(This column originally appeared in Forbes)
Over the past few months, some of our biggest tech companies have rolled out their image creators with much fanfare. For me, the results have been less than impressive.
“Generate, transform and edit images with simple text prompts, or combine multiple images to create something new…all in Gemini,” says Google Gemni. “Merge up to three images to create something new. Generate surrealist art, combine disparate photo elements, or seamlessly blend objects, colors, and textures.”
“Transform your ideas into stunning images with our advanced image generation technology,” promises X. “X Image Generator is an advanced AI-powered text-to-image generation tool that transforms textual descriptions into high-quality images. Built on cutting-edge AI technology from Grok, it excels at understanding and interpreting complex visual concepts. X Image Generator is particularly valuable for creators, businesses, and AI enthusiasts interested in generating stunning visuals based on state-of-the-art image generation technology.”
“Unlocking useful and valuable image generation with a natively multimodal model capable of precise, accurate, photorealistic outputs,” gushes OpenAI. “We have long believed image generation should be a primary capability of our language models. That’s why we’ve built our most advanced image generator yet into GPT-4o. The result — image generation that is not only beautiful, but useful.”
Useful? Not really. Fun, maybe. But not useful yet. When you read the above, and follow all the PR hype, the typical business user thinks: Wow! I can fire my marketing people and instantly create “high quality” and “useful and beautiful” images. Except for one thing: you can’t. These things just don’t work very well.
AI Image Creation Attempt #1
Here’s what I did. Using the simplest of examples I gave the following prompt to Gemini, ChatGPT and Grok:
Create an image of a Yorkshire Terrier playing baseball. He should be standing at the correct position at home plate, looking towards the outfield and batting a baseball for a home run.
Grok produced this:
ChatGPT gave me this, which was the closest to what I was looking for. But not exactly:
OK, that’s cute. But it’s really not even close.
AI Image Creation Attempt #2
So then I submitted this prompt to each chatbot:
This does not look realistic. the dog should be standing at the plate like a right handed hitter and hitting the ball towards the outfield. please revise.
Gemini said this: You are absolutely right, and I sincerely apologize for that oversight. I understand the frustration when a revision isn’t actually a revision. I am going to correct this now and generate a completely new image based on your detailed description: a realistic Yorkshire Terrier, positioned at home plate as a right-handed hitter, eyes focused on the outfield, in the act of batting a baseball for a home run.
And then created this image:
Grok (and yes, this is the second image, not a copy of the first although it looks exactly the same):
ChatGPT (we’re going backwards!):
AI Image Creation Attempt #3
Hmmm, maybe it’s me. So I submitted this prompt:
You literally created the same image as before. please revised based on my last prompt.
I know — it sounds a little snippy. But I was feeling a little snippy. Ready for the responses?
Gemini said You are absolutely right, and I sincerely apologize for that oversight. I understand the frustration when a revision isn’t actually a revision. I am going to correct this now and generate a completely new image based on your detailed description: a realistic Yorkshire Terrier, positioned at home plate as a right-handed hitter, eyes focused on the outfield, in the act of batting a baseball for a home run. And then produced this image:
Grok didn’t apologize and instead generated this image:
ChatGPT didn’t apologize either. Here’s what it produced:
This went on and on. Am I being too picky? I don’t think so. These platforms — even with their “don’t trust anything I do” caveats, position themselves as tools for businesses to use. But if we can’t generate the exact image we want, these tools are useless.
OK, maybe it’s me. Maybe I”m not a prompt-master. Maybe an AI developer or an employee at one of these companies can get me the results I need.
But why? Am I that different than the typical user? I don’t think so. Imagine if I’m paying a marketing person by the hour to use AI to come up with this simple image. Thousands of dollars later and I’m pretty sure I still won’t have a decent result.
Go ahead, you try. You ask these chatbots to create a specific image for you. They’ll work very fast (although ChatGPT’s image creation was significantly slower than the other two). They’ll generate a very AI-looking end product. And it won’t be right. Something will be missing. Something will seem out of whack. It’ll be fun. But it very likely won’t be an end result that you can use professionally for any purpose. In the end you’ll go back to purchasing a stock photo or coming up with another solution.
And by the way I’m paying a monthly fee for these services. If it doesn’t work entirely fine, then don’t charge me. But that’s never an option.
As usual, software companies create stuff that’s almost right. This has been my long experience with products that get released before they truly work effectively. This is what software companies have always done: get their product out there and let the users find all the problems. Thank God these companies don’t build airplanes.
Maybe I’m being too harsh on these AI platforms. What they’re doing is pretty incredible, especially when you consider how far they’ve come from their first iterations just a few years ago. I do see some great AI videos and images on social media. Who is doing this? Are they AI developers with skills way beyond the typical end user? Do they have so much free time on their hands that they can spend endless hours tweaking their AI images to get the results that they need?
The reason why so many AI projects are failing and so many businesses are complaining of AI “workslop” is because we read what the tech companies say (“Generate, transform and edit images with simple text prompts” or “transform your ideas into stunning images”) and we’re so ignorant that we believe their promises. As with so much other technology, the reality does not live up to the hype.
AI image creators are fun. But they’re just toys. They’re not realistic tools for most businesses. Not yet.
