Finding a relevant featured image is a step in every content marketers blog publishing process. Choosing the right one is critical, it’s what people see when an article is shared and ultimately what determines whether or not it captures attention.
Over the past few years AI image generation for featured images in WordPress has become a very common approach, but it’s easy to tell when AI was used. Images usually have subtle imperfections, oddities, and don’t quite feel right. There are ways to avoid this with the right prompt and the right LLM.
I’ve tried a bunch of them, so I decided to put them to the test and see how they all performed generating a featured image for a recently published blog post.
Here’s the prompt I’m using:
create a realistic HDR high res photo of the white house sized 1060x380 pixels - as realistic as possible and professional photography
And the results…
Google Gemini: Imagen 4
Very nice, all around. Good focus, feels real, and provides pretty much exactly what I would be after: a classic photo of the white house.
ChatGPT: GPT-4o
This is what we end up with using ChatGPT. The photo quality and accuracy is good, but there’s one obvious flaw: we’re not visually centered. It’s not difficult to correct, but worth noting as we’re comparing all the options. It’s also a little too dramatic on the shadows and feels unreal, to be honest these are the images that commonly stand out to me as definitely AI created.
Midjourney
Two examples are provided here, because they’re quite different and the second is very good.
Option 1
A little overly dramatic, with a lot of creative liberty taken. These are classic traits of midjourney, it’s very good at generating art. That said, sometimes the results turn out to be just what you’re looking for.
Option 2
A classic trait of Midjourney, they’ve taken a lot of creative liberty here but in this case the end result is nice depending on the specific topic of the post. This looks more like what a high-end photography might provide after digital filtering.
Ideogram
This one is one of the best in my opinion, good lighting and not overly dramatic. This may be the one I would have gone with if I were publishing the post.
Adobe FireFly
Technically it checks all the boxes, but there’s something about it that’s a little off. I think the dark levels and there’s too much contrast going on. It feels more like a late 90s photo. Still very good and accurate though.
Canva
Canva’s image generation AI is one I haven’t honestly used, but I’ve included it as an example because I’ve heard it mentioned a few times. It’s not bad, but I can’t say much more on it’s overall quality and use. This image has a bit of an unrealistic aspect to it, of those listed here I think it’s unfortunately the easiest to spot as AI generated. This is not to say that Canva isn’t a good option to explore, again this is the only time I’ve used it.
Conclusion
Overall they’re all good, and I suspect they’ll get even better. Determining the image to use is still a human aspect of the process that I haven’t been able to replicate with AI. If I provide an AI chat prompt with a request like:
provide me with a prompt to generate a relevant featured image for the following blog post: {pasted blog post here}
I have very rarely received a quality image as a result. There have been a few times, but it’s rare. Maybe this will get better, but I think there will always be a human aspect to this, even if it’s moderation and review.