Creating AI Art for Your Blog When Real Images Don’t Exist

Last Updated on April 10, 2025 by Candice Landau

I don’t use ChatGPT to write my articles but I have, of late, used it to create images for my blog.

I try not to do this when a real image will suffice—something a human has created and that I can support and credit—but every now and then, the image I need doesn’t exist, as in my latest post, Unmasked, a poem about diving with death, which incorporates death as a character. That’s when I turn to AI. I do also say when I create an image using AI. You’ll find this in the image caption. This is not negotiable to me and I hope it helps keep the trust.

Until recently, I couldn’t really do this. The images ChatGPT created often had extra fingers, odd angles, or were just factually incorrect. I remember asking it once to create an image of a whale shark. No matter what I did or said, it could not get the tail right. It kept illustrating a whale tail rather than a shark tail. It was immensely frustrating!

Before I used ChatGPT for image creation, I had a subscription to Midjourney. Integrated into Discord, it was easy to use and great fun. It made much better images than ChatGPT, though like ChatGPT it really struggled with scuba divers, only seeming to understand them as commercial divers with hard hats. I paid $20 a month to use it. When ChatGPT started launching better algorithms, I cut my Midjourney subscription and went instead for an OpenAI subscription, also $20 a month. I haven’t looked back (or tried Midjourney again), even though, until recently, ChatGPT was infinitely worse for image generation.

As of March 25, 2025, things have changed. OpenAI announced its new image generator, built into GPT-4o. Having tested it now for a week, I can tell you it is so much better than all the iterations that came before it. It creates actual text you can read. It understands prompts better. It allows you to iterate on images without completely changing them.

To put it to the test, I decided to upload poetry that required images I wouldn’t be able to find online, or create myself without being a graphic artist whiz.

Here are examples of some of my articles that use AI images. In each case, a real image wasn’t available on Unsplash, Pexels, or one of the other sites I regularly use to support artists trying to make a name for themself. That said, in a couple of cases, appropriate images likely were available to purchase via Shutterstock but I’m not a business that can spend $50 on an image or a subscription to these corporate-oriented stock photo sites and so, unfortunately, AI wins again.

Articles featuring images generated using ChatGPT:

Unmasked, a Poem About Diving with Death featuring the grim reaper and a Pacific Northwest diver.
This Geode, Us, a Poem About New Intimacy featuring a geode in spherical form, cracked open with crystal colors mentioned in my imagery.
Unanticipated, a Poem About the Death of Self featuring the grim reaper in purgatory-like lands.
A Visitor, a Poem About Making a Decision featuring a window, rain and an impression of wind, all with a gloomy atmosphere.

To be clear, none of these images came out right the first time. This is because you really have to have a great prompt in order to create good images.

So, iterate!

Okay, I hear you. Thing is, there’s a catch. It takes time to generate each image, sometimes a minute, sometimes longer. OpenAI also caps the number of images you can generate within a given time frame and freezes you out if you’re putting in too many requests. It’s much better to do the thinking first and take the time to write it all out so you can focus on perfecting a good image rather than repairing an awful image.

Based on what I’ve learned, here’s my hard-earned advice on how you can get images out the box.

How to get ChatGPT to create good images for your blog, faster

Though you’re unlikely to get the image you want the first time around, you’re going to have a better chance of it, if you incorporate all the things I mention below in ONE prompt.

Get clear about the style you want. Do you want a photo realistic image or an image that looks like an acrylic painting? Do you want a Pixar-style graphic or something that looks like an artist’s pencil sketch?
Mention how the image will be used/the size/orientation you want it in. For example, I tell the tool that I want it to create a horizontal image for my blog. I could give it dimensions but I don’t really bother as WordPress is good enough at resizing and I’m also a dab hand at screenshotting!
Mention the mood you want to set. Some of my images are pretty dark or ominous. Sometimes I want to be sure this comes across, other times I don’t. Be sure to tell it exactly what mood the image should evoke.
Mention what you don’t want it to include. One of my poems evokes imagery of a crocodile dragging a person down and drowning them. When I first asked ChatGPT to create an image for it, it was horrifying! Very gory. I decided in the end to just use an artist’s image of a crocodile. That poem was far too dark to get a real depiction.
Give it details. Paint a picture. One of the things I realized quickly was that unless you give very specific details about the things you’re imagining, it’s unlikely to include them. For example, when I asked it to create an image for my poem, Unmasked, it initially showed a close-up of a skeleton touching a male diver. Both the diver and the skeleton reaper were kneeling on the sand (not what I imagined when I wrote the poem). It was also quite unappealing and didn’t capture the feeling of being a diver in the Pacific Northwest, specifically. I wanted to evoke those dark waters, and an aura of gloominess. When I asked it to pull back—have the diver holding a line and kitted out in a drysuit, as well as a more vague depiction of a grim reaper with kelp and plumose anemones to set the scene—it created a much better version.
Frame the image. Is the image a close-up? Is it pulled back? Does it have a focal point?
Tell it whether or not you want writing in the picture. It’s much better at incorporating copy now, but I prefer not to have too much copy on my images, not least because I’m picky about fonts. If I want it I can add it in Canva or Photoshop or Procreate if handwritten is desired.

Here’s an example of a prompt I wrote to generate the image for this poem:

“Create a horizontal photo realistic image for my blog for the following poem. The image should be a scene in an underground subway like the British Underground of a train approaching. I don’t’ want to see details on any of the peoples’ faces—I want it to have an anonymous feel. I also want a departure board to show another train will be arriving in a couple of minutes, and another after that. The poem should show how people rush to catch a train even though there’s another on the way, always. The train is symbolic of entering a relationship. I want to see the whole scene: the train, the platform, the announcement board, the person/people rushing. I don’t want you to specifically mention the British Underground but I want it to have that feel. Here’s the poem:

TRAIN APPROACHING

You told me
These days people rush
Into everything.

I agreed because
I agree, but
It took a moment
Even a few days because
My mind spoke over me
Mind the gap
Doors closing
Train departing
And I dashed to get on board.

Why this frenzy
When the system is sound?
There is always another train
On the way.”

This was the first image it generated. It took about a minute:

I have to say, this was a much better image than if I’d just told it to create an image for my poem. However, it still needed refinement. Here’s what I told it to do:

“I don’t want it to have the title of the poem on the image. The only words should be the minutes. I also don’t love the photo realistic image as much as I thought I would, could you make it feel slightly dreamier, a bit more illustrated but still in a realistic style?“

The resulting image was perfect for what I wanted, well, minus the person standing on the track. Errr, not sure why it felt that was necessary. Perhaps I should have mentioned that? It still does strange things.

Here’s how I refined it:

“Okay, almost there. But I don’t want someone standing on the tracks!“

Final image? Almost exactly what I wanted. Though I didn’t ask it to get rid of the illustrations on the underground wall. Sometimes you can’t account for these things. That said, close enough! The poem is now published here.

A template for creating images using ChatGPT

Because you got to the very end of this article, I’m going to reward you with the template you can copy paste and use to generate amazing AI images:

Create a [horizontal/vertical/square] image in a [photo realistic, painterly, illustrated] style. This image will be used for a [blog post/Instagram post/8.5 x 11 poster] and should evoke a [dreamy, ominous, peaceful, surreal] tone. The scene should depict [describe the setting, characters, and any key objects or elements]. Please [show the entire scene, use a close-up, pull back the perspective]. Do not include [anything you want excluded]. The image should feel [any other descriptors].

Let me know if this works for you and if you had to change it in any way. Also, please know that NONE of this article was written using ChatGPT.

PS: For those of you interested in seeing what sort of image it creates when I give it no guidance, see below. The image is a technically pretty good image, with some minor issues, but it isn’t appropriate for a featured image for the blog post. It also has some quirks. Note the extra apostrophe added at the end of the poem. And, of course, I would have put “Mind the Gap” parallel to the gap. Honestly, I wouldn’t have put any other words around the poem if I’d wanted it written out like this: