diff --git a/collisions/examples/collisions1.mp4 b/collisions/examples/collisions1.mp4
new file mode 100644
index 0000000..38e0473
Binary files /dev/null and b/collisions/examples/collisions1.mp4 differ
diff --git a/collisions/examples/collisions2.mp4 b/collisions/examples/collisions2.mp4
new file mode 100644
index 0000000..09b28f7
Binary files /dev/null and b/collisions/examples/collisions2.mp4 differ
diff --git a/collisions/scripts/mp4s.py b/collisions/scripts/mp4s.py
new file mode 100644
index 0000000..7ec5371
--- /dev/null
+++ b/collisions/scripts/mp4s.py
@@ -0,0 +1,109 @@
+#!/usr/bin/env python
+
+# script to collide "MP4"-based files with specific `ftyp`
+# Ange Albertini, December 2019
+
+import struct
+import sys
+import hashlib
+
+
+def dprint(s):
+  DEBUG = True
+  DEBUG = False
+  if DEBUG:
+    print("D " + s)
+
+
+def relocate(d, delta):
+  # finds and relocates all Sample Tables Chunk Offset tables
+  # TODO: support 64 bits `co64` tables ()
+  offset = 0
+  tablecount = d.count("stco")
+  dprint("stco found: %i" % tablecount)
+  for i in range(tablecount):
+    offset = d.find("stco", offset)
+    dprint("current offset: %0X" % offset)
+
+    length   = struct.unpack(">I", d[offset-4:offset])[0]
+    verflag  = struct.unpack(">I", d[offset+4:offset+8])[0]
+    offcount = struct.unpack(">I", d[offset+8:offset+12])[0]
+
+    if verflag != 0:
+      dprint(" version/flag not 0 (found %X) at offset: %0X" % (verflag, offset+4))
+      continue
+
+    # length, type, verflag, count - all 32b
+    if (offcount + 4) * 4 != length:
+      dprint(" Atom length (%X) and offset count (%X) don't match" % (length, offcount))
+      continue
+
+    dprint(" offset count: %i" % offcount)
+    offset += 4 * 3
+    offsets = struct.unpack(">%iI" % offcount, d[offset:offset + offcount * 4])
+    dprint(" offsets (old): %s" % `list(offsets)`) 
+    offsets = [i + delta for i in offsets]
+    dprint(" (new) offsets: %s" % `offsets`)
+
+    d = d[:offset] + struct.pack(">%iI" % offcount, *offsets) + d[offset+offcount*4:]
+
+    offset += 4 * offcount
+
+  dprint("")
+  return d
+
+
+def freeAtom(l):
+  return struct.pack(">I", l) + "free" + "\0" * (l - 8)
+
+
+def isValid(d):
+  # fragile check of validity
+  return d.startswith("\0\0\0") and d[:32].count("ftyp") > 0
+
+
+
+fn1, fn2 = sys.argv[1:3]
+
+with open(fn1, "rb") as f:
+  d1 = f.read()
+
+with open(fn2, "rb") as f:
+  d2 = f.read()
+
+assert isValid(d1)
+assert isValid(d2)
+
+d1 = d1[0x20:]
+d2 = d2[0x20:]
+l1 = len(d1)
+l2 = len(d2)
+
+suffix = "".join([
+  struct.pack(">I", 0x100 + 8), "free",
+  "\0" * (0x100 - 8),
+  struct.pack(">I", 8 + l1), "free",
+  relocate(d1, 0x1C0 + 8 - 0x20),
+  relocate(d2, 0x1C0 + 8 - 0x20 + l1),
+  ])
+
+# 32b length prefix
+
+with open("mp4s1.bin", "rb") as f:
+  prefix1 = f.read()
+with open("mp4s2.bin", "rb") as f:
+  prefix2 = f.read()
+
+col1 = prefix1 + suffix
+col2 = prefix2 + suffix
+
+md5 = hashlib.md5(col1).hexdigest()
+
+if md5 == hashlib.md5(col2).hexdigest():
+  print "common md5: %s" % md5
+
+  with open("collisions1.mp4", "wb") as f:
+    f.write(col1)
+
+  with open("collisions2.mp4", "wb") as f:
+    f.write(col2)
diff --git a/collisions/scripts/mp4s1.bin b/collisions/scripts/mp4s1.bin
new file mode 100644
index 0000000..d3369db
Binary files /dev/null and b/collisions/scripts/mp4s1.bin differ
diff --git a/collisions/scripts/mp4s2.bin b/collisions/scripts/mp4s2.bin
new file mode 100644
index 0000000..226c96f
Binary files /dev/null and b/collisions/scripts/mp4s2.bin differ