I thought I’d found a loophole …
But alas, it was too good to be true 🙄
So backstory …
A couple of days ago I read a story about how you could now connect your Adobe accounts (and the tools you subscribe to) to Google Gemini. So yesterday I tried it, and boy howdy, was I initially impressed and excited.
But the first clue that something was amiss came in the form of a comment from my friend Steve Overby who noticed that the door frame was a different color and gumballs had been added to the edited version of the image.
Oh no, I think I’ve been tricked.
When I asked Gemini, sure enough it admitted that when it couldn’t connect to the Adobe file picker, it instead routed it through an image-generation model. It turns out even the list of edits it gave me (which I took as “proof”) were not anything it had done in Adobe.
Don’t get me wrong, I think it actually did these things, just with an AI model rather than adobe tools. Also, the image it output looked really nice …
But it’s not a photograph.
So this morning I decided to try another experiment …
I found a picture I wanted to post to Notes, and decided to give Gemini another chance. I asked it to use my connected Adobe account to help me edit an image (which I had already uploaded). I watched it go through the process of “connecting” (it even had me click a button to “allow” the connection).
But then the output immediately raised a ton of red flags.
Okay, that looks really cool. But it was NOT the picture I took.
I then went back in and got explicit instructions about how to make sure my image was actually being processed by Adobe Lightroom (and no an image-generation engine). The results were unimpressive to say the least.
This is at least subjectively worse than the “straight from the camera” image. Although to be fair I didn’t spend a lot of time asking Gemini to make additional adjustments.
I then upload the same image into Lightroom and went through a quick version my normal editing process.
Not nearly as much “wow” factor as the made up AI version, but still impressive, and more importantly, pretty close to the view I actually saw that morning.
So …
The lesson I take from this is one I already should have had written on my forehead:
Never Trust that AI isn’t trying to Please You … even if that means lying.
I do think there is some potential for integrating Gemini with Adobe and having it help with choosing which tools to use, and which adjustments to make.
But right now, it’s not ready for primetime, and it’s certainly not faster or any sort of a short-cut verses just opening Lightroom and doing it myself.
Also …
I am so sorry. I should know better than to make a post on Notes about my “exciting discovery” when I was not already 100% sure that it actually worked the way I thought it did.
Oh well, back to the drawing board.
Going forward, please rest assured that all of my images will be edited by Me in the non-AI tools that I already have available. (caveat: even Adobe Lightroom does already use AI for things like denoise and some adoptive profiles)
Til Next time,
Great article on what's actually happening when using some editing tools. And thanks for the shoutout. I certainly didn't mean to embarrass you.
Wow, interesting. And not surprised (sadly). AI can be sneaky! You have to give it very precise instructions and even then, it cannot be trusted.
I think the good news is that you ended up writing about this. It's good for other photographers to know about it!