I’ve been obsessed with AI image technology lately. Between creating visuals for articles and experimenting with new tools in my spare time, I’ve found myself wondering about a crucial question: can today’s AI detectors actually identify images from the newest generation models?
This isn’t just academic curiosity. As someone who works with digital content daily, I need to understand the capabilities and limitations of these technologies. The rapid advancement of AI image generators has completely transformed what’s possible in digital creation — and simultaneously raised serious questions about authenticity and verification.
When I discovered Undetectable AI had released their image detector around the same time OpenAI launched their groundbreaking 4o image generation model, I knew I had to put them head-to-head. The timing felt perfect for a real-world test of detection versus generation capabilities.
What is Undetectable AI?
At its core, Undetectable AI is an AI humanizer, a special subset of paraphrasing tools that can turn any text into something undetectable by common AI detection tools. This works by changing the tone of your input, rearranging it, removing repetition and common AI words, and adding a few intentional mistakes.
But we’re not here to talk about that. You see, Undetectable AI has lots of new features, but the one I’m most interested in is their AI image detector — and that will be our focus today.
What is 4o Image Generation?
OpenAI’s newest AI model is meant to replace their DALL-E model. Released in March 2025, GPT-4o’s newest offering is the most impressive AI image model I’ve seen to date. It understands context perfectly and writes text in images on a human level — something I’ve never seen before.
Undetectable AI’s Image Detector vs. 4o Image Model
So, all this has me wondering: I have a new AI image detector and a new AI image model at hand. What happens when I test them against each other?
Test #1
Undetectable AI: Correctly identified image as machine-generated.
AI Likelihood Score: 100%
Test #2
Undetectable AI: Wrongly identified image as authentic.
AI Likelihood Score: 0%
Test #3
Undetectable AI: Wrongly identified image as authentic.
AI Likelihood Score: 0%
Test #4
Undetectable AI: Wrongly identified image as authentic.
AI Likelihood Score: 0%
Test #5
Undetectable AI: Correctly identified image as machine-generated.
AI Likelihood Score: 100%
Test #6
Undetectable AI: Correctly identified image as machine-generated.
AI Likelihood Score: 100%
Test #7
Undetectable AI: Wrongly identified image as authentic.
AI Likelihood Score: 0%
Test #8
Undetectable AI: Wrongly identified image as authentic.
AI Likelihood Score: 0%
Test #9
Undetectable AI: Correctly identified image as machine-generated.
AI Likelihood Score: 100%
Test #10
Undetectable AI: Wrongly identified image as authentic.
AI Likelihood Score: 0%
Overall Score
|
AI Likelihood Score (Correctness) |
|
The Bottom Line
Well, this is awkward — Undetectable AI’s image detector only caught 4 out of 10 AI-generated images. That’s a 40% success rate, which isn’t exactly inspiring confidence.
I had high hopes for this detector, especially considering how well Undetectable AI’s humanizer works. But detecting AI-generated images is clearly a different beast altogether. The 4o model seems to be creating images that fly completely under the radar half the time.
But I just want to emphasize this: Undetectable AI’s image detector is still in beta. Meaning that this is the worst it’ll ever be. Considering how good 4o is, 40% is actually pretty good.
Does this mean 4o has cracked the code? Not necessarily. My testing was limited, and there are many variables at play — image complexity, subject matter, and even the detector’s own learning curve.
The reality? We’re still in the early days of this technology. Today’s undetectable image might be tomorrow’s obvious fake. As someone who’s been tracking this space closely, I’d recommend a healthy dose of skepticism toward both AI-generated images and the tools claiming to detect them.
What’s your experience with AI image detection? Have you tried these tools yourself? I’d love to hear your thoughts in the comments below.
