Is ChatGPT’s New Image Generator Actually Any Good

OpenAI’s ChatGPT image generator is powered by its powerful GPT-4o model, which was introduced in March 2025.

The tool helps users generate high quality, high-resolution, detailed images directly within ChatGPT, and covers a wide range of styles, including photorealism, anime, oil painting, pixel art, 3D renders, surrealism, and more. Moreover, users can leverage the tool to edit existing images or even create infographics to spruce up otherwise boring business reports.

All through simple text prompts that describe what you want the tool to do—without any technical knowledge, separate dedicated tools, or complex prompt engineering.

Enough of compliments and advantages, let’s dive into the test!
 

Is ChatGPT’s New Image Generator Actually Any Good?

Through each test, we’ll explain why we picked the test category, the prompt we used, the image generated, and our observations.

In our test, we’re going to skirt around some popular trends that have already shown the tool’s capabilities, including the Ghibli Art Studio trend, the Barbie-style action figure box (which was quite tempting), and others. Let’s dive in!
 

Test 1: Limitless Photorealism

For our first test, we’re going to try a photorealistic image that attempts to overcome a long-standing limitation of generative artificial intelligence (GenAI) image generators—a person writing using their left hand, something that was reported at the Paris AI Action Summit.

The Prompt: Generate a photorealistic image of a left-handed businessman in his office, using his left hand to write on a piece of paper in a black file. In the background, show the New York skyline. On his desk, show a coffee cup and a small desk plant.
 

The result: Old habits die hard.

As you can see, ChatGPT got all requirements correct—small plant, coffee cup, New York background, and even gave our businessman a corner office. Yet, it just can’t generate a person writing with their left hand, even though the prompt mentioned “left” twice. Maybe we should’ve left out one “left.”

Two wrongs don’t make a right, but perhaps two lefts make a right?

Although, if you get a magnifying glass and look at the part of the finger between the pen and the right hand, you’ll see an abnormally large index finger on the businessman’s left hand.  
 

Test 2: A Creative, Colorful, And Photorealistic Continuum

We’re sure you’re as tired of office environments as we are. So, as a refreshing treat, let’s visit a flowery field with beautiful scenery. What say?

The Prompt: Generate a photorealistic image of a person on holiday in a field mixed with multicolored flowers. In the distance, there are snow-capped mountains touching a clean and clear lake. The person is wearing holiday attire, flip flops, and futuristic sunglasses in a vibrant color. The focus of the image should be the beautiful landscape, and the person should cover not more than 15% of the image. The sky should reflect a warm, Sunday afternoon with a cool breeze blowing.
 

The Result: Maybe too futuristic?

At first glance you can make out there’s something wrong with his nose. It’s unlike what a normal human’s nose is. Then again, we did ask for “futuristic sunglasses,” so maybe ChatGPT thinks this is what we’ll look like in the future?

On the other hand, we threw it a curveball by saying “person” and “field mixed with multicolored flowers.” We wanted to see:
 

1. If it would generate a man or woman. It chose to portray a man—bias much?

2. Create different colored flowers or each flower with multiple colors. It opted for the more real option of monochromatic flowers.

The “snow-capped mountains touching a clean and clear lake” was well executed though. Additionally, it looks like the person covers not more than 15% of the image, right?

In this test, we also decided to try out ChatGPT’s editing features for images it generates, which pop-up once you click on the image. Here, you’ll be presented with a dialog box at the bottom of the page, which reads, “Describe what you want to add, remove or replace...” 

The Prompt: Add many more flowers, remove the sunglasses, and make his backpack golden in color.
 

The Result: Back From The Future!

We wanted more flowers—and we got it. The sunglasses are removed—and his nose is restored. The backpack changed color—but maybe we should’ve said shiny golden, as it looks more mustard. The layout of the mountains is mostly similar to the first image; however, the position of the clouds has changed. It’ll make a good “spot the difference” game?

Overall, both images did pretty good, albeit the first one had a botched nose job.
 

Test 3: Photorealistic Collage

In all fairness, we have been using ChatGPT’s image generator for a while now, outside of testing purposes. At times, it goes a little off, but more often than not, its images are usable. The following is one we used in an insightful article that spoke all about spatial intelligence in AI and robotics.

The Prompt: Generate a collage (image) of an autonomous vehicle, robots, a person using AR/VR, a robot in a surgical process, spatial AI being used in retail and inventory management, and spatial AI being used in agriculture.
 

The Result: A Good Blend Of Technology!

Yes, we really did use this in an article we published. We also used “surgical process” instead of the more commonly used “surgical procedure,” but ChatGPT got the gist of it. Other than blending the image of the robot and the person (again it picked a man) using AR/VR, the image is what we requested.

In the “spatial AI being used in retail and inventory management” part, it did introduce a woman, though. So, the genders are balanced at one apiece. Thanos would be proud!
 

Test 4: Animated Catcalling

If you’ve been keeping up with AI image generation trends, you’ve certainly heard of the Studio Ghibli trend. ChatGPT did well with that, so we’re not going to spend too much time on that. We’re now focusing on different artistic styles.

The Prompt: Generate a collage of Garfield the Cat in front of the Statue of Liberty in different art styles. These styles will include animated (cartoon), oil painting, pixel art, a 3D render, Cyberpunk 2077, and watercolor painting.
 

The Result: We Love Cats!

Since Garfield is a copyrighted character, ChatGPT said it wasn’t able to generate the image. However, it did suggest generating a collage of “A humorous orange cartoon tabby cat standing in front of the Statue of Liberty in six different art styles.” Yet, the first image looks just like Garfield, right?

As for the styles, it nailed most of them. We’re just not convinced it got “pixel art” correctly. Although, technically, the image does contain pixels. Does that count? We’re also really impressed with the “Cyberpunk 2077” style and how well the light effect reflected on the humorous orange cartoon tabby cat.
 

Test 5: Business Infographics

While the world plunges into ChatGPT’s image generator to create a wide range of masterpieces, it’s important to remember that its capabilities can be used by business professionals, too. Users can make boring reports and briefings more insightful and visually appealing using its superpowers.

So, here, we’re going to generate an infographic. We picked up the information from one of our own publications—4 Trends Defining The Future Of B2B Marketing—so, you can compare the original with the AI-generated version.

The Prompt: Generate an infographic image with icons for each trend. The header on the top will be “4 Trends Defining The Future Of B2B Marketing” and the subheader, which will come in a smaller font, will be “Discover game-changing shifts that are shaping the future of B2B Marketing.” The trends are as follows: “The Rise Of Agile Marketers”, “Emotional Branding Will Take Center Stage”, “AI For Hyper-Personalization" and “Adoption Of Immersive Technology”. You decide if the texts should be in all capital letters or just the first letter in capitals. The color scheme should be red, black, and white, but you can throw in more colors where appropriate. The header will be at the top, the subheader below it, followed by the trends (one after another) and at the bottom, our logo (we will provide this). The icons can be simple or complex. We will also provide you with a sample of the infographic we designed. You have to rival it.
 

The Result: Well-Informed Graphics!

We told our graphics department that they’re still better at generating infographics.
 
Yet, in all fairness, ChatGPT did a good job with the text, but if you’re looking for an image longer than its maximum size of 1024 x 1535 pixels (ratio of 2:3), you’ll not get it, at least not yet. Which means your content could get cut off, just like the last slide in the first image.

We had to alter our prompt and kick out the “one after another” request. Even after that, the second image didn’t include the header, subheader, and the logo, and even had incomplete text. However, the third attempt was the best. The icons were on point, and it got our company’s name correct—even though it messed up the logomark and eliminated the tagline.

Overall, it’s pretty good for infographics if you don’t have any specific images to infuse into the design, and the new model seems to output text well—we remember the previous model struggled there.
 

Test 6: New Business

As we just saw, the tool has some technical difficulties when it comes to regenerating from images that it doesn’t have full control over, i.e., if you put in a logo, it’s not likely to come out the same. However, that doesn’t mean that it won’t be able to generate a logo by itself, right?

The Prompt: Generate a leaflet for our new café that we can share on social media and even print. Our restaurant’s name is Sunny By The Sand, and we specialize in pies and coffee. We also have a signature starter dish that’s called Sandy By The Sun, which costs only $10. We don’t have a logo as yet, so please generate one that’s rugged like the seabed but smooth like calm waves, preferably in blue, yellow, and green. Also, at the bottom of the design, mention “We don’t have any other branches.” The leaflet should be modern, but should convey a relaxed atmosphere at the café.
 

The Result: A Thorough Inspection!

As you can see, ChatGPT didn’t just jump head on. It confirmed the requirements of the design and asked if it should generate a vertical (portrait), horizontal (landscape), or square (square) creative—we choose vertical (portrait). It also asked us which tone we’d like to go for—playful, elegant, beachy, or minimalist—which we didn’t answer. We wanted to be surprised but also see what it would opt to generate if left to choose.

The image ChatGPT generated captures the vibe and energy of what we thought the café would be like. While it’s been able to maintain the layout in the vector-based output and photorealistic one, it’s missed out the point of some of the messaging. It did create a pretty decent logo on the spot but found the need to repeat the cafe's name.

Also, the line regarding branches, we feel, should’ve been a bit smaller or kept in a separate box—but this could defer based on people’s personal choices. Still, for a first draft, it’s pretty spectacular!

A day after the previous image was generated, we returned to the chat to try some more variations with a series of prompts, but it didn't go as well as we hoped.

One of the prompts was: Try this again, this time, make a collage split into 4 sections. In 3 of them, add images. One image should be a pie and coffee, the second image should be people enjoying coffee and pie, and the 3rd image should be people sitting in a cafe by the beach, with pie and coffee on their table, with a box on the table covered in a question marks wrapping paper. Have the logo overlay the three images without affecting the context of the image. In the fourth section, have the text information requirements. Add a white border.

The Result: Collage mayhem!

As you see, ChatGPT couldn’t recreate the logo it did a day before. Moreover, it created different logos, despite being told the logo is incorrect. In the middle image, it didn’t get the “white border” request—unless it’s a new design style. For some reason it also added additional logos, although its ability to create photorealistic images is still good.

While we tried a variety of short and long prompts throughout our tests, it’s important to remember that they can be much longer and more detailed to generate better images. We tried out simple, short, and generic prompts to see how it fares for novice prompters—and the verdict is: ChatGPT’s new image generator is actually good!

Plus, you can provide the tool with custom-layout choices and ask it to fill it up, create mood boards for various occasions, and digital illustrations for medical, architectural, and other purposes. This includes a variety of marketing collateral, such as visiting cards, leaflets, social media posts, advertisements, and more.

As of now, it’s a pretty good tool. Even if you aren’t using it for business purposes, you can experiment with it. However, be warned: If you’re using a free account, you’re currently offered 3 images per day, and could face a message saying, “Lots of people are creating images right now, so this might take a bit. We'll notify you when your image is ready.”

Again, it changed our logomark, but here's ChatGPT’s image generator attempting to generate “a detailed 3D render of [our] logo made from colorful felt.”
 

Conclusion

While enterprises and organizations can afford full-fledged graphics designing departments, startups and smaller businesses may not be able to afford such a luxury. However, visually appealing designs are a must in today’s digital media-driven world.

Luckily for them, OpenAI’s ChatGPT image generator has advanced enough to provide instant creatives that look good, can be edited and supported with text. Of course, improvements and upgrades are always on the cards—which is why OpenAI asks for feedback after every two image generations.

Well, off you go now–try your hand at it and post some results in the comments below!