AlphabetMagic. My first AI experiment

Options
15678911»

Comments

  • Nick Shinn
    Nick Shinn Posts: 2,152
    edited May 2023
    Options
    @PabloImpallari
    How else could there be such an exact coincidence?

    Well, you might look a little further afield than the digital domain.

    Cooper Black originated in Oz’s lettering, which does not seem to have been traced over Bodoni!
    Cooper Black was published by Barnhart Brothers and Spindler, which did not have a Bodoni, AFAIK.
    There may be some other explanation for the “exact coincidence”; I suspect it is some kind of artefact of the digital fonts on which your AI was trained. Forming conclusions about RW history from such things is dubious.   





     

  • PabloImpallari
    PabloImpallari Posts: 783
    edited May 2023
    Options
    Cooper Black was published by Barnhart Brothers and Spindler, which did not have a Bodoni, AFAIK.
    That's a good point... it made me think.....
    Maybe is the other way around and Bodoni Black was draw having Cooper Black as a reference. Which sort of make sense given Cooper Black's success. Goudy did his Heavyface to compete and maybe the ATF guys tried jump inn with Bodoni too.

    Whatever the case, I stand my point that they are linked, even if I made a mistake in the chronological order. Thanks Nick for digging in to this!
    I will add a "notes.txt" on Github.

    Well, you might look a little further afield than the digital domain.
    Speaking about that... I'm trying to re-create the my list of specimen and lettering books. I used to had a huge one, but it wasn't online and got lost when my computer got stolen.  I don't what that to happen again, so this time is on-line.
    If someone has time and desire, I would like to invite you to collaborate with more links to complete it.
  • Nick Shinn
    Nick Shinn Posts: 2,152
    Options
    Those Cooper images are taken from:
    The Book of Oz Cooper, published by the Society of Typographic Arts, Chicago, in 1949.
  • PabloImpallari
    PabloImpallari Posts: 783
    edited May 2023
    Options
    Today I have been working in a model to create the Heavy master from the Regular master. I'm not yet there, but I think i'm on the right path to it.

    Cabin Regular:


    Make it Heavy!


    I need to keep improving the model to properly isolate the "Heavy" concept, so it is more respectful of the original forms (So it does not replace the /i dots, the /g, or the /v /w apex, /a tail, etc.... the "weight" concept needs to be fully isolated for the "shape" of the letters, and not to transform the original letters beyond a "reasonable point"... the thing is that finding that "reasonable point" is not so easy).
  • Thomas Phinney
    Thomas Phinney Posts: 2,778
    Options
    One thing is bugging me: the model needs to have optical correction at the joins. Normally this increases as the weight goes up.
  • PabloImpallari
    PabloImpallari Posts: 783
    edited May 2023
    Options
    Yes, is bugging me too.

    But also, another thing that bugs me even more is that the "heaviness" is not as heavy as it can be.... what I really want as a result form the model is more like an "Ultra" more than a standard "Heavy".

    Since there is not a standard definition of what a heavy weight is, I included many "heavys" in the training data that are not really all that heavy... for example, AtlasGrotesk Black or Acumin Black, are not as black as AGBook Black or AktivGrotesk Black. There are different degrees of "Blackness" ranging from font to font, even within the same design features.

    This is the Cabin Heavy master I did by node-dragging back in 2020:
    The goal I'm trying to achieve for the "MakeitHeavy" model will be to get a result as close to this as possible. (ignore the /g.. I can't remember why I decided to go with a single story g back in 2020)


    The logic to make it happen in theory is already there. I think I need to keep refining the training data selection to make it happen in practice.
  • PabloImpallari
    PabloImpallari Posts: 783
    edited May 2023
    Options
    Just For Fun:
    These are some of the RayWantsCrazy experiment result turned into fonts.

    If "Good text fonts should be invisible" and go unnoticed to the reader, then good display fonts should be all the opposite: Super visible to draw all the attention possible from the reader and catch his eye.

    Crazy Franklin Black:



    Just lines:



    Mecano / Futuristic:



    Old Greek inscriptional style fonts are -to me- cool, but boring.
    By adding a blackletter style emphasis to the initial uppercase, they become more fun 



    They may be crap, or not... but they look so much different from everything else....
    I'm loving n2 "Just lines", and can imagine it being used for vegan pseudo-food brands or things like that.
  • PabloImpallari
    PabloImpallari Posts: 783
    edited May 2023
    Options
    The Esoteric one:
  • PabloImpallari
    Options
    StyleDrop just released by by Google AI
    https://styledrop.github.io/
  • Dave Crossland
    Dave Crossland Posts: 1,395
    Options
    Any news Pablo?
  • Dave Crossland
    Dave Crossland Posts: 1,395
    edited August 2023
    Options
    The type shown on https://ideogram.ai looks very promising.


  • Nick Shinn
    Nick Shinn Posts: 2,152
    Options
    New and improved “pseudo-random” feature!
  • Dave Crossland
    Options
    And the spacing needs work ;)

    But the legibility, and the overall cohesiveness of forms, is very high, compared to stuff that was seen just a few months ago. 
  • James Puckett
    James Puckett Posts: 1,978
    Options
    I wonder if the AI learns well enough that it studied old lettering manuals available on the web. Try generating spencerian stuff.
  • Dave Crossland
    Options
    https://serce.me/posts/02-10-2023-hey-computer-make-me-a-font

    I expect another announcement some day from someone heretofore unknown to the type community who posts something like this which is really really good. Just personal speculation on my part. 
  • PabloImpallari
    PabloImpallari Posts: 783
    edited April 17
    Options
    "AI is like an immigrant... learns fast, works hard and is here to steal our jobs"
                                                                                                                    - Rafi Bastos
  • Cata
    Cata Posts: 1
    Options
    Hola Pablo! Te he enviado un mail interesada sobre el tema :)