29 December 2025 | Issue #119

In this issue:

Breaking down the top AI visualization tools:
- ChatGPT 5.2
- Midjourney 7
- Firefly 5
- Meta AI
- Gemini 3 (Nano Banana)
- Stable Diffusion (Base)
- Canva AI
The Visual Marketer Now in Full Color
The Visual Marketer is one of the top marketing books of the year
Design Quote of the Week

AI Image Generators for 2026

With the number of image generators available and the speed at which these models are improving, it's impossible to keep up unless you're using them all the time.

I have my preferences, but I wanted to see where we stand as 2026 gets started.

To test a handful of top generative AI models, I wanted to use the same prompt across seven tools.

Each week, I define my subject and add the same description to my prompt so my newsletter images stay on theme. So let's create options for this week's newsletter.

This is the prompt that I used for all of these images:

a woman is looking at an array of images on the wall, she's evaluating six different images. A vibrant, high-contrast illustration using flat vector-style shading with neon and duotone color blocking. The characters are stylized with exaggerated lighting in magenta and yellow tones, set against a cyan background. There’s a strong use of shadow and highlight to create depth without using gradients. It uses clean lines and a minimal background to keep focus on the figures and objects. --ar 16:9

Traditional printing uses cyan (blue), magenta (pinkish), yellow, and black to create a full color gamut. Because this newsletter has a design focus, the feature image for each issue uses colors based on cyan, magenta, and yellow. That's been the visual brand since I updated the logo in issue 100.

ChatGPT 5.2

The Basics

ChatGPT’s image generator was released in March 2025, with a major upgrade rolling out in December 2025. What sets it apart from OpenAI’s earlier image tool, DALL·E, is that image creation happens inside a conversation. You can describe what you want, see the result, then refine it using natural language without starting over.

This makes image creation feel more like collaborating with a creative partner than writing a perfect prompt. You can write things like “make it more minimal,” “change the color palette,” or “move the subject to the left,” and iterate naturally. DALL-E acts more like Midjourney.

I'm on the ChatGPT Pro ($20/mo) plan.

Image Creation

As for the image, it kind of followed the instructions, but not fully. The presence of reds and blacks shows it didn't fully understand my request to simply use cyan, magenta, and yellow.

Also, practically every ChatGPT image you see online uses the same illustration style. You might think you're creating something new and unique, but everyone else is using that same style. Here's a sampling from my LinkedIn feed that I saw while scrolling yesterday.

Almost all ChatGPT illustrations look the same

Midjourney 7.0

The Basics

Midjourney was one of the early winners in the generative AI image race. Originally, only available through Discord, users now have the ability to create the images right on the Midjourney site. It also has the ability to animate images.

Midjourney takes a prompt and outputs four options. You can set tolerances for how creative you want the image to be. I still use Discord because I keep separate Channels for different use cases. It helps me keep organized.

I pay $10/mo for Midjourney

Image Creation

This image did a good job following instructions. There was another option where I liked the composition more, but it didn't follow the color instructions properly.

I could have gone through a few more rounds of edits to get the composition and color combination that I liked, but for this experiment, I wanted to compare first drafts. I didn't use the Discord channel I usually use to create my Marketing by Design images because I wanted a fresh image so it could be properly evaluated.

Firefly 5

The Basics

Adobe Firefly is Adobe’s generative AI system, built specifically with commercial use in mind. It is tightly integrated into tools marketers already use, like Photoshop, Illustrator, and Express. Firefly focuses on tasks like image generation, background replacement, text effects, and generative fill.

What makes Firefly appealing to marketers is trust and workflow. Adobe emphasizes licensed and safe training data, which matters for brand teams. Firefly feels less experimental than some other tools, but much more practical for production work that needs to ship.

Firefly offers different models that can be used to generate the image. For this exercise, I chose Firefly Image 5 (Preview). When using generative AI in Photoshop, you can choose which model you want to use.

I pay for Adobe Creative Cloud (WAY too much per year) which includes Firefly, but you can get Firefly Standard starting at $10/mo

Image Creation

It's ... fine. It followed the basics of the prompt (except it included the color black), but it's not a good composition. I don't think anyone looking at art stands right along the wall. It looks like where GenAI image generators were a few years ago.

Meta AI

The Basics

Meta AI’s image generation tools are embedded inside Meta’s ecosystem, including platforms like Instagram, Facebook, and WhatsApp. The goal here is fast, lightweight creation rather than deep creative control. You describe an image and get something usable almost instantly.

For marketers, Meta AI is less about crafting hero visuals and more about speed and social-first content. It is not the most refined tool, but it lowers the barrier to creating images directly inside the platforms where content is published.

Meta AI offers four options and the ability to add animation.

Image Creation

I liked the image, but it's not really conveying what I was looking for. The concept isn't obvious.

As you can see from the options, they look cool, but they either don't nail the concept or the colors. Since Meta AI allows for back-and-forth conversation, it probably wouldn't have taken too long to get a result I would use.

Gemini 3 (Nano Banana Pro)

The Basics

Gemini’s image generation capabilities are part of Google’s broader AI ecosystem. “Nano Banana” is the informal name people use to describe Google’s fast, lightweight image generation model that emphasizes speed and responsiveness over heavy artistic styling.

Gemini shines when images are part of a larger workflow, like presentations, documents, or search-related content. It feels more utilitarian than expressive, which can actually be a strength for marketers who need visuals that support content rather than dominate it.

I use a Google AI Pro account.

Image Creation

I have to say, I was thoroughly disappointed in this result. Nano Banana has great for creating photo-realistic imagery, but this is just boring. The composition is clear, but it's not engaging.

I'll have to remember that for photos, Gemini is great. For illustrations, I'll need to use another service.

Stable Diffusion (Base)

The Basics

This is where I got my start with image generation. Stable Diffusion is an open-source image generation model, which makes it very different from most other tools on this list. You can run it locally, use it through third-party interfaces, or integrate it into custom workflows. For this test, I used StableDiffusionWeb(.)com.

This flexibility is powerful, but it comes with complexity. Stable Diffusion rewards experimentation and technical curiosity. For marketers, it is often more work than necessary unless you want deep control, custom styles, or complete ownership of the pipeline.

Because it's open source, there are free and paid options.

Image Creation

This is a decent image. It delivered what I asked. But it's not overly interesting. The woman isn't separated enough from the background. Because of the angle and colors, it takes a second to fully understand the image. The illustration style looks generic.

Canva AI

The Basics

Canva’s AI image tools are built directly into a platform many marketers already use daily. Image generation in Canva is designed to be simple, fast, and tightly integrated with layouts such as social posts, presentations, and ads.

What Canva does especially well is context. You are not just generating an image, you are generating it inside a design. For marketers who are already comfortable with Canva, this makes AI image creation feel like a natural extension rather than a new tool to learn. For this test, I just created the image, not within any specific visual.

I pay for Canva so I have access to its AI and Brand Kit tools.

Image Creation

This actually looks closest to the images I usually create. It did a better job mimicking my Midjourney images than Midjourney did. The color is what I wanted (even if the woman's hair is a little too dark). The images within the frames could be changed up a bit.

Results

Of the seven options, Midjourney, Meta AI, and Canva AI are the ones that work the best.

Midjourney does far more than what's expected. The figure is clearly on a different plane than the images on the wall. The lighting adds mood, even though it uses vibrant colors.

Meta AI doesn't quite get the composition correct, but it is an engaging image. It's more impressionistic than the others. The figure draws the viewer in more than the other options, but part of that is due to the fact it didn't fully follow the instructions. The images in the background add a level of texture that is missing from almost all of the others.

Canva AI shows depth and a good composition. The colors and abstract coloring fit the aesthetic that I've been using for the past few months.

Photos

Now, let's do a similar exercise, but see how it works when we request a photo.

This is the prompt I used across the different models:

a photo of a woman is looking at an array of images on the wall. She's evaluating six different images. A vibrant, high-contrast photo. The lighting is dramatic, and exaggerated in magenta and yellow tones, set against a cyan background. There's a strong use of shadows and highlights to create depth. Use minimal background to keep focus on the figures and objects. --ar 16:9

ChatGPT 5.2

Midjourney 7

Firefly 5

Meta AI

Gemini 3 (Nano Banana)

Stable Diffusion (Basic)

Canva AI

Photo Results

This one is harder to pick the best one. These are each drastically different.

Firefly, again, comes up with the least interesting one. I'm glad that Adobe used only licensed photos to train its model, but it's producing subpar images. It's good to use within Photoshop for removing the watermarks on images created by other AI models.

Canva AI, Stable Diffusion, and Gemini 3 produced expected results. Nothing too flash, nothing too eye-catching.

Meta AI, Midjourney 7, and ChatGPT 5.2 created photo-realistic images that are stand out from the sea of standard AI images.

One of the biggest challenges with AI is creating similar results attempt after attempt.

(BTW, if you ever want to see who the fake AI experts are, ask them to get similar results time after time.)

If you're looking to create imagery that uses the same type of imagery on a regular basis, these hyper-stylized images may be hard to replicate.

If you wanted to replicate the style, you can always upload the image to an AI tool, such as ChatGPT or Meta AI, and ask it to describe the image. After a few attempts, you might end up with a highly detailed prompt that will help you replicate the image style.

Which Works for You?

Unfortunately, the answer comes down to "it depends" (I know, I hate that answer, too). Depending on what you're looking to do, you may need to use different models. The other part to take into account is what models do you already have? If you're using ChatGPT or Gemini, you might already have access to these tools. The same goes for your Canva account if you pay for it.

For those who generate AI imagery, which models do you prefer?