I considered upgrading to Gemini Pro, so I asked Gemini to convince me (it did not work)

Phone Arena

21-05-2025

You've probably noticed that Google I/O happened yesterday. I mean, half the news today seem to be centered around AI, but that's no wonder. Just like last year, Gemini took center stage at Google's annual event.
So, of course, I got hyped. I mean: I work here. Tech is what I live and breathe, so I wanted to try all of it. I opened up Gemini on my PC, and I navigated to the Upgrade panel. I really wanted to just subscribe and forget, but experience has taught me to always ask to try before I buy.
And you won't believe what happened next.
Image by Google So, imagine my surprise when I said to Gemini's Brainstormer Gem:
The AI model, of course, told me that this is not possible. But I'm brave, so I just started doing whatever I wanted. To my shock: Gemini complied.
As some might know, I'm into art and design. One of my most-used apps is Photoshop, but I'm always looking for a replacement, so that I can be set free from Adobe's subscription model, which isn't quite my tempo.
Recently, my social feeds became flooded with examples of people utilizing Gemini specifically to edit images. The results I saw were downright impressive. I'm sharing something similar below, so you can get a better idea, but there are loads of examples out there, especially in shortform.
And, of course, this is the first thing I wanted to try:
Instead of a response, however, the model outright generated an image:
Fine then! And it even continued, saying (and I quote) "And here's the same image, but with the box open:"
Something is not quite right here.
You only get three demo attempts per type of request, so I had to abandon my entertainment. But this meant I could try something else instead.
So, I've got this photo, which has always been a favorite of mine. I've not been able to track down the author, so if you know who made this, please let me know. Here it is:
I just love Castlevania, man. And when I was a kid, this picture felt like the summary of everything the franchise stood for in terms of aesthetic. So, honestly, I would totally rock an HD version of this image as my wallpaper.
(Please play Mozart's Requiem for creative effect)
Again: not quite what I was looking for. The original contents of the photo were absolutely abandoned. Gemini seems to have grasped the pseudo-demonic mood of the image, but in truth: that is just a feeling you get. The actual image is just a sunset over the clouds, with a church and some trees. If you ask me, I'd say this should have been the focus.
One of the issues I've ran into when using AI, time and again, was the model's failure to understand nuance and depth. Everything is surface level. And no matter how many breakthroughs we've made over the years, this hasn't really changed. Yes, Gemini and ChatGPT can generate more detailed images, and they've gotten better at understanding what we want from them.
They have not, however, gotten any closer to understanding what we're looking to get out of it. To me, that would be the real breakthrough. But the thought of that happening is also kind of scary, so let's get back to funny stuff.
I do know the creator of this lovely image, though:
Sentinels by Jenna Barton Jenna Barton is hands down one of my favorite artists. Her creations are so moody, and meaning-rich. I've lost myself in many of them, so in retrospect, I wish I didn't attempt to ruin one of my favorites with AI for this experiment. Sorry, Jenna, I just couldn't think of a better example.
I mean, my Google Photos app has such a feature, which is confirmed to be powered by AI. And, honestly, it does a great job most of the time. Ergo, this should be a breeze, right? Well, what the AI did seems a bit backwards:
So, instead of editing the existing image for me, the model just generated a similar one. And let's be honest: it is not similar in any meaningful way.
Another one of my personal favorites is Mike Mignola. If the name rings a bell, you might know the guy as the creator of Hellboy. I adore the series, so if you like comics, myths and folklore: make sure to give reading the series a shot, it's very good.
The image above is one of my favorite works from Mike. It's just such a vibe. So I wanted to see if Gemini can generate something similar for me.
This isn't it. Yes, some aspects are there: the surface-level ones. The rest of the prompt, the parts related to the mood and style of the image, were completely disregarded for one reason or another. Can't lie, though, I'd totally use this for a character sheet in D&D.
But I had reached the end of my test run, so I had to stop. On the one hand, I was glad that Gemini was unable to replicate the art styles of my favorite artists. But that wasn't really the intent: the goal here was to see if it really understands substance any better. I think it's safe to say: not yet.
No. I've got the receipts, but going through an article, filled with screenshots of the chat itself, does not sound fun to read. Plus, like I said at the beginning: I was truly excited to try out all of these new things.
I also asked Gemini to let me try generating audio and video, but it gently turned me down. I did keep on asking clarifying questions, though, and as it turned out: not all advertised Gemini Pro features are available in every territory, so make sure to check before you sub. At least, the AI model seems pretty honest about it.
My biggest gripe with Gemini is that regardless of subscription, you cannot export stuff to Docs or Sheets. I have, in fact, successfully tricked Gemini into exporting results in a sheet for me, but the file was pretty much broken. In terms of Docs however: it just denies doing that. And this sounds like a pretty basic feature to implement, right?
I have to give credit where it is due. Despite the fun above, Gemini has gotten way better at a lot of useful, harmless things. It's getting great at helping you brainstorm, it's fantastic at providing feedback on writing (no, we don't use AI here, relax). And it has, indeed, helped me figure out how to do stuff that seemed scary-complex at the start, like modding my first Seiko NH35 wristwatch.
Plus, as Gemini itself said to me at a point:
So I might just have to repeat my experiment again down the line. In case it remains unclear: I did not subscribe... yet. I do feel that this is an imminent thing that will happen in a few years or so. AI is here to stay and it can be very useful, if not utilized with malicious intent. And it's only going to get better, so you all better get a lot less malicious! But until that time comes, I'm looking forward to trying Android XR and Google's Glass successor, the announcement put a really wide grin on my face. How about you, what are you excited to try in the coming months?