diff --git a/supervised-airfoils.ipynb b/supervised-airfoils.ipynb
index f817df9..1cf7c78 100644
--- a/supervised-airfoils.ipynb
+++ b/supervised-airfoils.ipynb
@@ -275,7 +275,7 @@
         "\n",
         "Here's a overview of the architecure:\n",
         "\n",
-        "![An overview of the U-net we're using for this learning task](https://github.com/tum-pbs/pbdl-book/blob/master/resources/supervised-airfoils-unet.jpg?raw=1)\n",
+        "![An overview of the U-net we're using for this learning task](resources/supervised-airfoils-unet.jpg)\n",
         "\n",
         "First, we'll define a helper to set up a convolutional block in the network, `blockUNet`. Note, we don't use any pooling! Instead we use strides and transpose convolutions (these need to be symmetric for the decoder part, i.e. have an uneven kernel size), following [best practices](https://distill.pub/2016/deconv-checkerboard/). The full pytroch neural network is managed via the `DfpNet` class."
       ]
@@ -808,4 +808,4 @@
   },
   "nbformat": 4,
   "nbformat_minor": 0
-}
\ No newline at end of file
+}