Tuesday evening. Claude limit hit. The week barely started.
This happens occasionally. I push it too hard on a project, and suddenly I'm staring at a usage wall. Usually I just wait it out. This time, out of mild desperation and the fact that I'm already paying Google for cloud storage anyway, I thought: why not try Gemini Pro?
The promise is genuinely attractive. Video generation, a frontier model, multimodal everything. Google has more compute than most countries have GDP. How bad could it be?
Fairly bad, as it turns out.
Why I Actually Gave It a Fair Shot
I didn't come in looking to hate it. I had real work to do — I wanted to create an Instagram carousel based on my article about selling on Vinted. Practical task. Clear output. The kind of thing Claude handles in a single back-and-forth.
Step one: ask Gemini to generate a prompt structure for an 8-slide carousel. That part worked. Clean output, reasonable structure. Promising start.
Step two: ask it to generate the actual content for each of those 8 prompts. What came back was one summarised graphic combining all eight ideas into a single image. Still, visually okay. So I asked the obvious next thing — can you now split and crop this into 8 separate images?
This is where the cracks appeared.
The Moment It Referred Me to Canva
Gemini tried. It produced another slightly-different summarised graphic. I clarified what I needed in more detail. It tried again. Different graphic, same fundamental misunderstanding.
Then it suggested I use Canva or Figma to do it manually.
I had to re-read that sentence twice.
An AI assistant, in 2025, solving a content generation task by offloading the work back to me with a tool recommendation. If I wanted to open Canva and do it manually, I wouldn't have opened an AI.
At this point, Gemini went from "promising trial" to "one-star review in my head." If Claude is a five-star hotel with a spa and a personal butler, Gemini is a party hostel in Southeast Asia that's technically in the same price bracket but has you sharing a bathroom with six strangers and figuring out your own breakfast.
Context window? Felt like talking to someone who forgets the previous sentence while you're still in it. The further I pushed, the more the outputs drifted from the original brief — not refined, just increasingly distant.
Okay, What About Video?
Fair point — Gemini's video generation is supposed to be a strength. So I pivoted. Let's try creating a Reels-style video output.
Same structural problem. Gemini works well one prompt at a time. If you want a cohesive, multi-part sequence, you're either accepting a single compressed 8-second clip, or you're doing it slide by slide, manually sequencing everything yourself. There's no conceptual guidance, no narrative thread being held across the outputs the way Claude holds context across a long conversation.
The one thing that might be genuinely useful: Gemini can add dynamic text overlays and generate voiceovers. That's real. But then the dynamic text hallucinated words mid-execution. Instead of "sold" it rendered something that looked like "soyd." On a content piece I'd be publishing. For a brand.
I mean, come on.
The real revolution isn't the AI. It's not needing to become an expert just to use it.
What This Is Actually About
Someone will read this and say: "You just need to learn how to prompt Gemini properly. There are whole courses on this."
And that's exactly the problem.
The reason Claude changed how I work isn't that it's marginally better at following instructions. It's that I never had to become a Gemini specialist to get value out of it. The whole point — the actual paradigm shift — is that the AI meets you where you are. You shouldn't need a certification to split eight slides.
Every major leap in consumer tech over the last 25 years was a leap in convenience for the non-expert. The iPhone didn't ask you to learn Linux. Spotify didn't ask you to understand audio compression. Vibe coding, Claude-style AI collaboration — the revolution is that the barrier moves, not that you get better at climbing it.
Gemini, right now, still requires you to climb.
Maybe that changes. Google has the resources to close this gap, and I won't pretend the race is over. But as of today, paying for Gemini Pro feels like seeing a Photoshop ad in 2003 and receiving MS Paint.
What I'd Actually Do
- Don't cancel Claude because you hit the limit on a Tuesday. Wait it out. The gap in quality is worth the patience.
- If you need Gemini for something specific — video voiceovers with dynamic text have potential, but QA everything before it goes live. Hallucinated words are not a minor bug when it's your brand.
- Use Gemini as a complementary tool, not a replacement. If you find a workflow where Claude handles strategy and Gemini handles specific generative outputs, that might be worth exploring — but it requires setup effort upfront.
- Don't let the hype cycle set your expectations. Benchmark claims are real. Practical usability for non-specialists is a different thing entirely.
I'll keep the Gemini tab open. But I'm not renewing that subscription just yet.
P.S: Image was created by Gemini Pro...so don't take all words to seriously :)!