10 Comments
User's avatar
Maryn McKenna's avatar

I've been a subscriber for at least one year, maybe two, and have found your recommendations interesting and thoughtful. But I'm really dismayed to see you uncritically using and implicitly recommending the "Studio Ghibli style" that OpenAI has introduced.

They of course have not licensed this from Ghibli.

It's against the entire ethos that Miyazaki championed, and it's theft.

Expand full comment
Shift Happens (Steph Peters)'s avatar

Yep I’ve been writing with Chat for about 2 months now and it’s been wonderful my first amazing novella eye of the beholder is coming out shortly and you can see on my posts that many have images and most from chat

Like this one…

Actually check out the image from book 2 post… was going to share it hère…

First chapter of book 1 right here:

https://open.substack.com/pub/shifthapens/p/eye-of-the-beholder-3ab?r=b8pvb&utm_medium=ios

Expand full comment
James Francis's avatar

As a cartoonist, I don't understand how you see the above as a collaborative tool. It's clearly a replacement. No skilled cartoonist needs such a service. But people who don't want to pay for cartoonists or bother to learn the basics of cartooning do...

Be honest: would you ever use ChatGPT's cartoon output and then still hire a cartoonist to realize the final product? I suspect the answer is no. So, why represent it as such in your newsletter, as experimenting? At least be honest to your readers.

Expand full comment
Jeremy Caplan's avatar

Hi James, this depends on what size organization we're talking about. A mid-size or large org might well experiment with GPT to rough out ideas before handing the concept over to a professional to meet their style, specifications and quality standards. I agree with you that solo writers aren't likely to have the budget to hire professional cartoonists for an individual newsletter or blog post, so if they use this, it would be as a substitute for having no cartoon/image at all.

Expand full comment
Michelle Silbernagel's avatar

I've been playing around with ChatGPT's infographic maker and found the same issue you mentioned in the newsletter about it changing the graphic when you ask it to make a small correction. I was wondering if you had any success taking the PNG it created for you and uploading it to some other image editor to fix or change it to be the way you want it. My final iterations still have typos and words cut off. After about three hours of playing around, I have a lovely graphic with misspellings and words cut off. I still want to use it, but I'm not sure what to do. I have a month of Canva Pro for free and am wondering if I might be able to figure something out there -- but I have rudimentary skills at best in Canva. 😂

Expand full comment
Jeremy Caplan's avatar

Canva has a bunch of new AI capabilities, as you've probably seen, Michelle, so it's getting more flexible and powerful at helping you generate and even revise images with AI. As for editing ChatGPT's AI-generated images, at present it's tricky, as you noted, to get a precise result when you have a very specific, detailed infographic in mind, especially if it has many concepts or distinct elements in it. I expect that will improve over time, as will the editing functionality. I've found that if I try multiple iterations, it's generally able to fix the cut-off text and typos, but it does take some patience. This is still early going for this new model.

Expand full comment
Muhlyssa's avatar

Would it be possible for you to share the prompts you used for some of the images in the article?

Expand full comment
Jeremy Caplan's avatar

Hi Muhlyssa, for the prompt to picture concept image — the flowchart — I started with "Create a wide image of a Prompt-to-Picture Concept. It's a visually engaging flowchart that helpfully and vividly illustrates the AI generation process—from a simple idea (a laptop with plants growing out of it, representing creativity) branching out into unexpected, delightful variations." That stemmed (no pun intended) from a prior chat thread I had been working on, thinking through how to convey in simple visual terms the way that ChatGPT 4o enables anyone to start with a concept and turn it into an initial visual idea followed by multiple visual ideas. After a few additional iterations and requests for further adjustments and improvements, I ended up with the Prompt-to-Picture flowchart you see in the published post. I hope that helps!

Expand full comment
Benny Haas's avatar

I'm surprised you don't mention Midjourney and Stable Diffusion here for image generation? 🤔

Also, I thought Dall-E and chatGPT shared the same 'AI engine' behind the scenes...

So will I get the same results from Dall-E soon?

Expand full comment
Jeremy Caplan's avatar

Hi Benny, those are great platforms/tools as well. There are so many out there with distinct advantages and limitations. I simply don't have enough space or bandwidth in individual posts to mention all of them, especially if I want to mention specifics or get into nuanced differences, etc.. And as the number of AI tools now numbers into the thousands, my capacity to keep up with all of them is challenged.

On the other point, Dall-E 3 was the old engine powering ChatGPT's image generation, but that has now been supplanted by the new 4o image generation engine, which is vastly improved.

Expand full comment