diff --git a/test/shaping/data/text-rendering-tests/extract-tests.py b/test/shaping/data/text-rendering-tests/extract-tests.py
new file mode 100755
index 000000000..cb003db9c
--- /dev/null
+++ b/test/shaping/data/text-rendering-tests/extract-tests.py
@@ -0,0 +1,45 @@
+#!/usr/bin/env python
+
+from __future__ import print_function
+import sys
+import xml.etree.ElementTree as ET
+
+# Can we extract this from HTML element itself? I couldn't.
+namespaces = {
+	'ft': 'https://github.com/OpenType/fonttest',
+	'xlink': 'http://www.w3.org/1999/xlink',
+}
+def ns(s):
+	ns,s = s.split(':')
+	return '{%s}%s' % (namespaces[ns], s)
+
+def unistr(s):
+	return ','.join('U+%04X' % ord(c) for c in s)
+
+def glyphstr(glyphs):
+	out = []
+	for glyphname,x,y in glyphs:
+		if x or y:
+			out.append('%s@%d,%d' % (glyphname, x, y))
+		else:
+			out.append(glyphname)
+	return '['+'|'.join(out)+']'
+
+html = ET.fromstring(sys.stdin.read())
+found = False
+for elt in html.findall(".//*[@class='expected'][@ft:id]", namespaces):
+	found = True
+	name = elt.get(ns('ft:id'))
+	text = elt.get(ns('ft:render'))
+	font = elt.get(ns('ft:font'))
+	glyphs = []
+	for use in elt.findall(".//use"):
+		x = int(use.get('x'))
+		y = int(use.get('y'))
+		href = use.get(ns('xlink:href'))
+		assert href[0] == '#'
+		glyphname = '.'.join(href[1:].split('/')[1].split('.')[1:])
+		glyphs.append((glyphname, x, y))
+	print("../fonts/%s:--font-size=1000 --no-clusters --no-advances:%s:%s" % (font, unistr(text), glyphstr(glyphs)))
+
+sys.exit(0 if found else 1)