Mixfont Update - seeking a designer to help complete an AI-generated font

ericlu · September 14

Hi Typedrawers,

It's been a minute since I've posted. Since I last shared about Mixfont, the site has grown to over 25,000 registered users from all over the world who have created tens of thousands of unique fonts. Mixfont is now the #1 result for AI font generator on major search engines. I've received countless messages of support from users who have been able to create their own imagined font designs from scratch, and the API has been integrated in several popular note-taking apps, enabling people to start typing in their own handwriting from just a sample image. Some of my work has also been cited in the Smithsonian Magazine, Dezeen, and Fast Company.

I'm super grateful for the support and feedback that I've gotten from the Typedrawers community. Thanks to the feedback I've gotten here I've been able to continually improve the service and make it more helpful for thousands of designers from around the world.

In the vein of this work, I'm currently looking to commission a type designer to help me take an AI-generated font over the finish line so that I can submit it to Google Fonts. I tried this initially with a font called Gothic Gumdrop and the font was tentatively accepted, but still contained certain spacing and outline issues. I'm looking to work with a typographer or font designer to help take a font like this over the finish line. The work would involve polishing the details of an AI generated font, fixing spacing and outline issues, and manually adjusting or redesigning any glyphs that need to be improved. It can be Gothic Gumdrop, or use a completely different AI-generated font as a starting point. My goal is to prep a submission that meet the quality standards of the Google Fonts team so that the "world's first AI-generated font" can be submitted and accepted to the library.

I realize that this is an unusual request, but I thought I would start in this forum to see if there would be anyone here who would be interested in partnering on this project. I have the budget to make this happen - please let me know if you or someone you might know might be interested.

Thank you and look forward to working together.
Eric

Thomas Phinney · September 15

Why not put that money/effort into improving Mixfont in the same areas?

There already exist auto-spacing, auto-kerning and especially outline-optimization systems that do better.

SCarewe · September 15

Even Florence Foster-Jenkins was less tone-deaf.

Simon Cozens · September 15

I tend to use Yofont for all my AI font generation needs these days. It does everything Mixfont does, and it also generates Hebrew, Arabic, Devanagari and CJK as well.

Kasper Pyndt · September 15

You do realise that if a designer manually adjusts and optimises the font, then it’s no longer “the world’s first AI generated font” - right?

yanone · September 15

I do admire the framing. “Take an AI-generated font over the finish line” — as if the machine had done the heavy lifting and now just needs an incarnation to receive the medal. But the work on offer — fixing the spacing, correcting the outlines, redesigning the glyphs that don’t work — isn’t the finish line. It’s the race. That’s not “polishing” a font. It’s designing one.

What’s being described is a professional quietly doing the actual type design while the result ships under an “AI-generated” headline. The 90% you’re calling polish is the craft. At that point the only thing the AI generated is the press release.

Welcome to the centuries-old craft of type design.

John Hudson · September 15

@Simon Cozens Excellent counter-troll.

ericlu · September 15

Thomas Phinney said:

Why not put that money/effort into improving Mixfont in the same areas?

There already exist auto-spacing, auto-kerning and especially outline-optimization systems that do better.

I have been working on improving these areas (with input from helpful folks here) but it's still difficult for the technology to create fonts that are on level of quality of professionally designed fonts with respect to all the details in one single prompt / pass. I think it's like AI in other areas where it can provide a quick and helpful starting point but still requires review. Hence seeking to partner with a human designer.

@yanone
When exactly did I say AI is doing 90% of the work? All aspects of the font design are an equally important part of the work, and because of that I'm trying to get it right. This is partnership work submission, where AI will generate the first draft and then I'm hoping to work with a human designer to turn this drafted design into a production ready font.

@Kasper Pyndt
It's still an AI generated font but it will be created in partnership with human designers. For the right collaborator I think this could be an interesting project to contribute to, as of yet there hasn't been a submission like this to any large professional font library. I believe this can be a big milestone to be a part of.

@SCarewe @John Hudson
This isn't a troll post. I'm seeking a collaborator on this and I wanted to see if I could involve this community first because I wish to build this in collaboration with this community of designers. It's fine if you're not interested but it's my right to ask in case someone else might be. I'm determined to make this happen so I think it's better if I can do it with someone here who understands the voices and concerns of the type design community.

Kris Sowersby · September 16

What is your budget to complete this work?

Simon Cozens · September 16

You may not realise that you're trolling. That doesn't change the facts. I'm not going to explain again why you're trolling because people have done that multiple times in great detail and it hasn't gone in. You don't care about the discipline; you just want the PR win of having the first "AI font on Google fonts".

Bad news, Eric. You're too late.

See, ten years ago, I wrote a program called Sans-o-Matic. It's been used hundreds of times. It automatically draws key glyphs for sans-serif fonts. All it took was a human designer to tidy up the spacing, kerning and draw the rest of the glyphset, and boom, an automatically generated font.

(To be explicit for people not trained on my sense of humour: this is a provocative way of asking "how much automation makes something 'an automated font'?" The answer is "100%")

Thomas Phinney · September 17

This is definitely a slippery-slope question.

Going the other way, synthesizing one or more glyphs on the fly, from a multi-axis variable (or MM) font, in a specific style so as to match the style of the host font, is something that was done as far back as 1999. https://patents.google.com/patent/US6760029B1/en (Yes, that’s me in there. I contributed the idea, heuristics and made the initial MM font from which euro glyphs were generated.) This approach was used by Adobe in their OpenType library conversion to add customized “symbol substitution” glyphs to fonts other than Adobe Originals, and also euro symbols for a more limited number of fonts.

Laurence Penney and Irene Vlachou came up with what seems to be pretty much the same approach for adding Greek to existing fonts with FauxFoundry (https://www.fauxfoundry.com/).

I would very much rather see the AI efforts go into essentially a better AI version of that. The “prompt” being not a text description, but at least in part actual glyphs as inputs. The number of input glyphs could be highly variable, anywhere from say 1% to 99%+ of the desired output glyph set. Different people might have different preferences for how big an input set to start with, what glyphs they wanted generated, and how much they wanted to revise the AI-generated content. One might cycle through repeated steps of output, revise and feed back, more output.

Adam Twardoch · September 17

https://yofont.com/ is probably the best AI font generator ever to exist. Best that had existed, and I think best out of all there to follow. And it’s among the best AI generators of anything.

ericlu · September 18

@Thomas Phinney
That's an awesome patent! Thank you for sharing the link, so cool to see that aspect of your work.

I have been working on that workflow - in fact this summer I created a Glyphs plugin (https://github.com/schriftgestalt/glyphs-packages/pull/199) that takes the glyphs as inputs, and using the Mixfont model converts them into editable glyphs (it can also expand upon the already created glyphs). I'm hoping to get this plugin accepted to Glyphs btw (in case anyone here might be able to help with that). The goal is exactly the workflow you're describing - use actual glyphs as input to either expand or improve, and be able to iterate on that.

@Kris Sowersby
I'm following Thomas' excellent article as a guide on the cost of developing new fonts for setting a budget for this project. I believe that the comp should be competitive - if you are seriously interested please send me a DM with some samples of work and we can go from there.

@Simon Cozens @Adam Twardoch
YoFont does look like a legitimate competitor. I would love your feedback on anything specific that you like about their product over Mixfont or any ways that Mixfont could improve. I am here in this forum so that I can learn and improve the product.

I'm not sure how and when I became the bad guy, but I do care deeply about getting this project right. Whether or not this submission to Google Fonts can truly count as a "world's first" might be subject to different interpretations, but I believe that it will be a notable milestone regardless.

Simon Cozens · September 18

ericlu said:

@Simon Cozens @Adam Twardoch
YoFont does look like a legitimate competitor. I would love your feedback on anything specific that you like about their product over Mixfont or any ways that Mixfont could improve.

Sorry, this time we're trolling. YoFont is bloody terrible. The fact that you can't see that is a problem. Or, in the context of this discussion, the problem.

The "Arabic" and "Devanagari" it generates is just completely wrong; glyphs in the wrong slots, the output is incoherent and unreadable, and the style variation is minimal (the Arabic is just Microsoft Arial Arabic with a few small outline variations) and utterly disconnected to the Latin. The only reason that YoFont can claim to generate a complete Arabic or Devanagari font is purely the Dunning-Kruger effect on the part of its creator - the fact that the creator of YoFont has no idea what Arabic or Devanagari actually looks like, but has been convinced by their AI companion that the job is good enough for production. And I suspect that you think it's a legitimate competitor only because you have no idea what good Arabic or Devanagari looks like either.

Turns out it's really easy to generate a production quality font if you have no idea what a production quality font looks like. Bears thinking about, doesn't it?

SCarewe · September 18

ericlu said:

I'm not sure how and when I became the bad guy

Well. Your unwavering motivation to build a product that everybody here is telling you is harmful, unhelpful and misguided. You don't seem to listen to any of the concerns, as all points made in the numerous threads are just the same ones, over and over again.

Please just stop for a moment and take the advice of the type enthusiasts here to heart. I really, really recommend you read Simon's posts and reflect on them. He means well and is giving you all the tools to come out of this a better person.

A note on your Glyphs plugin: the fact that you don't see how terrible the output is (even just from a design perspective) is exactly what people here are telling you about.

Erik · September 18

At the risk of parrotting SCarewe:

ericlu said:

I'm not sure how and when I became the bad guy

I donʼt know either, but I hypothesize that itʼs connected to the adoption of the philosophy expressed in this comment, in another thread:

ericlu said:
I think that if it wasn't me, it would be someone else who invents and productizes this technology.

Arguments of the form “if I donʼt do ＿, someone else will” are an ethical red flag, if not a sure one. Theyʼre especially suspect when rationalizing an action one already desires for personal gain amid ethical concerns against said action. (The same form might appear within a more ethically grounded argument, as part of the motivation to take an uncomfortable action, for example, “if I donʼt talk to my kids about drugs, the pushers will”.)

Whenever and however it happened, if you wish to become a good guy, itʼs of course ultimately up to you to proceed with identifying and implementing the changes to make that happen.

John Hudson · September 18

Arguments of the form “if I donʼt do ＿, someone else will” are an ethical red flag, if not a sure one.

If the British East India Company hadn’t seized control of India and looted it of its wealth and resources, the Dutch or the French companies would have.

AI is digital colonialism. There are big entities driving the plundering of creative wealth and cultural resources, but also opportunistic chancers trying to grab their piece of the loot.

Kris Sowersby · September 18

@ericlu Honestly mate, I wouldn’t touch this no matter how much you paid me. The letters you’ve prompted have no mana, no spirit, no dignity. There’s no concept, no connection to the people who have come before us and no respect for the people who will come after us.

For reference, and for anyone else reading this thread, it should be at least $10,000 USD to take this work on. If you’re paid any less you’re being severely underpaid.

ericlu · September 18

@Simon Cozens
I haven't tried out YoFont at all so I can't comment on their results. I'm focused on improving the results from the Mixfont model, and part of the reason that Mixfont hasn't expanded to non Latin characters is because I want to ensure high quality for a smaller subset before expanding to other languages.

@SCarewe
Did you end up trying out the Glyphs plugin? If so, I appreciate you testing it out! Since each generation is unique, some outputs will be higher quality than others. I would love if you could share the generation that you tried and felt was "terrible" - this would greatly help me improve the results over time.

@Erik
@John Hudson
I understand and empathize that there are many concerns with AI, but directing frustration entirely towards me is misguided. Part of my research has also been to create anti-AI fonts (like decoy font) that help people protect themselves from AI.

There's a lot of negativity in this thread so I thought I'd share some of the positivity that I've received privately from folks in Typedrawers as well as from users from all around the world. It's critically important to me to listening to all perspectives, consider feedback from all angles, and to learn and grow to ensure that the work I'm doing is meaningful and helpful. I wouldn't continue my research and work if I believed Mixfont wasn't net beneficial for humanity. Since I've launched I have been learning from type resources both online and in person to try to improve my understanding of the craft. Type Design Resources has been an invaluable resource for this.

Going back to my original request - I believe that submitting a font to Google Fonts will be a significant milestone for evaluating the quality and progress of the model and I would love to find an interested partner to work together on this project. For the record, I think that @Kris Sowersby comment on budget is more than fair and I was also thinking along these lines in terms of numbers. If you may be interested in this work please DM me.

Final

SCarewe · September 18

I didn't try your Glyphs plugin. I looked at the demo video and the material you show. Again, you not seeing why it is so bad is the actual problem. Or, rather, that you are replacing any form of process through which you could learn the craft and train your eye.

What is the net positive here? Humanity gets more slop. Really terrible quality, generated by machines trained on stolen data, with no soul whatsoever.

Does this really give you purpose? Other than being the one to do it rather than somebody else?

Out of curiosity, do you practice any form of art? Do you play an instrument?

Erik · September 18

ericlu said:

@Erik
@John Hudson
I understand and empathize that there are many concerns with AI, but directing frustration entirely towards me is misguided.

I did not “direct[…] frustration entirely towards” you (nor in any direction); I speculatively answered an indirect question which you posed (if but rhetorically). Misguided is mischaracterizing my motivations and especially my words that are available for all to see.

John Hudson · September 18

I understand and empathize that there are many concerns with AI, but directing frustration entirely towards me is misguided.

Trust me, this ‘frustration’ — actually, a developing social-economic and historical analysis of AI in the context imperialism and colonialism — is not entirely directed towards you. But you’re the one who has shown up in our village.

Ray Larabie · September 19

@ericlu Humanity was built on a foundation of double-edged swords and AI is another one. I'm not an opponent of AI in general and kind of an enthusiast...sorry to disappoint many of you. Current AI can already improve fonts and some of us are using it for that. Not designing, but for detecting technical errors, writing scripts etc.

What generally I don't think is good about this project is that it seems to be attempting to solve the wrong thing and doing it wrong way. We have already have a million fonts, and a certain percentage are good ones. This project isn't about making better fonts; it's about novelty, amusement and circumventing licensing. Instead of getting a quality font from a reputable source, it's generating the opposite.

People on this site have criticized the MyFonts ChatGPT plugin, but the intention was to help people or AI agents find solutions to their problems. The near zero-friction solution is Open Source, and that was the option that AI models often presented as the only viable option. But Open Source might not be the best solution. Sometimes it really is the best solution. Sometimes, a higher friction ($) option is the better solution. Either way, the ideal setup is that it's supposed to be solving a problem.

But by generating these fonts, you're presenting an option that appears, to the uninitiated, to solve a problem, but doesn't do it reliably. And this is a case where the AI-slop term truly applies.

Not having to pay for licensing is the service you're providing. But this is just making fonts that aren't good for people who don't know better, and that just makes the world a shittier place.

Thomas Phinney · September 19

Ray exaggerates. We only have about a quarter or a third of a million fonts. And a lot of them suck.

I also agree that AI is useful for a bunch of things. I also worry about its unpaid costs, but that doesn’t mean it doesn’t “work” for some things.

Besides the fact that this is a really, really hard task... another fundamental problem here is, you would need a font expert to help you teach the AI how to make fonts that don’t suck. You are not (yet) such an expert. It is not clear whether you have the particular turns of mind to be well suited to become one.

BTW, the best intro book on type design is finally back in print with a new second edition. How to Create Typefaces: from sketch to screen, by Henestrosa, Meseguer and Scaglione. You should buy it ASAP. I bought five or six copies.

Simon Cozens · September 19

ericlu said:

I'm focused on improving the results from the Mixfont model... I believe that submitting a font to Google Fonts will be a significant milestone for evaluating the quality and progress of the model

Sorry, but your position here makes no sense. So let's say that you are telling the truth and you're focused on improving Mixfont. Passing the Google Fonts quality bar is definitely a good milestone to evaluate where you are in that. And right now, you know that your model is not at that stage. That's a good awareness to have.

What would an honest person do here? They'd say, "My model is obviously not good enough, I need to improve it." They're be spurred on by the rejection, double down on the goal that they'd set themselves, and work harder to try to achieve it. Failure is what drives entrepreneurs towards success; real entrepreneurs don't just give up and accept their own mediocrity. They would not cut corners; they would work until they'd got it right.

What would a charlatan do here? They'd do two things. First, they'd deny the failure and project success. We know exactly what everyone's favourite charlatan would say when confronted with their model being sub-par: they'd say "This is a great model, there hasn't been a model like this, many people are saying that this is the best model ever. Many people."

Second, they'd try to smooth over their failure by getting a human to "get it over the line". They'd cut any corner available and they'd move the goalposts they'd set for themselves even while still publicly claiming they'd "got an AI generated font onto Google Fonts". Where's the success in that? Where's the achievement? Doesn't matter to them, because for the charlatan it's not about success or achievement, it's about deceit. The Great Oz hides a man behind the curtain because he's trying to deceive people.

I don't know which one you are, but this is the inflection point. This is the moment that you get to decide. If you're the honest person, get back to work until you've got the model right. If you're the charlatan, I'm not going to stop you, but please do it somewhere else. But trying to be both is incoherent.

Yves Michel · September 19

Simon Cozens said:

... but please do it somewhere else.

Then comes the final curtain...

Joshua Langman · September 19

BTW, the best intro book on type design is finally back in print with a new second edition. How to Create Typefaces: from sketch to screen, by Henestrosa, Meseguer and Scaglione. You should buy it ASAP. I bought five or six copies.

Thanks, Thomas — I just did.

Gaspar Muñoz · September 19

Ahhh, I’m going to write pretty much the same thing everyone else is saying. Apparently, you simply don’t want to understand that the problem lies in the ethical (and possibly illegal) dilemma posed by your platform and your “idea of helping.” I don’t understand how this would “help” type designers. As soon as you enter the website, it already demonstrates how to imitate the work of other designers.

I don’t want to add much or drag this out, because others have already made solid arguments and you haven’t been capable of accepting them. The argument that you’ve gained popularity seems completely irrelevant to me. Piracy and license theft are popular too, and that doesn’t make them any less unethical.

To finish, I’m simply asking you to read the ideas everyone has put forward, stop playing the victim, and actually reflect on your project. It seems to me that you have a lot of knowledge in general, and that it could be useful to many people, but you’re approaching it very, very badly.

Peter Constable · September 19

This thread makes me wonder if there's another possible motivation for C2PA manifests in fonts...

C2PA = Coalition for Content Provenance and Authenticity. A C2PA manifest can be bound cryptographically to a digital artefact to indicate its provenance. For example, a photo or video can include a C2PA manifest that can verify it was shot on a Nikon camera and published by CBC. A manifest can be removed, but the absence of a manifest simply provides no information about the provenance. It is presence of a manifest that can be useful for a consumer.

Now, a C2PA manifest can indicate that an item was AI generated (which was part of the purpose). The EU AI Act (or, rather, the code of practice for its enforcement) now requires AI vendors to embed a C2PA manifest in AI-generated images, videos and documents to make that clear to recipients. There is very strong incentive, and companies are starting to do this.

A few years back, folk from Monotype were proposing that C2PA manifests be added to font files to help identify their origin and identity. They suggested it would help in responding to tech support. At the time, it wasn't to some of us clear how broadly useful that would be. But embedded metadata indicating a font was AI generated could be useful information.

Of course, there aren't incentives for anyone making a tool for AI generation of font data to include such metadata.

On the other hand, a C2PA manifest could provide a production history demonstrating that AI was not used in creating the font.

Andreas Stötzner · September 19

I guess the label “AI-free human creation” will soon become a quality seal of fonts (…photos, videos &cª) as “GMO-free” is a quality label for yogurts.

Mixfont Update - seeking a designer to help complete an AI-generated font

Comments

Categories