I Reviewed the Best AI Image Generators for 2025

I tested the most popular AI image generators to discover their greatest strengths and weaknesses.

At Ahrefs, we have a team of extremely skilled (and very human) designers, but not everyone has that luxury. I wanted to know: are AI image generators useful for spinning up quick social media posts, creating blog post graphics, or saving a few bucks on expensive stock photography?

So I tested out the most popular cloud-based text-to-image tools: DALL-E 3 (available in ChatGPT), Midjourney, Canva’s Magic Media, Adobe Firefly, and the very new Gemini for Workspace.

All these tools generate images in a few clicks, without needing to do anything complicated like training custom models or running programs locally on your computer.

The best AI image generator is, in my opinion, Adobe Firefly. All the models had their own strengths, but Firefly offered maximum control over image generation and image editing.

Here are the pros and cons (and many, many images) sharing my experience with each.

I wanted to test each AI image generator in a range of different scenarios, so I created tons of prompts across three main categories:

I tested different levels of prompt complexity, but kept my prompts generally simple. The whole point of these text-to-image tools is to describe something that you want and have the AI create it for you, so I purposefully avoided PhD-level prompt engineering or professional design lingo.

Here’s a photo of me running these tests:

I then judged each AI image generator’s output across a few key dimensions:

Here are my findings.

Adobe Firefly has—by far—the best editing controls of the image generators I tested. This isn’t surprising, considering that Adobe makes Photoshop, and Illustrator, and Lightroom, and dozens of other market-leading design tools.

Here’s an example. The prompt “A cartoon character with ginger hair carrying a huge golden key to represent ‘keyword research’” generated a series of okay-but-not-great images. But in a few clicks, I was able to fix the biggest problems and dramatically improve the result.

Here’s the before:

In a few minutes using Firefly, I was able to:

And here’s the after:

Adobe Firefly also gives you a ton of control over the image-generation process. A big plus: you can use existing images as style and composition references, making it much easier to generate a series of images with a cohesive style.

Here’s the prompt “A cartoon character with ginger hair carrying a huge magnifying glass to represent ‘competitor research’”, but using my previous image generation as a reference:

The style is slightly different, but they feel recognisably similar. You can also specify particular reference styles, compositions, content types (like art versus photo), and even effects (colour, lighting, bokeh, camera angles, you name it).

That means you can use the same prompt but get very different results. Here’s the result for the prompt “Beautiful minimalist home office with a view of trees outside” when I’ve specified golden hour lighting and warm tones:

And here I’ve used the same prompt but asked for low lighting and cool tones for a very different vibe:

And because Firefly is made by Adobe, you can import your generated images into other Adobe products to add text or edit further. Pretty handy.

Midjourney is beautiful. I’ve been a paying Midjourney customer for three years for the simple reason that everything it generates is gorgeous, and more aesthetically pleasing than any other AI model I’ve tested.

I use Midjourney to illustrate my creative writing, and it excels at fantasy-style illustration. Here’s an image I created for one of my novels, with no editing or manipulation:

It’s also pretty handy for photorealism too. Here’s the prompt “Stock photo of a beautiful minimalist home office with a view of trees outside”:

There are a couple of AI-isms (how many wheels does that chair have?!), but I want to forgive them because the photo is so damn beautiful.

Here’s “Stock photo of a thoughtful person in a meeting at a software company”, featuring an AI-generated man so handsome I didn’t want to look in a mirror for the rest of the day:

Even Midjourney cartoon illustrations look classy, and almost good enough to be plucked from the frames of a Pixar film:

Midjourney does have weaknesses. It categorically cannot do data visualization. Feed it even simple data and it will generate nonsense (but it will at least be beautiful nonsense):

Midjourney’s editing workflows are much better than they used to be, but still not very sophisticated. As well as generating four images for every prompt, you have the option to:

Here’s an example of varying an image. There are small, subtle differences between each photo, like the number of wheels on the chair—helpful for minimizing any weird AI-isms in images you like:

These options are nowhere near as precise as Adobe Firefly’s editing workflow, but given Midjourney’s ability to make generally beautiful images from simple, single prompts, this workflow creates surprisingly useful images.

(And as a final bonus, you no longer have to rely on a janky Discord server to generate images—Midjourney’s web app works very well.)

Given the popularity of ChatGPT, DALL-E 3—the image generation model offered as part of ChatGPT—will be most people’s first introduction to AI image generators. That’s a shame, because it’s one of the worst.

To make this point, here’s what happened when I asked for a “Stock photo of someone working on their laptop in a New York coffee shop”:

This is pretty representative of DALL-E 3: most of its images look and feel like they’re AI-generated.

Look for a moment and you’ll spot nonsense text, furniture blending into the background, a weird uncanny-valley glow to the main character, straight lines that are never straight… and most of ChatGPT’s images suffer from the same issues.

Here’s ChatGPT trying to gaslight me into believing that this is a photograph of a home office (the trees look like a freaking pointillism painting):

These issues are at least less obvious in cartoon imagery. Here’s our character holding a key again:

Not bad, despite a couple of AI-isms, like the double-ended key and weird abstract backpack charm. Unfortunately, I couldn’t remove these little quirks, because even though ChatGPT recently added the ability to highlight parts of the image to selectively edit, this feature was super unreliable when I tested it.

On one occasion, ChatGPT even decided that, actually, no, it didn’t want me to do any image editing:

Without much control over image generation or editing, DALL-E 3 is a bit of a crapshoot, and it’s virtually impossible to carry consistent styles across images.

When I tried to make a new image with the same cartoon character, it changed style radically:

You can’t easily upscale your images either, and when I asked ChatGPT to resize a YouTube thumbnail to 16:9 resolution, it decided to write a Python script to stretch the image to landscape format.

Which, err… did not look good:

When I tried to refine the prompt to reflect Ahrefs’ brand guidelines, it gave me a lecture on designing thumbnails, and didn’t actually make an image.

Generating images with ChatGPT reminds me playing the video game DOOM on a calculator. It might technically be possible, but you probably shouldn’t do it.

ChatGPT had one big redeeming virtue, where its penchant for Python was extremely useful: data visualization. It was the only AI image generator capable of actually turning a list of data points into an accurate graph:

And it can handle more complex data visualisations too:

This is a different type of “image generation”, but for someone like me who wrangles data on a daily basis, incredibly useful, and a feature I use all the time.

Canva’s Magic Media is an AI image generator embedded directly within the main Canva app. To get started, you’re offered a choice of image, graphic, or video.

It handles stock photography pretty well: here’s our prompt for a beautiful home office:

You can pick one of around two dozen specific styles to emulate, and pre-set the aspect ratio of the photo. Here’s our New York coffee shop with the Moody style applied:

Here, we begin to see Magic Media’s biggest weakness creeping in: uncanny valley photorealism.

Here’s another stock photo attempt that almost looks good… except for the deformed hands, confusing arm physics, and background ensemble of melty-faced monsters:

It’s useful for generating vector art too, and the images can be exported directly as PNGs with no background, but the images themselves are a little amateurish.

Here’s our key-holding cartoon figure again, this time holding a perfectly smooth key in one hand and a smaller, seemingly melted key in the other:

Here’s the terrifying result of using the same prompt with the 3D Chrome style applied:

Because Magic Media is embedded in Canva, it’s incredibly easy to add text, resize the finished image, or add effects to the generated images. That’s a big plus, but in my opinion, not enough to compensate for amateurish quality of the image generation.

Here’s an example of how fast AI tools are developing. As I was writing this blog post, Google added AI image generation capabilities directly into Google Docs. Now, you can use the @image command and select “Help me create an image.”

It’s pretty simple. You can use one of three aspect ratios and specify one of six pre-determined styles, and Google returns four images to choose from.

Here’s a decent little image for the prompt “A cartoon character with ginger hair carrying a huge magnifying glass”:

And here’s “A cartoon character with ginger hair carrying a huge golden key” with the Watercolor style applied:

Although these cartoons are decent, Gemini seems to have a special skill: photography. It rendered beautiful scenes for my home office prompt with the Photography style selected:

And Gemini for Workspace seems to handle photos of people even better. Here’s a very realistic rendition of “Stock photo of someone working on their laptop in a New York coffee shop”—even down to the Apple logo on the laptop:

And here’s “Photo of a woman giving a talk on stage”. I cannot tell this image was AI-generated:

These images are small and low-resolution, but as a big plus, you can generate them in the flow of work—pretty useful for adding in a quick mock-up or placeholder to pass on to your design team or improve in the future.

This is obviously a very new feature (when I tested it, image generation failed for me about 70% of the time), but I’d expect it to improve pretty quickly and become a major contender for best AI image generator.

Final thoughts

AI text-to-image generators are at their best when you ask for straightforward designs and don’t have a particularly strong opinion of the exact image you want to see. If you want a quick stock photo or blog illustration, and don’t have to worry about pesky brand guidelines, most of these tools are up to the task (apart from maybe ChatGPT… sorry).

But the more specific detail you want from the image—words, numbers, particular brand guidelines—and the stronger your opinion about what you want the final image to look like, the more frustrating these tools become.

I think Adobe Firefly is the best AI image generator because it sits at the intersection between generative AI and traditional design tools. It pairs all the creative benefits of AI with the editing control of Photoshop or Illustrator. That means it can tackle complicated design workflows, like creating a series of cohesive characters, or applying particular styles or compositions. If you’re serious about using AI image generators for your brand or business, I’d start with Firefly.

I’ll keep updating this post as new AI image generators are released and existing tools continue to get updated. Want to ask me to review a tool for you? Let me know on LinkedIn.