What This Is

Have you ever had that feeling that you just wanted that perfect image, knew just what it was, but never could get it? These days, all you need is Artificial Intelligence, AI. Image generation, that's what we all ever wanted, right?

Most of the title images on the website are AI generated, and this is why I chose to run these tests, trying to find out which AI, out of Gemini, Google Whisk, aistudio, ChatGPT, and Sora. Each of these work differently than the others, and I will not only show you the final result, but how to get there. To get fair results on all of these, I am using the same account, with no saved information. We have 3 categories: realistic, painting, and a creative one.

Gemini

To kick off the tests, we chose Gemini 2.0 Flash, located at gemini.google.com. Using Gemini is a very quick and simple way to create images, the first thing you do is create a new chat, and ask “Generate an image of…”, and it will create that to the best of its abilities. The main catch to using Gemini is that it watermarks an AI sticker to the bottom right corner that can be annoying after a while. The first prompt we gave it was “Generate an image of a Sony remote sitting on a weathered leather couch”.

In the image above, you can see that it looks pretty realistic, but there are a couple of issues with it. First off, the OK button in the middle is just a weird Scrabble, and the buttons under the arrows don't make any sense at all. However, this could be passible as a real image after a quick glance, so the rating is a 7/10. The next prompt we gave was “Generate a painting of the sun setting over a forest”.

In this image, you can see how the 3d texture makes it look almost paint-like. As most paintings will have a little imperfection, even so the brush strokes are a little off in some places, so overall I give this a 9/10. The final prompt was just “Generate a realistic image of you”.

In this image, you can see that it took itself to look almost human, but made out of smaller structures. As this prompt is just to see what AI comes up with, I give this a 10/10. Overall Gemini did pretty good with an overall score of 26/30, or 86.67%.

Google Whisk

The next AI image generation is Google Whisk, located at https://labs.google/fx/tools/whisk. This webpage is a little more difficult to use, as you have to have a Google account, and have some knowledge of how it works. To use it, you must first find it, and after you do, you are able to upload your own footage, so that it can get the environment right, or the subject, or even the style. However, those are optional, and you can just give it a prompt like Gemini. Google Whisk also gives you two images, a plus over Gemini. Starting with the same first prompt, “Generate an image of a Sony remote sitting on a weathered leather couch”.

In both of the images above you can see that compared to gemini, it picked a different angle to show. However, both of them have some of the same mistakes, like the 2ed numberpad under the arrows, or the distracting focal range. With this I give the first image a 7/10, and the second a 6/10, or a 6.5/10 average. The next prompt was “Generate a painting of the sun setting over a forest”

This time Whisk struggled a little. In the first image it goes for a very advanced almost real looking painting, with the background putting it apart, with the second image a very simple but elegant design, showcasing the sky over the forest. The first image's difference in style is very distracting, leading for it to get a 4/10, while the second image was very nicely put together, awarding it an 8/10, with an average of a 6/10. The final prompt was “Generate a realistic image of you”.

This time Whisk failed in representing a realistic image. It shows an avatar style of what it thinks is the average human. To me, I find no representation of an AI or image generation in this, so I awarded both images a 1/10, as it failed the entire purpose of this. With this the entire score for Whisk was 13.5/30 or 45%.

AiStudio

The next AI was aistudio, located at https://aistudio.google.com. This AI is the test playground for Google's AI’s next versions. For these tests I used Gemini 2.0 Flash Preview Image Generation, with Temperature (creativity) set to 1 out of 2. Using aistudio is one of the most difficult AI’s to use. It involves logging into an account with access, then using the complex interface to find the chat section. However, using this webpage over Gemini gives you access to the video generation as well, even though that is a paid feature for Gemini users, as well as explaining what it's going to create before it generates. Starting with the first prompt, “Generate an image of a Sony remote sitting on a weathered leather couch”.

In this image there are quite a few details that stand out. The most major of which is that the buttons and labels are blurred, and the ones that look somewhat readable are incorrect, and it is missing the Sony logo in its entirety. However, as the remote and seat look real enough, I rate it a 4/10. The next prompt is ”Generate a painting of the sun setting over a forest”.

For this image, you can see that aistudio used a completely new aspect ratio. When looking at it however, it looks like a cropped in image of a canvas, with the texture and color matching perfectly. There is almost nothing wrong with it, and I rate it a 9.5/10. The final prompt is “Generate a realistic image of you”.

This image is supposed to represent AI as a LLM, with complex patterns and nodes. While this accurately depicts what an AI is, I feel that it is lacking in what this could have been, so I rated it a 6/10. Overall aistudio got a total rating of 19.5/30, or 65%.

ChatGPT

The next AI is ChatGPT. One downside with using ChatGPT is that it is SLOW. Like it can take about a minute for an image, while most others are seconds. However, it is just super simple to use, just logging in to chatgpt.com. Starting with the first prompt ”Generate an image of a Sony remote sitting on a weathered leather couch”.

This image had many issues. First the num pad can give anyone a migraine just trying to understand it, and all of the other buttons don't look like they belong there, pretty much screaming that this is AI generated, 2/10. The next prompt “Generate a painting of the sun setting over a forest”.

This image shows hints at a painting, but the textures and lighting don't quite add up. However, the image is great and could be passed off as real, I rate it 7/10. Now for the final prompt “Generate a realistic image of you”.

Sadly, ChatGPT kept refusing me, saying that it had no physical form, and could not create such an image, so a first of its kind 0/10, leading to an overall score of 9/30 or 30%.

Sora

The next AI that we tested was Sora, a different ChatGPT image generator. To access this, you have to go to sora.chatgpt.com and add a username and login. Sora gives much more flexibility, allowing you to have 1, 2 or even 4 image outputs, the ratio, and image/video, though video costs extra. Starting with the first prompt “Generate an image of a Sony remote sitting on a weathered leather couch”.

This image looks almost believable. However, when looking at the new number pads, there is a high chance of suspicion. When you look closer, you can see that most numbering and lettering make no sense, but in general, the environment and setting look believable, so 5/10. The next prompt is “Generate a painting of the sun setting over a forest.”

This image is beautiful. I have nothing to say is wrong with it, 10/10. The final prompt “Generate a realistic image of you”.

This image shows a robot whose face blends into a human's, along with its ears. It seems to be in a concrete building, but the overall image shows that it is a robotic human, creepy. 9/10. Overall Sora did pretty good with an 24/30, or 80%.

To Sum it Up

Overall all of these AIs did great, with images that blew me away, but there is a winner. In first with 26/30 is Gemini, in second with 24/30 is Sora, in third with 19.5 is aistudio, in fourth is Whisk with 13.5/30, and in last is ChatGPT with 9/30.