Mind-blowing AI in Galaxy phones makes fake photos look dangerously real - but why?

But the fun doesn’t stop there… You can also add real-life objects to photos by sketching a rough outline over your own (or someone else’s) pictures. The best part? The results are shockingly (!) realistic. The scary part? The results are… shockingly (!) realistic.

Sketch to Image: The pros and cons of Samsung’s unbelievable generative AI depend on how you use it

Before I move on to breaking down what excites and what freaks me out about Samsung’s new generative AI, let’s take a second to acknowledge how insanely real these “photos” (shared by X users) look…

Take the photos with “pets” in them. While the one with the guy holding an AI-generated cat at the mall could do with better lighting and shadows (on the cat), I’d bet money most people wouldn’t be able to tell he’s holding a nonexistent cat. And that’s if you think to check the photo for AI in the first place.

However, the other two cat images (eyepatch cat and posing cat) look absolutely indistinguishable from legit photos (to me).

Sure, the fact that (for whatever reason) a cat is wearing an eyepatch is a whole different mystery to be solved, but that’s not the point here.



Apart from the UFO prank (unless you do believe in aliens!), the rest of the Sketch to Image edits are also super realistic. Particularly the one with the fake “Y-A” print on the guy’s shirt, and the freakish “boat on fire” (courtesy of Michael Fisher), which is where Samsung’s AI gets pretty… scary.

Note: I added a little bit of artificial noise to the “burning boat” image just to prove a point. While the original edit looked a bit “too perfect”, the added noise easily makes the imperfections much harder to see.

And again, this is a high-res image you can zoom in and examine closely. But what if it was a low-res screenshot!

The good: Samsung’s Sketch to Image is the most fun and (potentially) most productive new AI feature on Android

And getting to the not-so-great part about Sketch to Image, as it stands, it seems like Samsung isn’t able to pull off the whole sketching thing using only the power of your Galaxy phone. In other words, your photos are sent to the cloud where more powerful algorithms and computers take care of the heavy lifting, which is becoming common practice nowadays.

Apart from the usual security concerns surrounding “cloud computing”, the other problem is that this process makes “Sketch to Image” rather slow. In fact, generating a single image, or altering a photo can take 20-30 seconds, which (in today’s society) is a long time to wait for just about anything - let alone to see your new picture.

Generative AI like “Sketch to Image” can boost the creativity of your projects while making you less creative

As with all of the AI that “helps” us get things done, I strongly believe Sketch to Image might turn out to be another creativity roadblock for humans. And that’s ironic considering generative AI is supposed to make you come across as “more creative”. However the key phrase here is “come across”.

Just like I can open my calculator at any time and work out how much is 86 - 77 x 5, not having to force my brain to think, I can also whip out my Galaxy Fold and let it fill in the blanks on my super rough sketch. And never learn how to draw - even if I wanted to.

Samsung’s magical Sketch to Image generative AI can be pretty… sketchy

Finally, it almost goes without saying… The super-powerful image alterations you can make using tools like Sketch to Image and the Magic Eraser, can end up being abused by some people. And they probably will.

For the record, the AI-generated images come with metadata and an “AI-generated” sticker in the bottom left corner - to let you know they are… AI.

But it’s not like you can’t crop out the caption; screenshot the image; and lose the metadata before trying to convince somebody this is a real photograph. And who knows what people will choose to use this for…

Perhaps adding or removing yourself from a photo can convince your partner you didn’t cheat on them! Or you can convince somebody you’re holding their dog hostage - but really, you’re holding an AI dog that looks like their dog.

Even more capable generative AI on phones brings up the same old questions: What does it mean “to take a photo” in 2024?

Finally, it’s only fitting to ask a question that’s become more and more relevant as generative AI keeps evolving… What is a photograph? What does it mean to “take a photo”?

This was a question I started asking myself all the way back when Instagram filters were cool. But Google and Samsung’s recent AI is far more sophisticated, letting you remove objects from photos; add objects; move them, expand a photo, and with Sketch to Image, add an original object to your picture.

In other words, even if my intentions are sweet and pure, and I simply want to enhance my photos and online presence by putting myself on a yacht, or putting myself on a yacht with Ariana Grande, or getting my subscribers to believe I own a pet alligator… Are those photos real?

I mean, they were “real photos”, and the person in them (a.k.a. me) is real… But is creating a different reality, which I’m a part of, a “photograph”, or just a painting of some sort? Or am I… AI Picasso?!