Hands on with Google Nano Banana
🧠How to use Google’s new model for product work. Storyboarding, UI mock-ups, visual assets for vibe coding, presentations and more. Plus, what is Google’s new Build Mode? #Knowledge Series 85
🔒The Knowledge Series is available for paid subscribers. Get full ongoing access to 80+ explainers and AI tutorials to grow your technical knowledge at work. New guides added every month.
Google has bombarded us with a flurry of new announcements over the past few weeks, so I thought it might be helpful to take some time to dig a little deeper into some of the ones that might be most helpful for product teams.
And the release generating the most noise over the past week is Google’s new Flash 2.5 image editing model, also known as Nano Banana. CEO Sundar Pichai cheekily announced the launch with a simple but straightforward tweet:
One commentator said this new model “made Stable Diffusion look like Microsoft Paint” and others have described it as “terrifying” - given just how powerful and realistic it is.
In this Knowledge Series, we’ll explore what this new model is, why it’s so powerful and most importantly, how you might be able to use it at work for practical things like storyboarding, creating visual assets you can use in vibe coded apps, sketching out concepts and diagrams, changing your LinkedIn professional head shots and more.
As well as this, we’ll also take a look at some of the other recent announcements from Google including its new “Build Mode” that lets you build and ship your own mini apps using the latest Gemini models.
Coming up:
What is Google Nano-Banana? A quick overview of the fundamentals and why it’s got people talking
Practical ways you can use this and other it at work including:
Storyboarding for user journeys / scenarios
Generating visual assets for vibe coded prototypes and UI mockups
Sketching out concepts and diagrams and learning about them together
Generating and editing your LinkedIn / professional head shots
Creating assets for use in presentations
How Google’s new “build mode” makes it easy to build and ship simple vibe coded apps. How to use Build Mode to build your own dedicated app that lets you create UI mockups in a few seconds.
What is Google Nano-Banana and why is it a big deal?
This is Google’s new Flash 2.5 Image model also known by the codename Nano-Banana. Announced last week, you can read more about it and its capabilities here, but it’s now the world’s leading image generating and editing model.
It's accessible through Google’s AI Studio, inside Gemini and now also through their API.
It beats other models on both image editing and generating and you can see its performance on editing in context here:
The model is ranked using the LLM Arena system - a ranking system that evaluates and ranks models by letting users compare the results of different models side by side. Users are shown the outputs from two different images ranked by two different models and asked to vote for their preference. So far, there are over 6 million votes included in the LLMArena so it’s a pretty dense collection of data on how models perform.
If you’re interested in digging deeper into some of the intricacies of Here’s the full links to each of the tests and their results:
Text to image - https://lmarena.ai/leaderboard/text-to-image
Image editing - https://lmarena.ai/leaderboard/image-edit
What Google is saying about it
DeepMind CEO Demis Hassabis has described the new model as “by far the best out there” with a model that “really excels at character consistency”. We’ll take a look at some of these capabilities in a bit more detail when we come onto the examples later.
For more context on how the model was built inside Google, this conversation with Google’s product and research leads from the Gemini team is definitely worth a watch. In it, they break down the technology behind its key capabilities, including interleaved generation for complex edits and new approaches to achieving character consistency and control options.
Why it’s got people talking
Aside from its overall performance, the model is garnering a lot of attention because of some specific new capabilities and details.
Firstly, it’s cheap. Gemini 2.5 Flash Image is priced at $30.00 per 1 million output tokens with each image being 1290 output tokens ($0.039 per image) - cheaper than rivals like DALL-E and Midjourney.
Secondly, it has what Google describes as “character consistency”. This refers to the model's ability to maintain the appearance and characteristics of an object or person across multiple different edits or scenes. In practice, this means you can ask the model to depict a person or object in multiple different places and the object or person will automatically adapt to fit its surroundings (more on that later).
This is possible in part thanks to the final new ability for this model: world knowledge. Nano-Banana is capable of understanding logical and cultural relationships - such as associating the “Mona Lisa” with the Louvre or incorporating up-to-date sports team colors, major brand logos, or current technology product designs into photo edits. It will use this world knowledge to ensure its creations are anchored in the real world, if necessary.
Practical ways to use the new image editing model at work
Before we look at some practical examples together, here’s some general guidelines on how to use it from Google’s AI product lead Logan Kilpatrick:
You can also check out the official prompting guide for engineers from Google for more details on how to prompt effectively.
Storyboarding for user journeys / scenarios
OK, now let’s get hands on with the model and start by putting its character consistency abilities to the test. At work, this could be particularly helpful for scenarios where you want to bring user research to life or articulate the jobs to be done for users.
In this case, I’ve uoloaded a simple reference image of my face and then prompted Nano-Banana in the Gemini Studio to see how well it depicts me in different scenarios.
When prepping for these examples, I wanted to purposely use smaller, shorter prompts to see how well it handled them. I took my headshot and simply asked it to put me in a coffee shop reading a book:
Put him in a coffee shop reading a bookThe result was pretty impressive:
Here’s how the model performed across different scenarios including reading a book, working in the office on a laptop and walking my dog in the street.
The only problem was that this isn’t my dog. So if I uploaded a photo of my actual dog, how well would the model cope with replacing the dog with my real dog?
Here’s my dog and here’s how the model performed when asked to replace it:
This took a bit of fiddling and I had to be pretty precise with my prompts to ensure the model knew which images I was referencing. Google’s guidelines suggest numbering your images in your prompt so that the model can reference them by number so I asked it to replace the dog in the first image with the dog in the second image.
Overall, it does a very good job - although there is some slight degradation in the quality of the image after a few edits which is a little disappointing given that model is being hyped based on its ability to preserve quality over multiple edits. The character consistency, though, is excellent, with the model able to put the character in various different scenarios with ease.
How to use this for storyboarding / bringing user journeys to life
Since it’s super easy to take one character and put them into different scenarios, this could be very helpful in cases where you’re building a storyboard of a typical day in the life of a target user and need to articulate that visually in presentations or elsewhere.
Imagine a Jobs to be Done board brought to life using these types of image edits, for example.
If there are specific scenarios that you want to really bring to life, you could also combine this with Google’s Veo 3 video generation models to generate video depictions of the same scenarios.
Generating visual assets for prototypes and UI mockups
Next, let’s take a look at how you can use Nano-Banana for crafting assets you can use in prototypes and UI mockups.
For this, we’ll use Apple Weather and ask the model to first make some skeuomorphic design changes - and then put these in-situ in an iPhone. These types of assets are helpful if you’re building a landing page or generating a vibe coded app that needs its own assets.
We’ll start with the standard UI for Apple Weather: