Can you tell the difference between AI-generated vs human-made fonts?

ericlu
ericlu Posts: 16
in Font Technology
Hi Typedrawers, 

In the replies to some of my previous posts about Mixfont, there were folks that were dismissive of the quality of AI-generated fonts. A lot of people dismissed the quality of the one example font I posted (Gothic Gumdrop) and I've also seen some articles shared discussing how high quality AI-generated fonts are still 3-4 years away. 

I wanted to challenge these claims as I personally believe that the raster->vector font generation pipeline that's being used in Mixfont today is already good enough for the vast majority of use cases for decorative, display, stylized, etc fonts. While I think that creating a legendary variable font is not quite possible, I do feel that for most projects using AI-generated fonts is already high quality enough for many designers.

To prove out this hypothesis I've put together an AI Font Quiz, where some of these fonts are licensed from real font foundries, while others in this quiz are 100% generated with AI using the Mixfont model. You can try typing and testing out all of the fonts, and when you're ready you can select whether you think the font is AI-generated or human-made. 





Obviously if you download the TTF and inspect it you'll be able to easily tell the source of the font, but the point of the quiz is to see if you can visually decide whether something is AI-generated or human-made. 

I would love for you all to try out taking the quiz and let me know your scores. My thinking is that if this is hard even for professional type designers than it will be a challenge for everyday folks who don't think as much about type. I want to challenge the assumption that font generation is years away because I really think it is here right now and can be used and applied in many new ways. 
 
Respectfully looking forward to your feedback,
Eric 



Tagged:

Comments

  • James Puckett
    James Puckett Posts: 2,051
    I got most of them wrong. This reinforces my plan to go back to school for a double major in economics and Asian studies. Which will probably also get wiped out, career-wise, by AI but at least it will be a fun way to kill time until I graduate in my mid fifties and find out that all of those have also gone to AI.
  • Nick Shinn
    Nick Shinn Posts: 2,360
    edited June 30
    The world is burning, and the broligarchy is hyperscaling an invasive technology that consumes prodigious amounts of water and energy.  
    Typeface “quality” is not the issue.
  • Ray Larabie
    Ray Larabie Posts: 1,492
    edited 1:15AM
    I got one wrong. If you haven't taken the test yet, use the size tool and change sentences. If you don't rush through, it's really obvious (except for one). Also, you can type your own text in there...the numerals...yikes.
  • Qwerasd
    Qwerasd Posts: 7
    By just selecting "AI" for the fonts with obvious issues with the spacing or glaring curve quality problems, I got 11/12 correct. I misidentified one of the non-AI font as AI because it has a ton of weirdnesses that don't feel intentional to me- strange baseline mis-alignments, lumps and divots in curves that feel arbitrarily placed, and inconsistent logic for the forms.

    I think it's also worth noting that that false positive is from a designer who only started putting out fonts in 2021 and whose website shows many other fonts with similar issues. I'm not saying they're using AI image generators to generate letter forms then auto tracing them in to fonts to sell, but I'm not not saying that. Even if it's all by hand, they're clearly all very quickly dashed out, as evidenced by their enormous catalogue of work in just 5 years- and the fact that they don't have much regard for quality is evident by the "blog" section of their website which has 3 posts, from January, February, and October of 2025, all of which are a single paragraph of lorem ipsum. Not to mention that the search bar on the blog page has the same "Search font by name.." placeholder text as the font list section of the website. Suffice it to say, the designer of that particular font is at the very least not someone with a keen eye for detail or quality.

    I'll say, as well, that I'm relatively new to - and not exceptionally experienced with - type design, and it didn't take me some kind of special knowledge or especially close examination of the fonts in the quiz to recognize the gap in quality between the AI fonts and the human fonts. Like I said, I didn't select based on what I thought "looked AI" but rather simply which of them had glaring flaws that no self-respecting human designer would consider acceptable for an end product.
  • John Savard
    John Savard Posts: 1,234
    I did very poorly on the test. But that only seems to prove that some humans have designed typefaces that are as ugly as AI-generated ones, not that AI is capable of creating good typefaces. Ah, I see the post before me said the same thing.
  • jeremy tribby
    jeremy tribby Posts: 292
    ericlu said:
    In the replies to some of my previous posts about Mixfont, there were folks that were dismissive of the quality of AI-generated fonts.

    I wanted to challenge these claims

  • ericlu
    ericlu Posts: 16
    Mark Simonson said:
    None of the fonts in the quiz are what I would call good quality fonts.

    I am curious if it's possible to get even more specific and define what makes a "good quality font". I do agree that some of the selected human-made fonts may not in the echelon of the Helveticas, Inters, Proxima Novas of the world but they are still sold by professional foundries and/or added to selective collections (like Google Fonts).

    Is the general consensus here that Google Fonts are not high quality fonts? Or is font quality simply a "I know it when I see it" type of assessment? Speaking as an outsider, I have to admit it feels a bit elitist to opine certain fonts as not high quality, especially when it feels like many font designs are situational and a matter of personal preference. 
  • Dave Crossland
    Dave Crossland Posts: 1,554
    As the chief curator of the Google Fonts type library, I would offer that the general consensus (and my own view) is that Google Fonts is a very very mixed bag - all blanket statements about the quality of the library as a single unit are flawed, because there are both super high quality fonts and super low quality fonts in there – there's stuff from the absolute top-tier most-elite foundries, and there's stuff from undergraduate graphic design students who made their first font and shared it with the world under the Open Font License. (I should also note that the GF type library includes fonts by Jeremy, John, Mark, Nick, and James.)

    The essential dilemma I see with the kind of critique that Jeremy offer 2 replies up, which was also a dilemma with the 2011-2012 era fonts published on Google Fonts, particularly those by Vernon Adams R.I.P., is that while type designers can see the differences these few-UPM-unit differences make, I'm not sure that the wider reading public can see, nor the subset of people who are paid to produce documents. Here I'm attempting a sort of descriptivist definition of 'professional designer'; typically non-type-designer visual designers, who went to the UK University of Reading or Yale School of Art or whatnot, can see craft quality in font development. But it seems to me that this is a secondary factor – what people, even people who care for font making craft, primarily value when choosing type is the "atmosphere value," the emotional/cultural associations it bring to their project. 

    Back in the day, my idea for folks like Vernon was to focus on that as they developed new type for the web (and mostly for US English), because the nature of the GF CSS API is that when a font family is updated in the API, all web pages linking to the API will render with the API's current/latest version only. So we could launch a font family and update it over time, redrawing the design and improving the craft and care in it, add kerning, add more language support, add more family styles... And by putting out a couple dozen type ideas of roughly equal craft and scope, Vernon could then look at the API usage stats and prioritize what to invest expansion effort into. I had some great debates with some of the titans (I think Bruno Maag was probably the most fun for me to argue with ;) who doubted the wisdom of this, since going back to the floppy disk days, once a half-cooked font gets out into the wild hands of the public, the file continues to float around a long time, and they tried to postpone font distribution as long as possible so that any file circulating will be the least-bad version possible. 

    Anyway, my point is that if when picking type people see these 2 A.I. generated type samples from Eric's quiz,



    and their "System 1" mind says "I like it [for my project]", well, maybe their System 2 mind will realize it lacks the glyph repertoire they require or the numerals are too low quality or whatever, but maybe not. If they vibe with it strongly enough, they may be willing to overlook some issues.

    Eric, I'm curious if you are logging the test results, as I will be interested to see what the aggregate % result trends towards. 
  • C.Fransen
    C.Fransen Posts: 16
    Why would I help train an AI model that will eventually make this beautiful profession obsolete?
  • jeremy tribby
    jeremy tribby Posts: 292
    Dave Crossland said:
    The essential dilemma I see with the kind of critique that Jeremy offer 2 replies up, which was also a dilemma with the 2011-2012 era fonts published on Google Fonts, particularly those by Vernon Adams R.I.P., is that while type designers can see the differences these few-UPM-unit differences make, I'm not sure that the wider reading public can see, nor the subset of people who are paid to produce documents. 
    I would expand on this, though: for the time being at least, fonts are software that have to interoperate with other software. fonts have many audiences with varying levels of sensitivity to type, but I think an important category of font users not represented here are the machines - the font renderers, printers, design tooling, etc. even if that /T had all the extra points in places that didnt matter visually, such that, ideally, they would be invisible to the eye, that's still no guarantee that the font would work well for a given use case / environment. well-behaved outlines seem like table stakes, far below the threshold of the kinds of fonts that are implicitly being discussed in this thread (for example). I find this new breed of autotracers really interesting but the output so far seems inadequate
  • Christian Thalmann
    Christian Thalmann Posts: 2,068
    ericlu said:
    Speaking as an outsider, ...
    Have you considered at some point listening to what the insiders have to say? Every time they do, you start a new thread and expect a different answer.
  • Dave Crossland
    Dave Crossland Posts: 1,554
    jeremy tribby said:

    I would expand on this, though: for the time being at least, fonts are software that have to interoperate with other software. fonts have many audiences with varying levels of sensitivity to type, but I think an important category of font users not represented here are the machines - the font renderers, printers, design tooling, etc. 
    Sure, it's on that basis that Eric's first Google Fonts submission was not immediately accepted. 

    https://github.com/google/fonts/issues/10635

    For now the "human in the loop" remains a requirement.
  • Jesse R Ewing
    Jesse R Ewing Posts: 15
    I got 10 out of 12. My main takeaway is that AI-generated fonts can be impossible to distinguish from poorly-drawn human made ones.
  • Mark Simonson
    Mark Simonson Posts: 1,798
    I am curious if it's possible to get even more specific and define what makes a "good quality font".
    A good quality font (in my opinion) is one that is not only technically well-made but also one that I can imagine using as an experienced graphic designer in terms of its visual appeal and utility—something I’d be willing to pay for. To equate “experienced professional” with “elitist” just indicates inexperience to me.
  • John Hudson
    John Hudson Posts: 3,702
    A font is a technical implementation of a typeface, which is a coordinated design expressing an idea about letterforms and their systematic arrangement in the making of text. A good quality font must meet a base standard of technical quality, partly defined by the format and partly by the requirements of the individual typeface and the system that typeface requires to be useful in making text. There are more or less well-established canons of what makes a good typeface design for various kinds of use, which lean on craft and scientific understanding readability. Some aspects of these canons are more relaxed for fancy display types than for those intended for running text, but notions around relative proportions of letterforms and their spacing are common to most uses. To develop as a type designer is to gain a feeling for these notions and an eye for well-proportioned and well-spaced type, which may ultimately get you to a point where you can do unusual things and go against convention while understanding what you are doing and why.

    A font is an implementation of an idea, and better ideas make better fonts. But the idea isn’t reducible to an AI prompt, because the idea is something that is explored in the process of making, in the many thousands of small decisions made in the crafting of shapes and their arrangement within the system of the typeface.

Categories