diff --git a/README.md b/README.md
index 6772d6b..034bc0a 100644
--- a/README.md
+++ b/README.md
@@ -16,9 +16,13 @@ invoices as well as both [SQuAD2.0](https://huggingface.co/datasets/squad_v2) an
 ## Non-consecutive tokens
 
 Unlike other QA models, which can only extract consecutive tokens (because they predict the start and end of a sequence), this model can predict longer-range, non-consecutive sequences with an additional
-classifier head. For example, it can extract the two-line address as below:
+classifier head. For example, QA models often encounter this failure mode:
 
-![Two-line Address](./demo.png)
+![Broken Address](./before.png)
+
+However this model is able to predict non-consecutive tokens and therefore the address correctly:
+
+![Two-line Address](./after.png)
 
 ## Getting started with the model
 
diff --git a/after.png b/after.png
new file mode 100644
index 0000000..fa222c1
Binary files /dev/null and b/after.png differ
diff --git a/before.png b/before.png
new file mode 100644
index 0000000..f28f8b4
Binary files /dev/null and b/before.png differ
diff --git a/demo.png b/demo.png
deleted file mode 100644
index b2cd5d9..0000000
Binary files a/demo.png and /dev/null differ