State of the art in AI image generation as we go into 2026.

What is interesting for me while interacting with ChatGPT on various topics is its pharisee position on general ethics, when advising one how to deal with real-life problems. While being pretty much workaroundish about his moves  

I do not complain about it, just notice how it kind of resembles a widespread parenting model when some of us were kids 30-50 years ago (do not do what I do, but do as I tell you).

Chris Lozos said:

To me, AI is only of value if it can do the drudgery part of the job, not the creative part.

It feels to me like it may be precisely the opposite: AI spews plausible-but-problem-filled output, and it’s up to the humans to clean it up.

The video at https://openai.com/index/introducing-chatgpt-images-2-0/ is worth a watch, as well as scrolling down the whole page 

It isn't "AI Generated Fonts" that are vexing type designers, I think, but that this creates final production-ready documents without ever touching any font files at all. 

That future is already here, it just isn't evenly distributed...

I didn't look with a microscope but there was only a single Latin glyph I noticed that was drawn incorrectly (an uppercase G)

I asked a few friends and the non Latin is still behind, like where Latin was about a year ago. It's questionable how much source material exists for non Latin vs the overabundance of Latin in the world... But ... Tick tock.

I did scroll down through all the images in the presentation, and it would be churlish not to acknowledge that the sheer productiveness and surface sheen is impressive. But they are also 100% pastiche of styles of images previously made by skilled human beings. Ironically, the style used for the explanatory image panels is pretty close to what I anticipated in 2023, in the conclusion of my chapter on text technologies for the Bloomsbury Handbook of Global Typography: a pastiche of modernism, lacking any of the spirit of modernism. The bad thing that the machine is very good at.

AI type tends towards the generic, unsurprisingly, merging things the machine has learned from multiple models. A lot of type designers are going to see familiar forms and proportions in the results, without necessarily being able to point to direct copying or a clear path of derivation. Apart from the legal and ethical questions, consider just how characterless and boring these generic letterforms quickly become.

Ray described a hypothetical business owner using AI to produce first a graphic identity and style sheet, and then using AI to produce a whole variety of textual, graphical and video material. It might not pass analysis by a typographic expert, but would be, he suggests, ‘satisfactory’ for the client and customer. We’ve been here before. In the 1980s, during the desktop publishing revolution, business owners or their staff were creating brochures, reports, signs, etc. and they often thought the results satisfactory or, at least, good enough, and cheaper than paying someone who had knowledge, skills, and experience. But the results didn’t remain satisfactory, and while DTP killed off typesetting per se as a business, professional graphic design and branding agencies flourished because companies wanted to differentiate themselves and their products from ‘good enough’.

The massive derivative productivity of AI may similarly produce a fresh demand for non-AI creative work — the work of making things that have not been made before —, precisely because it becomes too easy to make a masssive amount of very repetitive stuff very quickly. Things like Studio Ghibli films — to use an example that has already boomed and busted as something people find interesting to pastiche with AI —are special because they are infrequent, carefully crafted, and the product of the vision and inventiveness of particular people with particular experiences of the living in the world. Being able to rapidly generate things that superficially resemble Studio Ghibli animation in massive quantity is technically impressive and incredibly boring very quickly. I predict the speed of boredom will ultimately exceed the speed of AI generation. People will get bored of what they are making even while the prompt is still being processed.

I am not much fussed by the notion that some significant quantity of text will be created without fonts. Making fonts is how I make my living, yes, but I have always been interested in a broader history of text manufacture that encompasses all manner of tools and media. If people are going to make texts using AI without fonts, that AI is going to need new models. The first step will probably be direct reference to specific existing typefaces, i.e. prompts that include the names of the fonts the user wants the text to be displayed in, with the expectation that these will be accurately rendered. People like specific typefaces, not generic serif and sans. They have their favourites, and just as people running Monotype machines wanted types that were only available on Linotype, and vice versa, they are going to want the new machines to give them access to what they know and like. The next step will presumably be the design and creation of new type styles specifically as models for machine learning. This may well involve much smaller glyph or character sets than designing a complete font, leveraging AI to fill in the blanks, but probably with an iterative approach allowing the maker to fork AI-generated glyphs, edit them, and feed them back into the system.

Several years ago, I designed the Bodoni Egyptian® typeface. Sounds like a prompt, doesn’t it?
Easy enough for AI to mimic, with or without fonts, but I’d like to believe that there’s something in the way that I designed and drew it that represents an inimitable human quality that people who specify type style can relate to. My taste; not to everyone’s liking, but unique and special enough, as determined in a font, with all the specificities of spacing and kerning, and not the result of a user’s prompts spun out by algorithms.

When I designed Neology®, which mixes grotesque and geometric characters in a quite readable manner, I discovered, in using the typeface for various projects, such as my blog, that I didn’t like the mixture and really did prefer the specificity of the Neology Deco version. From this I conclude that the specificity that a type designer embeds in a particular font—a particular weight, too, not an elastic variable instance—is what makes a typeface useful to the highest, and most distinctive degree. As a conceptual entity that typographers can work with in creating layouts, useful because it has meaning for them, connecting with so many other signifiers of design culture, in their understanding of what that is. And this is surely what art directors want for the brand they’re promoting.