This was a kind of fun write up to read, but maybe instead of teaching people to use an LLM to make fonts we just teach them how to make fonts lol
3
Mechanismatic - 2w
Definitely not the world's first AI generated font. I imagine there are many out there already. Not everyone is likely advertising it though.
It's a relatively limited set of characters and the design is nothing new. Dripping fonts have been a thing for a while.
Looking at the font closer shows where an LLM can't replicate human effort well enough yet - the kerning is non-existent.
The space character in the set is blank:
The punctuation is disproportionate in size to the letters:
Which brings us back to the fact that you need a human with expertise/experience to understand that the LLM isn't performing as well as a human creator, just (potentially) faster and more recklessly, and to fix the issues with the output to make it actually usable/functional, which just feels like correcting someone else's homework for them while they get the credit for the result.
2
Justin (koavf)🌻🍉🇪🇭🌿 - 1w
Every rational person understands that tools like LLMs are useful for certain specialists and professionals to do a B+ job of some mundane work. It does not replace that expertise. Tragically, for many, a B+ that includes a few things that are just untrue and sometimes is actually a D+ is fine as long as it means that you can avoid paying someone.
koavf in typography
Worlds first Ai generated font, using nano banana.
https://constanttime.notion.site/Worlds-first-Ai-generated-font-using-nano-banana-2ba6f8e15af18012864bdb760fa9c9baUgh.
This was a kind of fun write up to read, but maybe instead of teaching people to use an LLM to make fonts we just teach them how to make fonts lol
Definitely not the world's first AI generated font. I imagine there are many out there already. Not everyone is likely advertising it though.
It's a relatively limited set of characters and the design is nothing new. Dripping fonts have been a thing for a while.
Looking at the font closer shows where an LLM can't replicate human effort well enough yet - the kerning is non-existent.
The space character in the set is blank:
The punctuation is disproportionate in size to the letters:
Which brings us back to the fact that you need a human with expertise/experience to understand that the LLM isn't performing as well as a human creator, just (potentially) faster and more recklessly, and to fix the issues with the output to make it actually usable/functional, which just feels like correcting someone else's homework for them while they get the credit for the result.
Every rational person understands that tools like LLMs are useful for certain specialists and professionals to do a B+ job of some mundane work. It does not replace that expertise. Tragically, for many, a B+ that includes a few things that are just untrue and sometimes is actually a D+ is fine as long as it means that you can avoid paying someone.