When I decided to test using Gemini Nano Banana Pro 3 to create still images for a project on patient care, I was expecting a great deal of detial — but what truly stood out was the consistency. I started with one main nurse character from a text prompt and was able to place her in a series of hospital settings: behind the nurses’ station, consulting with a patient, and leading a small team discussion.
I worked with Nicholas Janes from Duke’s Center For Teaching And Learning on the initial idea for the testing of Nano Banana 3 Pro. Nick provided this prompt:
A detailed, natural-style illustration of a Black woman in her 30s working as a charge nurse coordinating care in a hospital unit. She stands at the central nurses’ station, calmly overseeing patient care and communicating with her team. Her body language conveys confident leadership and empathy as she reviews patient information on a computer monitor and speaks with a colleague. Around her, diverse healthcare workers — nurses, a respiratory therapist, and a provider — move purposefully in the background, showing teamwork and coordination. The charge nurse wears modern hospital scrubs and an ID badge. The environment includes a softly lit nurses’ station with digital patient boards, charts, and subtle reflections of clinical lighting. The atmosphere feels warm, professional, and human. The art style should be natural and expressive — detailed and realistic in emotion and setting, but not hyper-realistic or photographic.
I decided to change the initial prompt but more importantly, I decided to add a photographic reference. Here is the updated Nano Banana Prompt:
A detailed, natural-style illustration of a Black woman in her 30s working as a charge nurse coordinating care in a hospital unit. She stands at the central nurses’ station, calmly overseeing patient care and communicating with her team. Her body language conveys confident leadership and empathy as she reviews patient information on a computer monitor and speaks with a colleague. Around her, diverse healthcare workers — nurses, a respiratory therapist, and a provider — move purposefully in the background, showing teamwork and coordination. The charge nurse wears modern hospital scrubs and an ID badge. The environment includes a softly lit nurses’ station with digital patient boards, charts, and subtle reflections of clinical lighting. The atmosphere feels warm, professional, and human. The art style should be abstract and water color
Using a photographic reference raised an important ethical question for me:
Is it appropriate to use a real photograph as a basis for an AI-generated image? Even when an AI reinterprets a scene, using copyrighted or personal photographs as input references can raise concerns about consent and ownership.
To create responsibly, it might be best to use:
- Royalty-free or Creative Commons–licensed images,
- Your own photography, or
- Prior AI-generated work you’ve already made.
Balancing technological possibility with ethical integrity ensures that real people are represented with respect, even in digital re-creations. Questions remain that the models themselves could be considered unethical since we do not know anything about their training data (foreshadowing)
Example Series: Gemini Nano Banana Pro 3
We were successful using the previously created AI image as the reference for a newly created scene. We took that first image and used this super simple prompt to generate the next image:
Create the same nurse at the bedside of an Asian American speaking to the patient.
The main character’s detail (even their ID card) was replicated extremely well in a new pose and location. The water color details were great and the location seemed to belong in the same hospital.
I took that image and used this prompt:
Now she is in ground rounds. Please include a diverse group of nursing students and nurses discussing a different case at the bedside.
Although we couldn’t see the ID card, it was similar, the bed and room were similar and all of the other people in the scene were well focused. Really well done.
The third prompt was equally well done:
Our same nurse now is in the commissary, taking a rare moment of reflection and having a cup of coffee. Her mood is not sad or happy, just circumspect and having some downtime to recharge her batteries. The commissary is mostly empty and it is night time you can see the window.
ChatGPT Version (Generated from the Same Prompt)
Next, I took one of our Gemini-generated images and asked ChatGPT’s image tool (v5) to create the same scene using the identical prompt.
I wouldn’t expect the two images to be the same, but I was surprised at how well and close her hair and overall look rendered. Stylistically, they are very different. It’s completely subjective but I felt that Nano Banana was more to my liking while ChatGPT added some details that I felt were less professional like the steam in the coffee cup and the moon. I think it would be difficult to compare the images objectively but would love to hear some comments.
I then used the new ChatGPT 5 image as reference and used the same prompt from Nano Banana:
Now she is in ground rounds. Please include a diverse group of nursing students and nurses discussing a different case at the bedside.
This floored me. Look at all of the similarities in this image!
This similarity across two entirely different AI systems is remarkable.
It doesn’t mean the models share data — they don’t — but it reveals something fascinating about how these systems learn.
Both Gemini and ChatGPT are trained on large datasets that teach them common visual associations — for example, what “a nurse leading a hospital team meeting by a patient’s bedside” usually looks like.
So when both are given that same concept, they converge on a very similar visual interpretation.
In other words, they’ve learned the same visual language of healthcare.
It’s not duplication — it’s convergence through shared cultural and visual patterns.
Gemini maintained a consistent, painterly style across all its images, while ChatGPT interpreted the same scene with a slightly different tone and lighting, showing how each model carries its own “creative fingerprint.”
✦ Observations and Takeaways
-
Gemini Nano Banana Pro 3 excels at maintaining visual consistency and a unified art style across multiple images.
-
ChatGPT’s image generation offers stylistic reinterpretation and flexibility in tone.
-
Both tools demonstrate how AI can create cohesive, human-centered visuals for storytelling and education.
-
Their similarity highlights a broader truth: as AI systems evolve, they’re beginning to interpret the world in remarkably consistent visual terms.
The real opportunity — and responsibility — for creators is to use these tools thoughtfully, pairing innovation with ethical awareness.
Working across both Gemini and ChatGPT revealed not just what AI can do, but what it means to use it responsibly.
When two unrelated systems produce nearly identical, high-quality visuals from the same prompt, it’s a testament to how advanced and standardized our visual language has become.
It’s also a reminder that while technology is rapidly evolving, our creative and ethical judgment must evolve alongside it.
