jpg differ
diff --git a/Images/3DDFA_V2_800.jpg b/Images/3DDFA_V2_800.jpg
new file mode 100644
index 0000000..14b0816
Binary files /dev/null and b/Images/3DDFA_V2_800.jpg differ
diff --git a/README.md b/README.md
index 726f832..4b10112 100644
--- a/README.md
+++ b/README.md
@@ -103,6 +103,10 @@ Pytorch weights for use with `model.py` can be found [here](https://mega.nz/file
 
 More samples: [Results3.png](https://raw.githubusercontent.com/emilianavt/OpenSeeFace/master/Images/Results3.png), [Results4.png](https://raw.githubusercontent.com/emilianavt/OpenSeeFace/master/Images/Results4.png)
 
+Here's an informal [comparison](https://raw.githubusercontent.com/emilianavt/OpenSeeFace/master/Images/3DDFA_V2.png) with [3DDFA_V2](https://github.com/cleardusk/3DDFA_V2) on five arbitrary, non-cherry picked examples from my test data:
+
+![3DDFA_V2_800.png](https://raw.githubusercontent.com/emilianavt/OpenSeeFace/master/Images/3DDFA_V2_800.png)
+
 ## Face detection
 
 The landmark model is quite robust with respect to the size and orientation of the faces, so the custom face detection model gets away with rougher bounding boxes than other approaches. It has a favorable speed to accuracy ratio for the purposes of this project.