The Lines Are Blurring – How Good Are You At Spotting Real vs AI Generated Images?

With Generative AI tools getting better with every model iteration, it’s getting harder and harder to spot the real stuff from the AI generated content.

Take The Quiz – How Many Did You Spot?

Here’s a simple test ** you can take to see if you can spot the real photos from the AI generated ones…this one is quite easy in comparison, so leave us a comment and let us know how you scored!

** The quiz will be live for a couple of weeks

What The Best Generative AI Tools Can (Already) Do

I did a post recently, comparing some of the best Text to Image Generation tools to see how they stack up against each other. Save for the continued struggle rendering human hands, the results from some of the better tools and models are getting pretty darn good (see samples below).

The interesting thing is that much of the exponential improvements have happened over less than 6 months…here you can watch how Midjourney model output has become nearly photo realistic since Nov 2022.

Text to Video (and other cross-modal conversions) are also developing fast. Companies at the cutting edge (like Metaphysic) can already create some pretty compelling deepfakes…and now have the ability to do this on the fly (below you can watch the demo at TED this year).

So What’s Next? Generative AI Goes Mainstream.

You’re probably wondering what comes next. Save for filters embedded in popular social media apps, most of the Generative AI Image/Video tools out there today still only cater to the early adopters and early majority…but not for long.

Adobe has really accelerated access and ease of use by embedding Generative AI capabilities into Firefly…and now the Beta version of Photoshop. To understand just how easy it is to use these tools, you can watch a quick demo below…I was able to create some pretty compelling images with a very basic knowledge of Photoshop. Normally, this would have taken expert level knowledge and hours of work.

Why Does This Matter?

Generative AI promises some pretty amazing applications (creating new kinds of art, designing innovative products and generally letting our imaginations run wild). But it also poses some challenges, especially when it comes to telling the difference between real and fake images. As generative AI models evolve (and as you’ve hopefully seen from the examples above), it is going to be incredibly hard to spot the subtle clues that reveal their artificial nature.

So why does it matter if we can tell real from fake images? Well, for one thing, it can affect our trust and credibility in the information we consume online. If we can’t be sure that an image is authentic, we might be misled by false or manipulated content. For another thing, it can affect our creativity and originality. Some creative professionals argue that if we rely too much on generative AI to create images for us, we may lose the human touch and the personal expression that actually make art meaningful.

What We Can Do About It

Here are some possible ideas that may make the identification of AI images easier in the future:

• Develop standards and guidelines for labeling and disclosing AI-generated images (e.g. Meta data, Watermarks, etc). This can help us know the source and purpose of the images we see online, and to make informed decisions about how to use them.
• Create tools and techniques for detecting and verifying AI-generated images. This may help check the authenticity and quality of the images we encounter online, and to spot any signs of manipulation or alteration.
• Educate and raise awareness about generative AI and its implications. This could help us understand how generative AI works and what it can do, and to appreciate the ethical and social issues involved in its use.

I suspect standards and regulation are not far behind, with the EU perhaps leading the charge like they did with privacy and GDPR. Tool providers will also likely need to adopt certain guidelines with regards to labeling and disclosure.

In the travel industry, given the lower bar to doctoring content, it’s likely that we may lean even more into review sites and user generated content, provided they’ve proven trustworthy and continue to do so through robust moderation and community guidelines.

We certainly live in interesting times…and it’s going to get even more interesting in the months and years ahead.

PS – in case you were wondering, I created this article’s lead image (Mona Lisa on the streets of New York, holding her own painting) using a combination of Midjourney and Photoshop.