diff --git a/README.md b/README.md
index 513dde250847f1c0b2010af6633ae9c0b4a0334b..fc10ac73f3ba92c04d1554fe2275ec9d22c9cfb2 100644
--- a/README.md
+++ b/README.md
@@ -1,10 +1,10 @@
 # open-frame-scan
 
-<img src="img/frame.jpg" width=90%></img>
+<img src="img/frame.jpg" width=70%></img>
 
 This device is an open-source approach to scanning flat surfaces. The frame is simply placed on top of the surface to scan, and a computer vision software analyzes the picture to extract a rectified picture with real dimensions (pixels/unit).
 
-<img src="img/comparison.jpg" width=90%></img>
+<img src="img/comparison.jpg" width=70%></img>
 
 This project is inspired by the [Shaper Trace](https://www.shapertools.com/en-us/trace), but offers an open source interface and lets users scan color images, not just vectors.
 
@@ -30,21 +30,21 @@ If you have access to digital fabrication tools, the most straightforward way to
 
 The first step is to weed the vinyl in place:
 
-![](img/vinyl1.jpg)
+<img src="img/vinyl1.jpg" width=50%></img>
 
 The result is then stuck onto transfer tape and laid flat:
 
-![](img/vinyl2.jpg)
+<img src="img/vinyl2.jpg" width=50%></img>
 
 After joining to the frame, the transfer is peeled off, revealing the finished frame:
 
-![](img/vinyl3.jpg)
+<img src="img/vinyl3.jpg" width=50%></img>
 
 ### Inkjet printing
 
 Alternatively, you can print the design of the frame using an inkjet printer, then glue it to a thin piece of cardboard:
 
-![](img/cardboard.jpg)
+<img src="img/cardboard.jpg" width=50%></img>
 
 The small frame is designed to be printable on an A4 page (or letter), and the medium frame fits on an A3 page.
 
@@ -52,20 +52,20 @@ The small frame is designed to be printable on an A4 page (or letter), and the m
 
 To quantify the accuracy of the scanning, a 25mm calibration grid was built:
 
-![](img/grid.jpg)
+<img src="img/grid.jpg" width=70%></img>
 
 Here is the image used for analysis, showing the detected features (aruco markers in blue, corners in red):
 
-![](img/features.jpg)
+<img src="img/features.jpg" width=70%></img>
 
 Here is the extracted image:
 
-![](img/result.jpg)
+<img src="img/result.jpg" width=70%></img>
 
 The grid feature points are automatically extracted:
 
-![](img/grid_features.jpg)
+<img src="img/grid_features.jpg" width=70%></img>
 
 Finally, the distance between neighboring grid points is measured, revealing the accuracy of the scanning process. Here is a comparison of all frames, including the Shaper Trace, showing improved accuracy and less variance:
 
-![](img/plot.png)
+<img src="img/plot.png" width=70%></img>