Why are there no font managers that can sort the fonts with AI?

elenalauder
elenalauder Posts: 1
edited November 2023 in Type Design Software

Hello everyone,

I hope you are well?

I wonder why there is no font manager that sorts the fonts by style with the AI. There used to be a software for years, it was called TypeDNA from Sweden, it worked quite well, it searched for similarity of a font with the fonts you owned. Unfortunately, the company no longer exists and the software could only be used with an internet connection to the company's server to verify the license.

AI is all the rage, AI there, AI here, AI then, everything is AI.

Has AI even reached the font manager developers?

I mean, this is such a good function.

You search your font collection for a font that should look similar, which shortens the design process immensely.

What do you think? Or is there already something like this, or how should I write directly to the Font Manager developers and ask?

Thanks in advance for your input.

Have a good time,

Elena


Tagged:

Comments

  • Thomas Phinney
    Thomas Phinney Posts: 2,883
    Extensis Connect (formerly called Suitcase) has a feature called QuickMatch that looks for “similar” fonts. They have had that feature for about a decade now. It’s ok! 

    Desktop font management is a tough business. The OS level functionality is “good enough” for many users, so there is less room for retail products to make money. Adobe dropped ATM Deluxe because there was no room for growth and it wasn’t a big enough product. Monotype got rid of FontExplorer as a separate product. (Extensis is still around, but pursuing cloud-based solutions for teams, not so focused on individual users.)

    It is not a great sign of a healthy market when repeatedly, the biggest players just decide it is not worth playing.   :'(
  • John Butler
    John Butler Posts: 290
    My favorite tool for finding similar fonts is fontsinuse.com. Search for a typeface by name, then check out the entries under Related Typefaces. They’re related and not necessarily similar, and they’re human-curated, but it’s the most vendor-neutral, well-maintained resource for this task in 2023.
  • Nick Shinn
    Nick Shinn Posts: 2,207
    Identifont is also an independent, “neutral” site, run by an actual person, David Johnson.
  • James Puckett
    James Puckett Posts: 1,992
    Programmers who can code AI software get paid hundreds of thousands of dollars a year (I read yesterday that one of the senior programmers at OpenAI is getting paid $800,000 per year). Making this work would require purchasing special licenses for most digital fonts and analyzing them with the AI; that’s going to cost millions of dollars. The primary market for this software would be people who have collections of thousands of fonts and pay for their software, which probably isn’t a big enough market to cover the costs of developing AI based font management software.

  • Thomas Phinney
    Thomas Phinney Posts: 2,883
    Exactly what I was getting at. AI is expensive, and font management in particular is not a big enough market. AI is coming for the bigger markets first.

    AI will eventually hit type design, but it is taking longer than some other areas, again because of the cost vs payoff problem. In this case it is a bigger market than font management features—but the problem is vastly more complex as well.
  • Floor van Steeg
    Floor van Steeg Posts: 12
    edited November 2023
    As a solo developer of a font manager for macOS: it's technically complex. AI (or rather, machine learning and classification) is hard to get right.

    The classification process should have a high accuracy, otherwise it doesn't make sense to auto classify fonts. For example if the classifier would be correct 80% of the time to determine if a font is serif or sans-serif, you'd still need to go in manually to adjust all the false positives and false negatives.

    Would that be worth it? It totally depends on the accuracy.

    Furthermore, training the classifier is not very straightforward either. Most machine learning stuff works with classification based on images, not vector data. So characters would need to be rendered to images and then used for training. What images to generate? Just a few characters? The full Latin alphabet? What about other scripts?

    Not to mention that you'd need a big library of licensed fonts to train on.

    Of course it would be great if it would 'just work' — this is all responsibility of the developer and not the end user. Just saying that it's very time consuming to make it 'just work' and that's probably the reason why it isn't available right now: it's hard!

    And I'm not getting paid $800,000/yr ;)
  • Jens Kutilek
    Jens Kutilek Posts: 361
    Some years ago, I' seen a presentation by mathematician Jörn Loviscach about automatic sorting/grouping by similarity of fonts, based only on features extracted from the vectors alone. I think it could be a basis for auto-classification without machine learning. There is a PDF available: https://j3l7h.de/talks/2010-07-27_The_Universe_of_Fonts.pdf
  • Dave Crossland
    Dave Crossland Posts: 1,429
    http://vecg.cs.ucl.ac.uk/Projects/projects_fonts/projects_fonts.html

Categories