Thursday, January 16, 2020

Wall Street Journal's Hedcuts

Top row: hand-drawn portraits; Bottom row: computer-generated versions.
The Wall Street Journal has developed an artificial-intelligence system for creating its distinctive 'hedcut' portraits.
Human-created hedcut of Grumpy Cat, 2013, courtesy Wall Street Journal
Hedcuts have a wood-engraving or scratchboard look, made up of dots and dashes.

Left: human-created 'hedcut' of actress ChloĆ« Grace Moretz
Right: AI-created hedcut, courtesy Wall Street Journal
The AI learned the style and produced adequate results in most cases. But there were difficulties. One challenge was "teaching the tool to render hair and clothes differently than skin, which was often a matter of whether to crosshatch versus stipple."

Error cases caused by AI working with too limited set of data,
courtesy Wall Street Journal
"The most harrowing issue of all was overfitting, which happens when a model fits a limited set of data too closely. In this case, that meant the machine became too satisfied with its artistic ability and began producing terrifying monstrosities like these."
----
Read more on Wall Street Journal: What’s in a Hedcut? Depends How It’s Made.

12 comments:

Erich Jon said...

Progress I suppose. But it's going to affect the livelihood of five talented artists.
http://www.hedcut.com/blog--news/meet-the-wsj-hedcut-illustrators

Wendy said...

Amazing! What I am seeing here is that when you get a machine to make an image you get an incredibly mechanical, sterile, and boring result. Of course think of the savings in not paying someone to make something that is attractive and has appeal. Revolutionary!

Bob Easton said...

The AI machine certainly doesn't know enough about eyes. All portrait artists, and most of the rest of us, know that eyes strongly express personality and energy. Get them right and we see the real person. Do them like the AI machine does, and we get only lukewarm renderings.

Angeline said...

AI portraits look cross-eyed and a bit off.

I heard from an employee at a major publisher and toy company that the books they were publishing had covers created by AI. The covers usually feature a girl. The sales of the books were decreasing but the company didn't understand. The employee showed me a cover done by a human artist versus the AI artist. There were definitely differences between the two cover arts. It doesn't matter that you can't quite articulate what's wrong...it matters that there is somethings wrong.

If the publisher(s) keep using AI for arts, AI will have to get some kind of "feelings" to help. Perhaps buyers will know something is wrong and simply stop buying from these publishers until said publishers figure out we need and prefer human artists?!

Roberto Quintana said...

I think I’ll add that to my resume’: “Hire me, I produce adequate results in most cases.”

They are going to have to turn up (or down?) the sensitivity on that algorhythm, and add more data on anatomy and form. That cross-hatching on the faces looks like scarified face-tattoos, not delicate shadow contours defining the form and anatomy of the facial structure.
I think I can guess which portrait-artist Chloƫ Grace Moretz would hire.
(She might still think there was something wrong about the mouth, though ;)

I am much more impressed by the ’overfitting’ technique. At least the machine can produce a creative image, instead of a poor copy.

Reminds me of Sir Francis Bacon or Yoon Miseon.

Sir Francis Bacon:
https://www.google.com/search?biw=814&bih=530&tbm=isch&sxsrf=ACYBGNQ3no5LYhqQG2AgYSeuwW0kYb1qfg%3A1579198469599&sa=1&ei=BaggXsLiI4WStAaJvaaICQ&q=francis+bacon+distorted+portraits&oq=francis+bacon+portraits&gs_l=img.1.4.35i39j0j0i8i30l6.0.0..196505...0.0..0.108.108.0j1......0......gws-wiz-img.Ur0_JNkaidU

Yoon Miseon:
https://www.google.com/search?q=Yoon+Miseon&sxsrf=ACYBGNQ19t_f19Xci91xySN0pDlYc0n_sA:1579198874454&source=lnms&tbm=isch&sa=X&ved=2ahUKEwiJy8Of3ojnAhWbVc0KHYCRAEsQ_AUoAXoECAsQAw&biw=814&bih=530

@ Erich: Thanx for that link. -RQ

Lou said...

Sigh...
To me it no longer matters how well AI can produce an "artistic" likeness of anything. I'm over it. It bores me in the same way a plastic cup or a car made by robots bores me. No soul, no uniqueness, no delightful irregularities. A big yawn. Eventually AI "art" will get so good it will be impossible to tell between a real John Singer Sargent portrait and one created by AI. I wouldn't care to see it.
Every artist that's unemployed by AI and is forced to accept non-skilled low paying work adds yet another person to the mounting numbers of lower-class wage earners. I do what I can to support artists and craftsmen. I hope that others will find their way to do the same.

James Gurney said...

We all can see slight ways that the human work is aesthetically more pleasing than the AI version, but in the realm of newspaper illustration, what really matters is which approach art directors find more useful from the point of view of cost, efficiency, and creative options. Also, keep in mind this technology is just beginning to iterate, so it will quickly get better.

James Gurney said...

Another way to look at this is from the consumer's perspective. Does it matter to you or me if most of the weather forecasts and sports summaries in newspapers are written by AI? (because they already are). I'm OK with it as long as it does the job. But somewhere out there, a sportswriter is reinventing his or her career.

Drake Gomez said...

What I'm curious about is how the "hand-drawn" images were done--scratchboard, straight pen and ink, or...are they digital? Not AI, perhaps, but not really drawn, either, at least not in a traditional sense. Viewing them on a monitor, I'd bet on the latter. Unlike other commenters, I don't necessarily find the hand drawn images more satisfying than the AI ones, nor do I prefer the AI images. Both are too closely tied to their photographic source for my taste. I agree that the overfitted images are frighteningly like a Francis Bacon!

JaneBucci said...

http://www.hedcut.com/ I can't imagine a computer getting to this level -- Yikes!

Bill Marshall said...

Love to see an AI hedcut of a Screaming Pope.

Bill

James Gurney said...

Fascinating how these systems can do a pretty good job at simulating a style without knowing the "rules." No one sat down and taught them to do A,B, and C to achieve that style. If the computer has a big enough data set, it can figure out what to do, but it can't tell us what it knows or how it learned.