Nano Banana Pro vs. Midjourney: Testing Educational Explainers and Prompt Accuracy

Over the past nine months, I have been using Midjourney for image generation across a few freelance and side projects. Overall, Midjourney has performed extremely well when it comes to creative flair, atmosphere, mood, artistry, and cinematic lighting. It is especially strong for abstract visuals and highly stylized images.

That said, I have consistently found that Midjourney struggles in a few areas:

• Following detailed or complex prompts and following literal instructions, such as: “The character needs to sit in a chair, look relaxed, look to the left, and hold a book”
• Generating realistic photography, especially scenes involving people
• Editing existing images based on specific instructions, adding or modifying elements within an existing image
• Creating educational illustrations and infographics based on provided instructions.

A while ago, I came across Nano Banana Pro, and I finally had time to give it a try.

Nano Banana Pro is Google’s advanced AI image generation and editing model, built on the Gemini 3 Pro architecture. From what I understood, Nano Banana Pro includes a built-in “reasoning process” that helps it interpret context and instructions before generating an image, making it especially useful for visualizing ideas, creating prototypes, producing educational explainers and infographics, rendering legible text directly inside images, and editing or transforming specific parts of an existing image.

Nano Banana Pro can be accessed through several Google and partner platforms, including the Gemini app, Google AI Studio, Vertex AI, and Adobe Firefly. I am currently accessing it through my Higgsfield.ai account.

So, let’s look at what it generated.

Test 1: Precise Adherence to Instructions

Since detailed prompt-following is one of the areas where Midjourney often struggles, I started with a fairly specific prompt for a random idea that came to mind:

“Carl Jung sitting on a chair by a tree, on the edge of a cliff. The edge of a cliff is in the middle of the image vertically. Below is the abyss with misty landscape and houses far away. Above is the sky with the moon and stars. Behind him are mystical archetypal creatures — human figures or animals. Surreal, liminal, alchemical motifs.”

Yep, that prompt was a random bowl of Jungian word soup, but here is what Nano Banana Pro came up with:

I was genuinely wowed by this result. The model followed the details of the prompt remarkably well, captured the mood, and interpreted the symbolic and surreal elements in a distinctly Jungian way — even though I had not explicitly asked for that, it seemed to infer it naturally from the prompt. Overall, the image has strong artistic quality and carries exactly the vibe I was looking for.

For comparison, here are the variations Midjourney produced from the same prompt (by default, Midjourney generates four variations for a single prompt):

The Midjourney images are artistic, but they missed several key details requested in the prompt — the mystical archetypal creatures, the stars. Also the central figure does not resemble Carl Jung at all.

Test 2: Split-Screen Illustration / Basic Explainer

Next, I wanted to test how Nano Banana Pro handles illustrations, especially images that need to communicate a specific psychological or emotional concept.

Here was the prompt:

“The frame is split into two. The first displays a highly sensitive elementary-school-age boy at school, masking and trying to keep it together. The second displays the same child at home, melting down from emotional exhaustion. Anime cartoon style drawing.”

Nano Banana Pro result:

Midjourney result (4 variations):

In this test, Nano Banana Pro did an obviously better job of understanding the concept behind the prompt, filling in missing contextual details, and presenting the illustrated idea. It even added a couple of text bubbles with the language that fit the context.

And here is Nano Banana Pro result for the same prompt without the “anime cartoon style drawing” detail in the prompt:

The photography is realistic-looking, and the emotional context came through clearly in both scenes. Nano Banana Pro also maintained continuity between the two panels: it is clearly the same person in both situations (and wearing the same pants!)

For the next prompt, I wanted to see how well the model could represent a more subtle psychological context:

“The image is meant to represent a girl who looks cool on the outside — piercings, colored hair, confident expression. The frame is split into two. The first displays her looking confident and almost like a bully in a social setting. The second displays the real her — a kind, tender, sincere, and deep person at night, with the window open and the moon. Cartoon style drawing.”

Nano Banana Pro result:

Midjourney result (4 variations):

Once again, Nano Banana Pro seemed WAY stronger at interpreting the underlying idea, understanding the context, and presenting it visually, including the facial expressions. It knew to add the cat, the tea and the books.

From this point on, I’ll only include Nano Banana Pro results in my tests.

Test 3: Photorealistic Split-Screen Conceptual Image

Next, I tested with a prompt for a realistic editorial-style image rather than a drawing. I came across a similar image on Reddit and thought it would be interesting to try playing with this concept.

Prompt:

“An image that represents how programming / software engineering was done before and after AI. The frame is split into two. The first displays a programmer in a flow state typing on a computer. The second displays the same programmer in front of a computer with several monitors running AI agents/code, while the programmer is leaning back in the chair checking social media on his phone.”

Result:

The overall concept came through well, but there are a couple of issues. The vintage-looking PC from the 90s felt out of place, and I also noticed that the books lying on the desk had nonsensical gibberish titles, apparently must-reads in the field of imaginary AI literature.

So I used the inpainting tool, highlighted the two problem areas, and asked the model to replace the old computer with a laptop and remove the books from the desk.

Final Result:

The final result looks pretty good, and this is one of the areas where Nano Banana Pro feels especially useful: allowing you to refine specific parts of the image without starting over completely.

Test 4: Timeline Infographic

Finally, I tested whether Nano Banana Pro could generate a more structured visual — something closer to a timeline infographic or an educational visual. I asked it to create a history timeline of the IT industry.

Here was the prompt:

“A history diagram that shows the development of the IT industry, including mainframes, personal computers, smartphones, cloud computing, and AI. It needs to show the dates / years, icons, and look interesting. At the bottom of each period / stage, display the names of people or companies that influenced that period / stage. At the top of each period, display the years of the period. The image needs to be in a historic / old style, like old brownish paper.”

Result:

The result was surprisingly strong even on the first pass. It captured the structure, historical progression, requested visual style, and infographic-like layout much better than I would normally expect from an image generation model. What stood out most was that my prompt only specified the overall stages and the type of information to include, yet the model still filled in many of the concrete details associated with each era (although I would make a few adjustments for the Mainframe and AI periods).

Test 5: Infographic / Educational Illustration

And finally, I ran one more educational illustration test — this time explaining the Bohr effect, which is the science behind some popular breathwork methods. In simple terms, it has to do with how CO2 levels in the blood affect how easily oxygen is released from hemoglobin into the tissues. So yes, this one was definitely on the geekier side. Before handing it to Nano Banana Pro, I decided to clean up the prompt with ChatGPT first, mostly to test how Nano Banana Pro does with following precise detailed instructions.

The prompt:

“Create a clean educational split-screen illustration showing why fast, heavy breathing can reduce effective oxygen delivery to tissues.

Left side: a person breathing rapidly, with low CO2 in the blood, hemoglobin holding onto oxygen more tightly, and less oxygen reaching the tissues.

Right side: a person breathing calmly, with healthy CO2 levels, hemoglobin releasing oxygen more easily, and better oxygen delivery to the tissues.

Include lungs, blood vessels, red blood cells, oxygen, and body tissues, with simple arrows and labels.

Add a small note explaining the Bohr effect: higher CO2 helps oxygen unload into tissues; lower CO2 makes oxygen stay bound to hemoglobin.

Style: clear explainer infographic, visually appealing, simple, accurate, and easy for a general audience to understand.”

Result:

I’d say this is not bad at all, and usable on the first pass.

I also tested what happens if I just give it a very basic prompt asking to explain this fairly complex physiological / biochemical concept with the following prompt:

“A visual explaining the Bohr effect”

Result:

This visual is a pretty accurate explanation of the Bohr effect.

I Still LOVE Midjourney for Artistic Imagery

Midjourney has been an essential tool for me when generating creative, artistic, free-flowing imagery for freelance and side projects — including web and mobile apps, blog visuals, YouTube Shorts, and social media content. It is still excellent for mood, atmosphere, cinematic visuals, abstract art, and artistic experimentation.

One of Midjourney’s strengths is that it gives you a lot of creative control through parameters that influence the style of the generated image. For example, the --stylize parameter lets you control how strongly Midjourney applies its own artistic interpretation. A lower stylize value keeps the image closer to the prompt, while a higher value gives Midjourney more freedom to make the image feel dramatic, or visually expressive.

Its image generation is also fast (probably because there isn’t much “reasoning”), and it produces multiple variations from a single prompt, giving you several options to choose from right away. That makes it especially useful when you are experimenting with different moods or styles, or simply want the model to surprise you with unexpected creative options.

Final Thoughts

After testing Nano Banana Pro across several use cases, my impression is that it is especially strong at creating visuals that communicate a specific idea. It performs well both when you give it a fairly simple concept and let it infer the missing details, and when you provide a highly specific prompt with precise instructions. Nano Banana Pro’s photography-style images look realistic, with natural lighting, believable people, realistic facial expressions, and enough visual consistency to feel like real photography.

In the near future, I’ll continue using Midjourney for more artistic, free-flowing imagery — the kind of “surprise me” visuals where mood, style, and creative interpretation matter more than precise instructions. But for images that require accurate prompt-following, conceptual clarity, educational storytelling, editing, or readable text inside the image, Nano Banana Pro appears to be a strong choice.