diff --git a/README.md b/README.md
index 1b508bb..a2ac5b9 100644
--- a/README.md
+++ b/README.md
@@ -3,8 +3,6 @@
 "Yinyang" K-means and K-nn using NVIDIA CUDA
 ============================================
 
-![source{d}](img/sourced.png)
-
 K-means implementation is based on ["Yinyang K-Means: A Drop-In Replacement
 of the Classic K-Means with Consistent Speedup"](https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ding15.pdf)
 article. While it introduces some overhead and many conditional clauses
@@ -23,6 +21,9 @@ defined in `kmcuda.h`: `kmeans_cuda` and `knn_cuda`.
 It has the built-in Python3 native extension support, so you can
 `from libKMCUDA import kmeans_cuda`.
 
+[![source{d}](img/sourced.png)](http://sourced.tech)
+<p align="right"><a href="img/kmeans_image.ipynb">How this was created?</a></p>
+
 Table of contents
 -----------------
 * [K-means](#k-means)
@@ -44,7 +45,6 @@ Table of contents
 * [Python API](#python-api)
 * [C examples](#c-examples)
 * [C API](#c-api)
-* [How "source{d}" image was created?](#how-sourced-image-was-created)
 * [License](#license)
 
 K-means
@@ -533,10 +533,6 @@ KMCUDAResult knn_cuda(
 
 Returns KMCUDAResult (see `kmcuda.h`);
 
-How "source{d}" image was created?
-----------------------------------
-Check out this [notebook](img/kmeans_image.ipynb).
-
 License
 -------
 MIT license.
diff --git a/img/sourced.png b/img/sourced.png
index 94ba26a..56800c8 100644
Binary files a/img/sourced.png and b/img/sourced.png differ