Home› Font Technology
Options

How many type designers would use an AI assisted plug-in/app

TypofactoryTypofactory Posts: 56
in Font Technology
Hello,
I was reading about automated type design (Link) and looked into AI that could do that (DeepVecFont and DeepVecFont-v2). I asked one of the authors of the paper if he would make a Glyphs plugin, and his concern was whether or not people would use it. 
Heres the link to a Google Form for more standardized information gathering, but feel free to answer in the comments!
https://forms.gle/sVbnP3ouzvAWjyBr6

Comments

  • Options
    Igor PetrovicIgor Petrovic Posts: 263
    As far as I can tell from the recent AI mentions, type designers are not so interested to use automation to replace the "fun" part (drawing letterforms i.e) as they are to replace the "grind" part (mostly kerning). 

    Kerning is a perfect candidate for AI, but is a pretty delicate matter. AI still has to learn a lot about it since sloppy, partial, and unprecise autokerning makes more problems than solutions. It has to be really good in order to be used at all.

    Understanding the importance and complexity of this topic, and focusing AI on it might have much more potential than trying to automate the top level of the type design process. That's my five cents :)
  • Options
    Chris LozosChris Lozos Posts: 1,458
    Automate fair marketing instead of drawing.

  • Options
    John ButlerJohn Butler Posts: 248
    edited June 2023
    IMO the best place for automated type design is for OCRing old books. You can now buy open-book scanners that compensate for the curvature of pages; my PDF reader can already OCR text bitmaps into a searchable text layer, and the next logical step is to derive a decent-quality outline font from that large sample of low-res scanned glyphs. Somewhere diffuse among several applications, the code probably already exists to do this. I’d prefer to read a clean-looking PDF where the font has already been vectorized, than to read some grainy bitmap scan of an old book.
    I’ve not messed with Acrobat proper in forever, so for all I know there’s already something in it that does something like this, or at least with some thousand-axis variable font approximating the metrics of what it scans.
  • Options
    Vasil StanevVasil Stanev Posts: 762
    edited June 2023
    Igor Petrovic said:
    As far as I can tell from the recent AI mentions, type designers are not so interested to use automation to replace the "fun" part (drawing letterforms i.e) as they are to replace the "grind" part (mostly kerning). 

    Kerning is a perfect candidate for AI, but is a pretty delicate matter. AI still has to learn a lot about it since sloppy, partial, and unprecise autokerning makes more problems than solutions. It has to be really good in order to be used at all.

    Understanding the importance and complexity of this topic, and focusing AI on it might have much more potential than trying to automate the top level of the type design process. That's my five cents :)
    I disagree somewhat, spacing and kerning are just as much a part of the design process and a reflection of the designer's personality, as is drawing the glyphs. Some fonts need special kerning that doesn't compute from the positive space of the glyph itself. You could certainly make a better autokern feature, but to rely fully on that, as many would be tempted to do, is actually making them less crafty. This ultimately shortens their wings.
    But someone will do it anyway, giving to non-designers the ability to create bespoke font. Wether this will lead to more or less work for actual professionals will, IMPO, depend on the nature of the font. If you prompt the AI to combine a Bastarda Schwabacher with a pixel font, I don't see how a client will not turn to us to finish it. But a pixel font + a grotesk - then still there will be lots of coding issues. 
  • Options
    John HudsonJohn Hudson Posts: 2,979
    We’ve already talked through a lot of AI-in-type-design topics in this other thread. The most positively received ideas there related to using AI for things like glyph-set extension or completion, e.g. for roughing-in extended diacritic sets based on the characteristics of the design so-far established by the human designer. Put another way, we’re looking for tools to help us do our work, not to do the work for us.
  • Options
    John HudsonJohn Hudson Posts: 2,979
    I think about machine learning in the context of the kinds of roughing-in for which I already use a variety of scripts, and would expect similar amounts of manual refinement of results.
  • Options
    Chris LozosChris Lozos Posts: 1,458
    If you want to design type, then design type. Design your own type.
  • Options
    Florian PircherFlorian Pircher Posts: 173
    It may not even matter what type designer think. The text systems of the future may just automatically synthesize matching glyphs instead of relying on fallback fonts and they could generate “missing” bold, italic, etc. styles on the fly. Webbrowsers already synthesize bold and italic styles, just very crudely.
  • Options
    John HudsonJohn Hudson Posts: 2,979
    I don’t disagree, Florian, but the question in this thread is specifically about AI in the context of font tools, not AI in text synthesis.
  • Options
    Igor PetrovicIgor Petrovic Posts: 263
    edited June 2023
    Vasil Stanev said:
    I disagree somewhat, spacing and kerning are just as much a part of the design process and a reflection of the designer's personality, as is drawing the glyphs. 
    I agree, just wanted to point out that many type designers are interested in having decent auto kerning, which is the fact I believe (since the topic is about what type designers would like to get from AI).
  • Options
    James PuckettJames Puckett Posts: 1,971
    edited June 2023
    I could see using AI to bang out math characters. They’re not too annoying in a single font but I don’t want to fuck with them in a typeface that has width and weight axes with matching obliques. And I’d gladly use AI for for small caps if the results didn’t require more fiddling than scaling+interpolation does.
  • Options
    Ray LarabieRay Larabie Posts: 1,379
    That Google form won't open for me in Edge: You need permission
    This form can only be viewed by users in the owner's organization.

    This question seems to be about Glyphs but as for FontLab, I'd like a checker tool that can point out problems during development. Currently, I find problems by exporting instances, testing the family relationships in TransType, curve glitches in FontCreator, and instances in FontBakery. But a tool that could find inconsistencies during development would be helpful. I don't need it to fix errors, just find them.

    Examples of things things that recently slipped past me:
    • class omissions (I missed an accented Y in a class recently and it was hard to spot)
    • typo in an instance (ExtraBold instead Extra Bold in one field)
    • mismatched weight in an instance (one of my obliques weight axis locations was slightly higher than the upright)
    • Panose set to none/variable for all masters but one
    • typo on linked metrics (minus instead of equal)
    • superfluous anchors
    • flipped anchors on schwa
    • questionable vertical metrics
  • Options
    Enrico SogariEnrico Sogari Posts: 47
    edited July 2023
    AI artists are BS.
  • Options
    Yves MichelYves Michel Posts: 155
    Enrico Sogari said:
    AI artists are BS.
    English is not my mother language.
    Is BS for "brother's son", "Bahamas", "Big Smile" or "bullshit"?


  • Options
    Enrico SogariEnrico Sogari Posts: 47
    English is not my mother language.
    Is BS for "brother's son", "Bahamas", "Big Smile" or "bullshit"?
    Bulbus Stercum
  • Options
    Dave CrosslandDave Crossland Posts: 1,394
    It was for bullshit, in this case.
  • Options
    Christian ThalmannChristian Thalmann Posts: 1,941
    I am ever in awe of the capabilities of Midjourney and the likes (and, to a lesser degree, of ChatGPT), but their main weakness is details (e.g., getting hands right). Type design is pretty much all detail, so I don't see this happening anytime soon.
    I agree that an AI assistant to make all the rarely-used special symbols sounds attractive, but to a certain degree Glyphs already offers that capability.
  • Options
    Enrico SogariEnrico Sogari Posts: 47
    Christian Thalmann said:
    ... but their main weakness is details...
    I wouldn't count on that, they will improve very quickly.

Sign In or Register to comment.