In November, I tested the image generation capabilities within Google’s Gemini, which was powered by the Imagin 3 model. When I liked it, I ran very fast within its limits. Google recently – my successor – Imagin 4 – and I have been passing it at my pace in the past two weeks.
I think the new version is definitely an improvement, as I am thankful for some of my problems with Imagin 3. But some frustrations are still there, that is, the new version is not as good as I want.
How often do you make pictures with AI?
93 votes
This is a daily thing for me.
11 %
Maybe once every week.
15 %
A few times a month.
31 %
Never
43 %
So, what better?
The quality of the pictures has generally improved, though the improvement is not large. Imagin 3 was usually good at making pictures of people, animals and scenes, but the new version produces permanently fast, more detailed photos.
When it comes to producing images of people-which is possible only with Gemini Advanced-I had permanent problems with Imagin 3 where it will produce cartoon photos, even when I was not asking for this particular style. It was often a losing war to change it into a more realistic thing. I have not experienced any of these with Imagin 4. All the pictures of the people that arise from it look very professional – probably a little too much, which will make us touch later.
With the old model, I had limited control over the proportion of the biggest frustration aspects. I often trapped with 1: 1 square images, which strictly restricted their use. I could not use them for online posts, and printing them for a standard photo frame was out of question.
Although the Imagin 4 is still pre -defaulting 1: 1 ratio, but now I can only indicate it to use it only a different, such as 16: 9, 9:16, or 4: 3. This is the feature I am waiting for, as it makes the images far more versatile and useful.
Imagin 4 also works very easily. Although I haven’t found this quite faster – though a sharp model is allegedly at work – it has very few mistakes. With the previous version, Gemini sometimes showed an error message, saying he could not produce a picture for an unknown reason. I have not found any of the 4 people in the image. It just works.
Still look a bit retuched
Although the Imagin 4 produces better images, is more reliable, and allows the proportion of different aspects, the problems I have faced when testing its predecessor are still there.
My main problem is that images are often not as realistic as I want, especially when producing close parts of people and animals. The images are completely saturated, and many people have a significant impact on Bokia that fades the background professionally. They all feel as if they were taken by a photographer with a 15 -year experience, instead they point to the camera on my cat and press the shutter.
Certainly, they look good, but a “comfortable format” will increase a wonderful increase. I indicated Gemini to make an image more realistic by eliminating the Bokia effect and making it less perfect. The AI tried, but after pointing it three or four times on the same image, it seems that it reaches its limit and said that there can be nothing better. Every new image made by it was a bit more comfortable, but it was still quite polished, clearly indicating that it is AI-seatored.
You can see that in the pictures above, leaving left to right. The first one involves a strong bobkia effect, and the skin of the person is very clear, while the other two develop with the developing person and get more tired. Even in the last image, he started a little room. When it was not really real when giving Gemini to make the image more realistic, it is more comfortable.
Imagin 4 works greatly with random images such as scenes and city scyles. These images, which have been taken from afar, do not include such close details, so they look more real. Still, it can be hit or miss. A picture of the Sydney Opera House looks very good, though the saturation has been slightly collided-the garbage is extra green, and the water picture is perfect blue. But when I asked for the image of the Grand Valley, it looks completely artificial and does not fool anyone to think that this is a real picture. However, he performed better after some efforts.
Editing is better, but not enough
One of my grips with the previous version was its clumsy edit. When asked to change a minor thing – like a hat color – AI will do it, but it will also produce a very new, completely different icon. The ideal scenario has to be made an image and then every detail will be allowed to edit, such as changing a piece of clothing, adding a particular item, or changing the weather conditions while leaving everything exactly.
In this regard, the image 4 is better, but not more. When I indicated it turned the jacket to blue, he made a new picture. However, by saying to keep all the other details, in particular, it managed to maintain a lot of scenes and themes from the original. This is what happened in the above examples. In the third icon, the woman was the same, and she appeared in the same room, but her camera and the camera angle was different, making it more than amending the shoot.
Here is another example of a cat’s popul meal. I urged Gemini to change the color of Papal, and he did, and he had a lot of details. The cat is the same, and so is the background of the same. But the cat’s ears are now sticking, and the hat is a bit different. Still, a good effort.
In spite of its shortcomings, the Imagin 4 is an excellent toll
Even despite a long list of its problems and lost functionality, Imagin 4 is still among the best AI image generators available. Most of the problems I have mentioned are also present in AI image generation software, so it is not as if the Gemini is behind the competition. It seems that this type of tools seem to be important technical barriers that need to be overcome before reaching the next level of precision and realism.
Other limits still exist, such as failing to make famous people images or produce content that violates Google’s guidelines. Whether it is good or bad, it is a matter of opinion. For users who receive low restrictions, there are alternatives like Grook.
Have you tried the latest image generation in Gemini? Let me know your thoughts in the comments.
