Advertisement blocked. I get it, ads suck. However, this website is run by one person and these ads are the only way we can keep publishing. If you like this website, please consider becoming a supporting member to remove the ads or safelist the website in your content blocker, we will not be able to exist otherwise.

ChatGPT: Images! Text! Copyright Infringement!

AI

Maxwell Zeff, writing for TechCrunch:

It’s only been a day since ChatGPT’s new AI image generator went live, and social media feeds are already flooded with AI-generated memes in the style of Studio Ghibli, the cult-favorite Japanese animation studio behind blockbuster films such as “My Neighbor Totoro” and “Spirited Away.”

Parker Ortolani:

OpenAI dropped an all-new image generation system for ChatGPT today and man is it good. One of the biggest problems with artificially generated images has been the inability to generate accurate text within them. There have historically been problems with inaccurate characters, spelling, or even complete graphical errors. Today’s update to image generation with GPT-4o fixes these. You can now generate charts, signs, logos, word marks, text graphics, pretty much anything you can think of with ease. It nails spelling, seems to set type well, and generally abides by your instructions.

The latest update from OpenAI does a speed run through all the company’s greatest hits. It’s impressive. And it’s morally and legally, at best, in a gray area. The Ghibli stylized images all over social media lack art, lack soul, but are a technically impressive achievement. The kind of thing you’d use as a forum avatar but never hang on your wall. A novelty.

A few years back Hannah commissioned a hand drawn piece of art of her, me, and the cats for a present. It’s framed and in my office on a shelf I can see from my desk. It brings me joy every time I look over and see our old condo and the time, attention, and care that went into the creation of it. Not just from the artist, but from Hannah in working with the artist to craft something very much us.

I tossed some images to GPT and it generated a Ghibli version of us. It wasn’t bad. It butchered the Funkos that were in the background pretty badly. But it lacked character. And I didn’t like how it rendered my body. I then asked it to create something in the style of Bill Watterson. That it balked at. Told me it couldn’t do it. Why it would take instructions to copy/steal from Ghibli but not the famed Calvin and Hobbes artist, is … odd? But at the end of the day it’s an LLM, and if you can describe it without using the magic words it’ll still give you want you want:

Again, not bad. Kinda fun? A passing resemblance to the style. But it feels more like a paint by numbers template used by a caricaturist with less style. And it lacks any of the punch, the actual artistic flourish and genius of Bill Watterson’s art.

And the Funkos are still hilariously rendered.