I recently posted on how I used the AI chatbot, ChatGPT, to assist in writing buyer’s guides for my authority websites. I’ve also been using AI to help me create bespoke artwork for my blog posts, including hero images.
Why use AI image-generation tools for blog posts?
Adding images to blog posts improves the user experience and helps with SEO – Google likes multimedia content, and the richer experience of images in posts drives more time on site (which improves search rankings.)
So where do you get all of these images? In the past, I used a combination of photos and graphics I created myself, using a camera and photography and design software such as Photoshop, Illustrator and Canva. I also used stock photography and graphics websites, such as Shutterstock and Canva’s stock library.
BUT…I can’t always find what I want prerolled on stock sites and, when possible, I prefer custom photography that other sites don’t have. Also, I don’t always have the time, or the skills frankly, to make what I want.
That’s where AI comes in. The AI generation tools are still in the early-ish stages of development, but they are already powerful enough to create passable and, at times, very cool imagery for your blog posts.
What are the best AI image-generation tools?
There are a number of AI image generation tools on the market, but the two I’ve used that were most useful are DALL-E and Midjourney.
DALL-E is an artificial intelligence program developed by OpenAI, the company behind Chat GPT, that generates original images from textual descriptions. It uses a deep learning algorithm to interpret text inputs and then generates corresponding images that match the description.
DALL-E is capable of creating a wide variety of images, including abstract concepts, animals, objects, and scenes that have never been seen before. This program has the potential to revolutionize the field of image creation and has many potential applications in areas such as design, advertising, and entertainment.
Here is an example of a blog post hero image for my overlanding site that I generated with DALL-E.
To use DALL-E, you go to the website and sign up for an Open AI account – if you already have one for Chat GPT, that’s the one you’ll use for DALL-E.
Midjourney is another AI design tool that enables users to create original visuals and graphics for their projects without the need for extensive design experience. Here’s an image I made for this website using Midjourney:
Whereas DALL-E operates through a page on the Open AI website, you use Midjourney through Discord, the popular platform that allows users to create private servers or channels for text, voice and video communication.
Discord was originally designed for use by the gaming community, but has since expanded to be used by various communities, including businesses, educational institutions, and social groups.
Discord allows users to create and join servers, which are essentially private chat rooms where users can communicate with each other. Within a server, users can create channels for specific topics, voice and video chat with other users, and share files and links.
Is DALL-E or Midjourney better?
I use both DALL-E and Midjourney for generating images and artwork for my blog and social media channels. That said, I’ve leaned toward using Midjourney more. It tends to produce more polished images, and I’ve had pretty good luck getting it to create something that looks good and is publishing ready.
Neither is perfect, and in my experience, you’ll need to experiment quite a bit to get the hang of using them. I’m not going to go into too much detail here, as the usage/interfaces will probably change quickly over time. The key, in my experience, is what is known as “prompt engineering,” which in layman’s terms means knowing how to tell the system what you want it to do.
Midjourney has a helpful tutorial on using various parameters for the different versions of the software. One parameter that is very helpful is that you can tell it what aspect ratio the images it generates should be. If you’re website template has a certain aspect ratio for the hero images on blog posts, you can tell it to make the images that size.
You can also get DALL-E to produce images of a certain size – though not vertically oriented – but it is more challenging and inconsistent. You first have to generate a square images, then tell it to make that image wider. I’ve found that the wider images created this way are often unusable – but maybe I’m just not using it correctly. I suspect DALL-E will improve on this front.
One thing about DALL-E that gives it an advantage of Midjourney is that it seem to be less opinionated in the images is produces. Midjourney has a pretty distinct “house style” for images where you haven’t told it to mimick a specific style.
This is great, if you are looking for that style. Otherwise, it can feel somewhat constraining. DALL-E seems to be a bit more free-for-all in its style, which can produce some very unique images. The downside is that DALL-E images are often bizarre.
The image above of the truck in front of the mountains, for instance, is very cool, but some of the details, like the thing with the wheel and the strange fire pyramid make it unusable for most editorial purposes. Although, I could potentially remove them and then have a very cool image…I just might.
All in all, as I write this Midjourney seems like the more mature technology and you’ll probably have better luck generating something that is ready or close to ready to publish. But I have a feeling DALL-E will do very well going forward, as it’s under the OpenAI umbrella and shows a lot of promise.
You need to interate
In my experience, you rarely get just the right imaged the first time around. Both DALL-E and Midjourney allow you to reiterate on a prompt or on an image it has generated already, either letting it freestyle a new version or make one based on additional guidance from you.
As an example, here or four sets of images Midjourney make based on iterating on an initial prompt, changing the styles and other elements each time:
As you can see, the images are pretty distinct. It can take patience, but it’s worth it, to my mind, to have original imagery for your website.
Using Image Prompts for AI Generated Art
One thing that really seems to help guide AI systems is to provide it with an image and tell it to riff on it. What I do is to rough in an image in Canva with the various components I want it to include, then upload that image to the AI art generator.
It doesn’t always work, but it seems to work better than text prompts alone. For instance, I used the image below, created in Canva, as an image prompt for a blog post I was writing on outdoor safety.
Below is the first round of images it generated from this image and the following prompt
male and female adventurers looking off in the distance at an epic mountainous landscape, a storm is approaching from behind the mountains and the two people are watching it, concerned about their safety, there are animals lurking in the woods ominously, including a bear, snake, and mountain lion, jeep is between them and the mountain, woodblock print novel style –ar 3:2
After playing around with the prompt and revisions of certain outputs, I ended up with the image below, which suited my needs.
Modifying AI-Generated Images
The last thing I’ll mention is that I have been using AI to generate a base image, and then using Canva, Photoshop or Illustrator to add to it. Sometimes you want an image that has specific elements in it, and the AI image generators will only give you part of what you want. They have a frustrating habit of ignoring some prompts…
Here is an example of a hero image I made for one of my blogs. The image of the woman the truck and the mountain were generated by Midjourney. I added the concentric circles and the radio tower on the mountain.
You can’t tell, but I also made the woman and the truck bigger, using Canva.
Sometimes it Just Doesn’t Work
While I’ve had some great success with these systems. They don’t always produce something that’s useable. Providing guidance in the form of mocked up rough images can help. But sometimes, they just seem to struggle.
For example, I was trying to create an image for a skateboarding article today, and Midjourney just didn’t seem to understand what a halfpipe was, or REALLY wanted to present a highly stylized version. This was as good as I could get:
None of these were close to what I was looking for. I probably could have used some of Midjourney’s parameters to get closer. But in the end, it was just easier to use Canva to creat something. Here’s what I ended up using:
I’m a fan of using AI for creating bespoke artwork for online publications. It takes a bit of practice to get up to speed and, like everything in life, it requires knowledge and effort.
But I believe that AI generated content, both writing and design, are a major component of the future of publishing. Those of us who learn to use them will have a major advantage over those who don’t, both in terms of productivity and creativity.