This is from my series of questions "Can TeX do weird stuff?"

Today I want to know if TeX can simulate bad typography. I'm not talking about ugly templates, like wordlike or (god forgive me for this) abntex2, these are easy and not ugly enough.

I'm talking about nightmare-inducing poorly-kerned typography. I have seen lots of documents that the letters overlap each other and have irregular spacing, but could not find them for this question.

The closest thing I could find was (behold!) this:

enter image description here

and this:

enter image description here

I tried \letting and \deffing \kern and \glue to \relax and to a fixed amount, but none of these worked.

Is it possible to disable, or even better, completely messing up TeX's inter-letter spacing? Random inter-letter spacing would be perfect!

  • 20
    beauty is in the eye of the beholder... tex.stackexchange.com/questions/344214/… Feb 6, 2018 at 12:33
  • 16
    There are also the Cthulu worshippers
    – Skillmon
    Feb 6, 2018 at 12:49
  • 10
    This could be an interesting proposal for chickenize
    – egreg
    Feb 6, 2018 at 13:33
  • 3
    Most of the ugliness you illustrate in your examples is simply from font rasterization (i.e. it's simply not possible to do any better under the constraints of the number of pixels available). TeX will produce exactly such output, given low enough dpi (actually the output will not come from TeX itself but when the DVI file is processed, or when the PDF is rendered to screen) Feb 6, 2018 at 15:59
  • 2
    @egreg You can't fool me! Surely chickenize involves sacrificing live chickens to pagan deities by the light of the full moon. Otherwise, I cannot imagine how TeX could possibly work.
    – user139954
    Feb 6, 2018 at 19:07

2 Answers 2


An approach with XeTeX and \XeTeXinterchartoks (i.e. compile the below with xelatex):


  \kern \dimexpr(\randnum pt/10)\relax

\XeTeXinterchartokenstate = 1
\XeTeXinterchartoks 0 0 = {\randkern}



You can make the range smaller if you'd like less aggressive miskerning, e.g. changing {-30}{30} to {-10}{20} gives:

less miskerning

The idea is simply to insert a random kern between any two characters. For example, if we typed something like:

L\kern 0.4pt o\kern 1.3pt r\kern -1.0pt e\kern -0.4pt m

and so on, we'd have the result from the image above. The above code just does this, with two shortcuts:

  • Defines a \randkern, using package random to generate random numbers, and the \dimexpr primitive introduced in e-TeX).
  • Uses \XeTeXinterchartoks (documented in the XeTeX reference guide). Setting \XeTeXinterchartoks 0 0 to \randkern inserts the token \randkern between any two characters of class 0 (most characters are of class 0 by default). For example, when our text contains the character L followed by the character o, XeTeX treats it as if you had typed the token \randkern between them (like typing L\randkern o).
  • Beautiful! Can you please just explain briefly what does \XeTeXinterchartoks do? Feb 6, 2018 at 17:37
  • @PhelypeOleinik Yes sure, added to the answer with link to the documentation. Feb 6, 2018 at 18:02
  • 11
    The funny thing is I've got PDFs that end up like this because of bad font subsitutions (in a couple of cases only when printed). So you could have someone very confused
    – Chris H
    Feb 6, 2018 at 19:55

With LuaTeX, there’s a kerning callback provided, which we can use for exactly this purpose. There is no built-in equivalent of \XeTeXinterchartoks, but the appeal of LuaTeX is that a lot of such functionality we can implement ourselves, in Lua. We can get:


with (compile the below with lualatex)


where randomkern.lua is:

function rekern(head)
   local i = head
   while i~=nil do
      j = i.next
      -- Skip over discretionary (hyphen) nodes
      while j~=nil and node.type(j.id)=='disc' do
         j = j.next
      -- Insert a kern node between successive glyph nodes
      if node.type(i.id)=='glyph' and j~=nil and node.type(j.id)=='glyph' then
         k = node.new(node.id('kern'))
         head, i = node.insert_after(head, i, k)
      -- Tweak existing kerns (including ones we inserted) by a random amount
      if node.type(i.id)=='kern' then
         i.kern = i.kern + math.random(65536*-1, 65536*2)
      i = i.next

luatexbase.add_to_callback('kerning', rekern, 'Introduce random kern nodes')

The idea is that the kerning callback gets a list of nodes: for example,

  hlist  indent {}
  glyph  L
  glyph  o
  glyph  r
  glyph  e
  glyph  m
  glue  <spaceskip: 218235 plus 109117^0 minus 72745^0>
  glyph  i
  glyph  p
  glyph  s
  glyph  u
  glyph  m
  glue  <spaceskip: 218235 plus 109117^0 minus 72745^0>
  glyph  d
  glyph  o
  glyph  l
  glyph  o
  glyph  r
  glue  <spaceskip: 218235 plus 109117^0 minus 72745^0>
  glyph  s
  glyph  i
  glyph  t

and so on. We simply traverse this list, and between every two consecutive glyph nodes (possibly with a disc node between them), we insert a kern node, and give it a random value. (TeX stores all dimensions (lengths, etc.) internally in scaled points, where 65536 sp = 1 pt.)

Note that this version handles hyphenation automatically: only the kerning changes; the set of valid hyphenation or line-break points remains the same.

  • 1
    (Posted as separate answer because it's for a different engine and everything is different…) Feb 7, 2018 at 0:01
  • The kerning callback is supposed to apply kerning to the list, so the identity would be function(head) node.kerning(head) return head end. This also means if you just do function(head) return head end, TeX will apply no kerning at all, which should give similarly poor results. Aug 28, 2019 at 7:40
  • @HenriMenke Ah I see, I think I may not have been aware of node.kerning; thanks for the info. In LuaTeX it's currently often hard to know what the “identity” callback would be, so that's definitely useful to know. (BTW did you try your suggested no-kerning approach? The results don't look poor, i.e. they look similar to the default, at least to my eyes...) Aug 28, 2019 at 9:50

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .