I examined the preferred AI picture turbines to find their largest strengths and weaknesses.
At Ahrefs, we now have a group of extraordinarily professional (and really human) designers, yet now not everybody has that luxurious. I sought after to grasp: are AI picture turbines helpful for spinning up fast social media posts, growing weblog publish graphics, or saving a couple of greenbacks on dear inventory images?
So I examined out the preferred cloud-based text-to-image equipment: DALL-E 3 (to be had in ChatGPT), Midjourney, Canva’s Magic Media, Adobe Firefly, and the very new Gemini for Workspace.
These kinds of equipment generate pictures in a couple of clicks, without having to do the rest sophisticated like coaching customized fashions or operating systems in the neighborhood in your pc.
The most efficient AI picture generator is, individually, Adobe Firefly. All of the fashions had their very own strengths, yet Firefly introduced most regulate over picture era and picture modifying.
Listed here are the professionals and cons (and lots of, many pictures) sharing my revel in with each and every.
AI picture generator | Easiest for… | Pricing |
---|---|---|
Adobe Firefly | Easiest for optimum regulate over pictures | 25 unfastened credit monthly; $4.99 for 100 credit |
Midjourney | Easiest for gorgeous pictures | From $10/m for 200 generations |
DALL-E 3 / ChatGPT | Easiest for information visualization | 2 unfastened pictures in line with day at the Unfastened plan; complete get entry to begins at $20/m at the Plus plan |
Canva Magic Media | Easiest for producing vector pictures | 50 pictures to be had for Canva Unfastened customers; 500 pictures monthly for paid customers (from $14.99/m) |
Gemini for Workspace | Easiest for fast concepting | To be had as a Google Workspace add-on from $20/m |
I sought after to check each and every AI picture generator in a variety of various eventualities, so I created lots of activates throughout 3 primary classes:
- Inventory images (e.g. “Inventory picture of an attractive minimalist house place of business with a view of bushes out of doors”)
- Graphics and illustrations (e.g. “A cool animated film personality with ginger hair sporting an enormous golden key to constitute ‘key phrase analysis’”)
- Knowledge visualizations (e.g “Graph of web site visitors information: January 946, February 1071, March…”)
I examined other ranges of urged complexity, yet saved my activates in most cases easy. The entire level of those text-to-image equipment is to explain one thing that you wish to have and feature the AI create it for you, so I purposefully have shyed away from PhD-level urged engineering or skilled design lingo.
Right here’s a photograph of me operating those checks:

(Simply kidding, this was once generated via Adobe Firefly the usage of a photograph of myself as a reference. We are living in odd instances.)
I then judged each and every AI picture generator’s output throughout a couple of key dimensions:
- Accuracy: how smartly did the picture generator apply my path?
- Ease of modifying: how simple was once it to edit and refine the output?
- Uncanniness: did the output glance bizarre or clearly AI-generated?
- Legibility of textual content: how smartly did the type maintain textual content era?
- Consistency: may just I reproduce equivalent pictures on more than one events?
- Usefulness: may just I in reality use the output in actual lifestyles?
Listed here are my findings.
Adobe Firefly has—via a ways—the most productive modifying controls of the picture turbines I examined. This isn’t sudden, making an allowance for that Adobe makes Photoshop, and Illustrator, and Lightroom, and dozens of different market-leading design equipment.
Right here’s an instance. The urged “A cool animated film personality with ginger hair sporting an enormous golden key to constitute ‘key phrase analysis’” generated a sequence of okay-but-not-great pictures. However in a couple of clicks, I used to be ready to mend the most important issues and dramatically support the outcome.
Right here’s the sooner than:


AI is famously dangerous at drawing palms—so dangerous that Firefly bailed midway thru.
In a couple of mins the usage of Firefly, I used to be ready to:
- Resize the facet ratio from 1:1 to 4:3 the usage of generative fill.
- Repair a lacking hand via prompting Firefly to regenerate that particular portion of the picture.
- Upscale the small, low-quality picture to a a lot more helpful 2k decision.
And right here’s the after:
Adobe Firefly additionally provides you with a ton of regulate over the image-generation procedure. A large plus: you’ll use current pictures as genre and composition references, making it a lot more uncomplicated to generate a sequence of pictures with a cohesive genre.
Right here’s the urged “A cool animated film personality with ginger hair sporting an enormous magnifying glass to constitute ‘competitor analysis’”, yet the usage of my earlier picture era as a reference:
The manner is reasonably other, yet they really feel recognisably equivalent. You’ll additionally specify specific reference kinds, compositions, content material varieties (like artwork as opposed to picture), or even results (color, lights, bokeh, digital camera angles, you identify it).
That suggests you’ll use the similar urged yet get very other effects. Right here’s the outcome for the urged “Gorgeous minimalist house place of business with a view of bushes out of doors” once I’ve specified golden hour lights and heat tones:
And right here I’ve used the similar urged yet requested for low lights and cool tones for an excessively other vibe:
And since Firefly is made via Adobe, you’ll import your generated pictures into different Adobe merchandise so as to add textual content or edit additional. Lovely at hand.
Midjourney is lovely. I’ve been a paying Midjourney buyer for 3 years for the straightforward explanation why that the whole lot it generates is lovely, and extra aesthetically satisfying than every other AI type I’ve examined.
I exploit Midjourney for example my ingenious writing, and it excels at fantasy-style representation. Right here’s a picture I created for one in all my novels, without a modifying or manipulation:
It’s additionally lovely at hand for photorealism too. Right here’s the urged “Inventory picture of an attractive minimalist house place of business with a view of bushes out of doors”:
There are a few AI-isms (what number of wheels does that chair have?!), yet I need to forgive them for the reason that picture is so rattling gorgeous.
Right here’s “Inventory picture of a considerate particular person in a gathering at a device corporate”, that includes an AI-generated guy so good-looking I didn’t need to glance in a replicate for the remainder of the day:
Even Midjourney cool animated film illustrations glance stylish, and nearly excellent sufficient to be plucked from the frames of a Pixar movie:


Let’s forget about the backwards left hand.
Midjourney does have weaknesses. It categorically can not do information visualization. Feed it even easy information and it is going to generate nonsense (yet it is going to a minimum of be gorgeous nonsense):


I stay up for sharing this graph of weblog publish visitors in my Q1 recap.
Midjourney’s modifying workflows are a lot better than they was once, yet nonetheless now not very subtle. In addition to producing 4 pictures for each and every urged, you will have the approach to:
- Range any unmarried picture, both robust or delicate (principally regenerate a picture that’s similar to the former).
- Upscale pictures you favor to better decision.
- Take away portions of the picture (yet now not specify what you’d like to switch it with).
- Exchange the facet ratio (sq., 4:3, 16:9, and so on).
Right here’s an instance of various a picture. There are small, delicate variations between each and every picture, just like the collection of wheels at the chair—useful for minimizing any bizarre AI-isms in pictures you like:
Those choices are nowhere close to as actual as Adobe Firefly’s modifying workflow, yet given Midjourney’s skill to make in most cases gorgeous pictures from easy, unmarried activates, this workflow creates strangely helpful pictures.
(And as a last bonus, you now not need to depend on a janky Discord server to generate pictures—Midjourney’s internet app works very smartly.)
Given the recognition of ChatGPT, DALL-E 3—the picture era type introduced as a part of ChatGPT—will probably be the general public’s first advent to AI picture turbines. That’s a disgrace, as it’s one of the most worst.
To make this level, right here’s what took place once I requested for a “Inventory picture of somebody operating on their pc in a New York espresso store”:
That is lovely consultant of DALL-E 3: maximum of its pictures feel and appear like they’re AI-generated.
Search for a second and also you’ll spot nonsense textual content, furnishings mixing into the background, a peculiar uncanny-valley glow to the principle personality, immediately traces which are by no means immediately… and maximum of ChatGPT’s pictures be afflicted by the similar problems.
Right here’s ChatGPT seeking to gaslight me into believing that this can be a {photograph} of a house place of business (the bushes seem like a freaking pointillism portray):
Those problems are a minimum of much less obtrusive in cool animated film imagery. Right here’s our personality keeping a key once more:
Now not dangerous, in spite of a few AI-isms, just like the double-ended key and bizarre summary backpack allure. Sadly, I couldn’t take away those little quirks, as a result of although ChatGPT just lately added the facility to focus on portions of the picture to selectively edit, this option was once tremendous unreliable once I examined it.
On one instance, ChatGPT even determined that, in reality, no, it didn’t need me to do any picture modifying:
With out a lot regulate over picture era or modifying, DALL-E 3 is a little of a crapshoot, and it’s just about inconceivable to hold constant kinds throughout pictures.
Once I attempted to make a brand new picture with the similar cool animated film personality, it modified genre radically:
You’ll’t simply upscale your pictures both, and once I requested ChatGPT to resize a YouTube thumbnail to 16:9 decision, it determined to write a Python script to stretch the picture to panorama structure.
Which, err… didn’t glance excellent:
Once I attempted to refine the urged to replicate Ahrefs’ emblem tips, it gave me a lecture on designing thumbnails, and didn’t in reality make an picture.
Producing pictures with ChatGPT strikes a chord in my memory playing the video game DOOM on a calculator. It could technically be imaginable, yet you almost certainly shouldn’t do it.
ChatGPT had one large redeeming distinctive feature, the place its penchant for Python was once extraordinarily helpful: information visualization. It was once the handiest AI picture generator in a position to in reality turning a listing of information issues into a correct graph:
And it could actually maintain extra complicated information visualisations too:
It is a other form of “picture era”, yet for somebody like me who wrangles information every day, extremely helpful, and a function I exploit all of the time.
Canva’s Magic Media is an AI picture generator embedded immediately inside of the principle Canva app. To get began, you’re introduced a selection of picture, graphic, or video.
It handles inventory images lovely smartly: right here’s our urged for an exquisite house place of business:


3-legged tables are very “in” proper now.
You’ll select one in all round two dozen explicit kinds to emulate, and pre-set the facet ratio of the picture. Right here’s our New York espresso store with the Moody genre implemented:
Right here, we start to see Magic Media’s greatest weak point creeping in: uncanny valley photorealism.
Right here’s some other inventory picture try that nearly seems excellent… except for for the deformed palms, complicated arm physics, and background ensemble of melty-faced monsters:
It’s helpful for producing vector artwork too, and the photographs can also be exported immediately as PNGs without a background, however the pictures themselves are slightly amateurish.
Right here’s our key-holding cool animated film determine once more, this time keeping a superbly clean key in a single hand and a smaller, apparently melted key within the different:
Right here’s the terrifying results of the usage of the similar urged with the 3-D Chrome genre implemented:


Expensive Canva: no-one requested for this.
As a result of Magic Media is embedded in Canva, it’s extremely simple so as to add textual content, resize the completed picture, or upload results to the generated pictures. That’s a large plus, yet individually, now not sufficient to catch up on amateurish high quality of the picture era.
Right here’s an instance of ways speedy AI equipment are growing. As I used to be penning this weblog publish, Google added AI picture era functions immediately into Google Medical doctors. Now, you’ll use the @picture command and make a selection “Lend a hand me create an picture.”
It’s lovely easy. You’ll use one in all 3 facet ratios and specify one in all six pre-determined kinds, and Google returns 4 pictures to choose between.
Right here’s a tight little picture for the urged “A cool animated film personality with ginger hair sporting an enormous magnifying glass”:
And right here’s “A cool animated film personality with ginger hair sporting an enormous golden key” with the Watercolor genre implemented:
Even supposing those cartoons are first rate, Gemini turns out to have a distinct ability: images. It rendered gorgeous scenes for my house place of business urged with the Images genre decided on:
And Gemini for Workspace turns out to maintain pictures of other folks even higher. Right here’s a very practical rendition of “Inventory picture of somebody operating on their pc in a New York espresso store”—even right down to the Apple emblem at the pc:
And right here’s “Photograph of a girl giving a chat on level”. I can’t inform this picture was once AI-generated:
Those pictures are small and low-resolution, yet as a large plus, you’ll generate them within the waft of labor—lovely helpful for including in a handy guide a rough mock-up or placeholder to cross on for your design group or support at some point.
That is clearly an excessively new function (once I examined it, picture era failed for me about 70% of the time), yet I’d be expecting it to support lovely temporarily and grow to be a big contender for very best AI picture generator.
Ultimate ideas
AI text-to-image turbines are at their very best whilst you ask for simple designs and don’t have a in particular robust opinion of the precise picture you wish to have to look. If you wish to have a handy guide a rough inventory picture or weblog representation, and don’t have to fret about pesky emblem tips, a lot of these equipment are as much as the duty (except for perhaps ChatGPT… sorry).
However the extra explicit element you wish to have from the picture—phrases, numbers, specific emblem tips—and the more potent your opinion about what you wish to have the overall picture to seem like, the extra irritating those equipment grow to be.
I believe Adobe Firefly is the most productive AI picture generator as it sits on the intersection between generative AI and standard design equipment. It pairs all of the ingenious advantages of AI with the modifying regulate of Photoshop or Illustrator. That suggests it could actually take on sophisticated design workflows, like growing a sequence of cohesive characters, or making use of specific kinds or compositions. When you’re fascinated about the usage of AI picture turbines on your emblem or trade, I’d get started with Firefly.
I’ll stay updating this publish as new AI picture turbines are launched and current equipment proceed to get up to date. Wish to question me to check a device for you? Let me know on LinkedIn.