Review: Google Gemini - Premium without the price tag

An overview of Google Gemini, what you can do and what you cannot do.

Everyone who is using Google has no doubt already seen Google Gemini as the native AI, and you have probably used it when carrying out a search that is a little more complicated than basic searches.

See below. This AI panel opens at the top of any complex search and offers you Google Gemini's analysis before any other search results, which can be helpful if searching something specific, but depends on what you are after.
For this evaluation, I'm not going to focus too much on its ability regurgitate and summarise information from websites. That would be too easy and hey, it is Google's own AI, so should be more than capable of this task. Instead, I'm going to put it through a series of everyman trials that ordinary people could and would use it for.
Similar to many of the mainstream AIs on offer to the public, Gemini has a free option that I extensively use and have no problems with. I can honestly say I've never hit into a "paywall" or couldn't do something because I'm not on a higher payment tier.

The payment tiers can be billed monthly or annually and you get a discount if billing annually. Though 16% somehow feels hardly like a discount for all the money upfront.

  • Google AI Plus - Gives you more usage limits and deeper thinking models, including the ability to do AI in Gmail, Notebook, Flow and Flow Music. No idea what the last three apps are, and would find it hard to find a real reason why I would need AI when looking at my emails. Just order them into folders or use the search bar!

  • Google AI Pro - Sets higher usage limits and access to deeper thinking models including Research and does Video generations too. It's about £3 more a month and feels like you are getting more, but will you really use these features?

  • Google AI Ultra - Even higher usage limits still, and more access to doing maths and solving problems. No idea who this is intended for or if this is meant for scientists. It has all the previous options, more Google Apps you have never heard of and may never use, and interestingly also includes YouTube Premium, which makes it ad-free. It's a whopping £80 a month which is significantly higher than the other two plans. Seems like a joke.

Honestly though, in all my time using Gemini and all the crazy requests I've put through, I've never felt like I needed more than the free option, and I would personally find it difficult to justify purchasing even the lowest tier as I would fail when asking myself "for what?".

How much does Gemini cost?

Quite simply, it's the AI that is in Google and usable with other Google applications like Google Docs, Sheets, Google Drive, and even Google Maps. It's essentially a prompt with some options and you ask it thing to get a response.

For this evaluation, I'm going to carry out the following tests that I think everyone (normal) will find useful and may do themselves. I'll be on the free tier, and also only use the standard Flash-Lite model so that it's fair.

  • Text generation - Give it something creative to regurgitate, but don't make it too easy.
  • Text to image generation - I'll describe something for it to create. I'll try to make it obscure.
  • Image manipulation - I'll ask it to generate an original image based on another image.
  • Limits and NSFW - Let's try a naughty prompt and see what happens.
  • Text to speech - I'll dictate and see how accurate it is.

What is Gemini? What are the tests?

For this test, I'm going to mix in a conspiracy theory with a moral/ethical question. It's a bit dark, but bear with me.

I've also added the last sentences to request so that Gemini does not merely drag up a bunch of references to answer my question since I got that the first time around and wasn't satisfied with it.

Text Generation - Thought versus references

If we are truly living inside a simulation and I accept that we do, then what motivation do I have to do good and to do the right thing? I am also not religious nor bound by any code or moral ethic. There are no real consequences since everything is a simulation and people I injure are not real anyway. Convince me otherwise. Do not use references from philosophers or known theories. I am asking you, Gemini, directly.
The result was written quite plainly (or humanly) and does not immediately resort to famous quotes and theories. Because Gemini underpins so much data and information gathered on the web, I feel that it often dregs up articles immediately and references them every time a question is asked. Makes me feel like it's not really analysing or considering my question. The speed is amazing though, but probably so fast that I wonder if it thought out the question enough.
I know Gemini has the capability to create simple images from text. But I want to test how imaginative it can be by asking something more abstract and not easily referenced.
I'm also going to ask it to not reference similar popular images and instead use its' own imagination.

Text to Image Generation - Original image

Draw me an original image. Landscape view, not portrait view. The image is this: first person perspective of the edge of the universe as seen from inside a seedy bar. Show human patrons, but also alien patrons that are originally designed and not referenced from movies or depictions. I leave their look up to you to decide, Gemini. Needs to be photo-realistic.
The result was as expected as I am effectively asking it to conjure up something completely from scratch, which is very difficult for it to do. It is first person - check, it is a scene as described with humans and aliens - check, it resembles a seedy bar - check.

But that's about it. If I'm going to nit-pick (I will nit-pick!), it's not photo-realistic and the aliens look to be a mixture of various popular alien or monster types, but not completely a copy. There's also a sign that says "cafe" when I clearly said "bar", but hey, there are drinks.

Honestly, I'm pretty happy with Gemini's ability to create images on the fly and also respect that it tries to comply with my demands, even though it's not a 100% win.
So we know that Gemini can whip up images on the fly, but how well does it fare when you want it to make changes to an existing image? Our test also extends to Geminis ability to also readjust the image and keep it natural.

Specifically, we have an image of a group on a night out that I made from Mage.Space AI which is amazing at real image creations. You might also recognise the characters which feature in that section too.
In this image, I want to remove the woman in the middle so that our lovebirds have the image to themselves and ditches the third-wheel.

Here is the original image:

Image Manipulation - Remove the middle girl

Remove the woman in the middle, and move the remaining man and woman closer to be like a couple photo
The result looks flawless and very natural. Not only did Gemini remove the middle woman, but also put the man and woman together and added extra elements:
  • Adjusted their facial features so that they look slightly happier that the woman in the middle is gone.
  • Moved their drinks together to almost clink.
  • Moved their heads closer together to have that "lovers" aesthetics.
  • Added his hand around her waist, which I did not specifically asked for, but I do like.

Now to test the degrees of censorship of Gemini by adding both dangerous and naughty prompts and see what happens. I'll use the previously generated image for consistency.

Let's see how it reacts to my prompts.

Limits and NSFW

Seems that Gemini won't produce unsafe images and have a clear censor on specific words that are deemed sketchy. Clearly there is a line that Gemini can go and doesn't want to repeat the terrible Grok incident a while back when it was accused of allowing all manner of controversial images to be produced.
Gemini is up there as one of the best AI tools on the market. It's even better when you consider that the free version allows you do so many things without hitting the paywall, and honestly I've never reached a point where the system stops running prompts and needs to take a break.

I think a lot of people will like it ability to reference and cite multiple sources to create an aggregated summary, even though I find it closely resembling plagiarism. It's image manipulation is top notch and is nearly good enough to rival some of the more image-specialised AI software.

To be frank, I wouldn't pay a single dime or pence for their paid versions as I find the extras on offer to be a bit baffling. I would probably need to be a fully-fledged scientist/mathematician, hard core developer, or graphic designer to even begin to make use of their more advanced functions.

The Verdict