Christmas Cards, Dirtbikes, and the Oracle of Omaha (Part 2 of the How and Why of AI)

I was very excited to try Nano Banana the moment it came out, and here is an example of how you can use it. If you want a low-stakes way to have a ridiculous amount of fun this holiday season, go play with Google Gemini's image generation. That is the whole point of this post, and I am leading with it so you do not have to wait for the punchline.

A little backstory first. In 2006, I took a trip to Omaha with a group of my classmates from MIT Sloan and got to meet one of my heroes, Warren Buffett himself. That trip deserves its own post someday, but this is not that post. This one is about all the fun I had feeding real photos into Gemini's "Nano Banana" model and watching it build scenes that never actually happened.

I started with three real images. The first was a vintage picture of me standing next to the Oracle of Omaha, a photo I have hung onto for years.

The second was a shot of me in my favorite ugly Christmas sweater, which also happens to be my only ugly Christmas sweater. I landed it at our annual 4H white elephant party, and I was thrilled to finally own a proper Christmas-themed sweater of my own.

The third was a picture of my dog, because no holiday scene is complete without her.

Three real photos. That is all I gave it.

My first ask was simple. I told Nano Banana to combine the three into a Christmas card, something with Mr. Buffett, my dog, and me all in one frame. What came back looked like a photo we could have actually taken. The lighting matched, the faces held up, and my dog looked like she belonged there instead of being pasted in. If you have ever tried to wrangle three subjects into one holiday card, you know how much that would have cost me in real life.

Then I decided to go over the top. I asked it to make an image of Mr. Buffett and me riding dirtbikes together, both of us in our Christmas sweaters. This is a scene that will never happen, and yet there it was: two riders kicking up dirt, sweaters and all. It made me laugh out loud, which is a pretty good return on a few seconds of typing.

Here is what struck me most. The realism is the part that sneaks up on you. These are not obvious cartoons or clumsy cut-and-paste jobs. The model holds onto the details that make a face recognizable, keeps the lighting consistent, and fills in the surrounding scene in a way that feels plausible. It is the difference between a gag and something you might actually print.

A couple of practical notes if you want to try this yourself. Start with clear source photos, because the better your inputs, the better the output holds together. Be specific about the scene you want, then get playful once you see how well it handles the basics. And do not overthink the first few prompts. Half the fun is asking for something absurd and seeing whether it can pull it off. Spoiler: it usually can.

I will say the obvious thing about all of this too. It is worth keeping in mind that these images are fabricated. They are a blast to make and share with family, but they are inventions, not memories. I like knowing the difference, and I think that is part of what makes them fun rather than confusing.

If you have not played with Gemini image generation yet, this is my nudge. Grab a few photos, think of the most fun scene you can, and let it rip. Worst case, you get a laugh. Best case, you get a Christmas card of you and Warren Buffett on dirtbikes.