Sign up here to receive our free Work Tech newsletter in your inbox.

Charter Work Tech reviews—including our picks and the products we choose to review—are editorially independent. If you purchase something through a link on our site or in our newsletter, Charter may earn an affiliate commission.

I remember playing with OpenAI’s DALL·E 2 for the first time in late summer 2022. The AI-powered image generator was impressive, creating cool images based on a simple string of words. It was also funny: The images were often clunky and weird-looking (we’ve all seen the seven-fingered people at this point). Roughly a year later, though, image generators are in a completely different league. 

Casey Newton, founder and editor of the technology newsletter Platformer, recently wrote that text-to-image generators are a better way to understand rapid improvements in AI than chatbots because you can actually see the differences. For example, here are two images I created with DALL·E 2 and DALL·E 3 11 months apart, based on the same, very basic prompt. 

Image generated by DALL·E 2
Image generated by DALL·E 3

We put five different image generators to the test, challenging them to produce a series of images that would be useful in different settings. Our top picks are DALL·E 3, Magic Media, and MidJourney, with DALL·E 3 coming out on top.

We had the winner, DALL·E 3, produce an image based on the text of this article, and we told it to emphasize the fact that it won. Here’s what it gave us:

Image generated by DALL·E 3

We’ll try not to let its achievement be overshadowed by the fact that it misspelled its own name.

Our picks



Overall DALL·E 3  is the strongest AI image generator of the ones we tested. It was the top performer on two out of four of the image generation tasks, and it is by far the easiest tool to use. 


  • You don’t need a special prompt—just tell it in simple terms what you want.
  • It has the highest floor for image quality, always producing images that are at least solid and often better.
  • You can converse with it through ChatGPT Plus to edit the image, get new ideas, etc.


  • Its images often look the same unless you push it to mix things up.

Magic Media


  • It produces solid images. 
  • You can use it for free 50 times through Canva Free.
  • The objects in its images are realistic.


  • The faces of people in its images are distorted.
  • You need more detailed prompts than you do with DALL·E 3.



  • Its images are often very intricate, with lots of details and rich colors.
  • Its Basic Plan is a reasonable price for ~200 images per month.


  • You access it through Discord, which isn’t a great interface for the user.
  • You need more detailed prompts than you do with DALL·E 3.
  • It has a very specific look, giving you less variety in outputs.

Our process

We evaluated five AI image generators: 

  • DALL·E 3 (beta), by OpenAI, which can be accessed through ChatGPT Plus,  and Microsoft’s Bing Chat and Bing Image Creator.
  • Firefly Image 2 (beta), by Adobe.
  • Midjourney, by Midjourney.
  • Magic Media, by Canva, which is powered by Stable Diffusion but also has its own customizations built on top.
  • DiffusionBee, which lets you run Stability AI’s Stable Diffusion model on your computer.

We first tested their ability to generate two images:

  1. An open office full of employees working at their desks.
  2. A castle nestled between two mountains, with a beach in the foreground.

Based on those two tasks, we eliminated DiffusionBee, which produced images of a much lower quality than the other tools, and tested the remaining four tools on two more tasks:

  1. A logo for a futuristic company named ‘Charter.’
  2. An image of a city in 2100 with futuristic buildings and technology.

We then chose our top three based on how they stacked up on all four image tasks. We used similar prompts for each text-to-image generator, but adjusted them based on each tool’s restrictions and the results we saw. For example, Canva’s Magic Media required shorter prompts. DALL·E 3, meanwhile, took our prompts and improved them (see the “Our recommendations” section for more information about this process).

Our recommendations

Our top pick is DALL·E 3, which costs $20 per month through ChatGPT Plus, or can be accessed for free through Microsoft’s Bing Chat and Bing Image Creator. Magic Media, our second pick, can be accessed for free with a Canva Free account or through a Canva Pro account for $14.99 per month billed monthly (see Pricing Deep Dive for details). Midjourney, which came in third, starts at $10 per month for the Basic Plan billed monthly. Here’s how each of our picks stacks up on a few key features:

Overall image quality: DALL·E 3 takes the top prize on quality. OpenAI’s image generator is very good at producing passable images quickly. That doesn’t mean its first pass is always something you would use—it often isn’t—but it does mean that DALL·E 3 has the highest floor for image quality of our three top picks. The image generator sometimes struggles to make realistic images, though. For example, when we asked it to make an image of a bustling open office look more realistic, it kept giving me different versions of this:

Image generated by DALL·E 3

It’s a good image, but it looks like a video game.

Canva’s Magic Media comes in second for image quality, and it outperformed DALL·E 3 for generating realistic images, like this image of an office:

Image generated by Canva’s Magic Media AI image generator

The image has flaws—e.g., the people’s faces—but the layout and objects are quite good.

Midjourney comes in a close third for image quality, though it was competitive with DALL·E 3 for certain types of images, like the image of the futuristic city. 

Ease of use: DALL·E 3 is the winner here. The biggest barrier to entry for using text-to-image generators is prompting. DALL·E 3, however, can generate images from very simple descriptions by essentially translating your instructions into a prompt that it can work with. For example, I said:

Generate a high-resolution image of a busy office setting, showcasing people working, office furniture, and technology devices.

It then translated that instruction into this prompt:

Photo of a bustling office space with a diverse group of men and women working diligently at their desks. Laptops and monitors light up the room, while overhead lights illuminate the space. In the background, a large glass window lets in natural light. Bookshelves filled with binders and files line the walls. Potted plants add a touch of greenery.

And when using DALL·E 3 in ChatGPT Plus, you can converse with it, giving it feedback and making edits until you get the image that you want. Canva’s Magic Media comes in second, with its easy-to-use interface. Midjourney had the most friction of our top picks, because you have to access it through the communications app Discord.

Variety of outputs: Magic Media and DALL·E 3 are tied here. Midjourney comes in third, with a specific style that makes many of its images look the same.

Customization and control: DALL·E 3 is the leader here, because you can converse with it in ChatGPT Plus to customize the image and get the edits that you want. Magic Media and Midjourney are tied here. The former lets you select different styles for your image, such as “photo” and “minimalist”—you can then edit the image with Canva’s editing tools. With the latter, you can easily alter the image by adding more details to your prompt and upscale the image with the upscale tool.

Pricing deep dive

DALL·E 3 is available through ChatGPT Plus for $20 per month and through ChatGPT Enterprise (inquire for pricing). You can access it for free through Microsoft’s Bing Chat and Bing Image Creator.

Magic Media’s image generator can be access through Canva Free, Canva Pro, and Canva for Teams:

  • Canva Free gets you 50 lifetime uses of the AI image generator.
  • Canva Pro gets you 500 uses per month. It costs $14.99 per month (billed monthly) or $119.99 per year (billed annually).
  • Canva for Teams gets each user 500 uses per month, with a sliding scale fee starting at $300 per year for five people, $1,050 for 10 people, and so on—all billed monthly. You can calculate the cost for your team on Canva’s website.

See plans and pricing for more information about each plan.

Midjourney has four different pricing plans: 

  • Basic Plan gets you 3.3 hours per month of “fast GPU time.” According to Midjourney, the average image request takes about one minute of GPU time, meaning that the Basic Plan gets you roughly 200 images per month. The plan costs $10 per month billed monthly or $8 per month billed annually. 
  • Standard Plan gets you 15 hours per month of fast GPU time, but it also gets you unlimited “relax GPU time,” which means you get unlimited images. When you’re in relax mode, your requests get placed in a queue of up to 10 minutes. This plan costs $30 per month billed monthly and $24 per month billed annually. 
  • Pro Plan gets you 30 hours per month of fast GPU time and unlimited relax GPU time. The plan costs $60 per month billed monthly and $48 per month billed annually. 
  • Mega Plan gets you 60 hours per month of fast GPU time and unlimited relax GPU time. The plan costs $120 per month billed monthly and $96 per month billed annually.

See subscription plans for more information about each plan.

How we chose what to review

With many image generators on the market to choose from, we focused on the newest versions of the main image-generation models: DALL·E, Stable Diffusion, Midjourney, and Firefly. We also included Canva because of its position as a leading graphic design platform. The company’s Magic Media image generator is powered by Stable Diffusion, but it has its own customizations built on top. 

A note about copyright: There are several ongoing lawsuits regarding AI models and the data they’re trained on, including AI image generators. For example, Getty Images has sued Stable Diffusion creator Stability AI, alleging that the company scraped millions of Getty images without a license. Some artists are bringing their own lawsuits against AI companies, though they appear to be facing an uphill battle

Companies are starting to take measures to address those criticisms. OpenAI now allows artists to opt out of allowing the company to use their images to train future models, and its DALL·E 3 model will not make images in the style of any living artist. Adobe seems to have gone the furthest to alleviate concerns: It has only trained its models on licensed imagery, and it compensates Adobe Stock creators whose work is used to train Firefly. The company even says it will compensate businesses if they’re sued for copyright infringement for images created with Firefly.

You can find more information about this in the terms of use and service for OpenAI, Canva, and Midjourney

Sign up here to receive our free Work Tech newsletter in your inbox.