Built This Week/AI Tools

AI Tools

Nano Banana

Nano Banana is the image model we reach for when the subject has to stay the same across edits: product shots, costumed selfies and ad images, usually right on the first prompt.

Our verdict on Nano Banana

Nano Banana is Google's image model, and it is the one we use when an edit has to keep the subject the same. Sam called his product photos "extremely high quality" and said they were nearly always right on the first prompt. Its weak spot in our tests was faces: it keeps pose and hands, but not always your exact likeness.

In December 2025 Jordan said it had the best image generation of the largest models. By 2026 ChatGPT and Meta had caught up enough that neither of us could see a clear gap, so it is now one strong option rather than the only one.

What we built with Nano Banana

  • AI product images for a 3D-printed hoodie charm store (September 2025): Sam and his 11-year-old daughter 3D print hoodie charms for a store whose proceeds go to an orphanage. Every image on the site started as a phone photo of a charm. Nano Banana in Google AI Studio added the charm to a hoodie, then put the hoodie on a model. Each image took minutes.
  • Trick-or-treat AI costume builder (October 2025): Jordan's Replit app snaps a webcam photo, turns you into a pirate, ninja or anything you type with Nano Banana, then uses that image to start a Google Veo 3.1 Fast "Happy Halloween" video. The team used it heavily in its first 48 hours, and one teammate picked her real costume from it.
  • HipTrain Meta ad concepts (April 2026): Sam had Claude Design write three campaigns, each with an image prompt. He ran one prompt through Nano Banana and handed the image back to Claude Design, which composed the finished ad.
  • Episode backgrounds. Jordan generates a new artificial background for almost every episode, and for a long stretch Nano Banana made them.
So, basically, I'm just I print out the the little charms, take a picture, throw it into Google AI Studio, give it a quick prompt, add these charms to a yellow hoodie.

— Sam Nadler, Episode 12

How we use Nano Banana

  1. Start from a real photo. Nano Banana's strength is continuity, so give it the actual object or person. Sam photographs each printed charm; Jordan's app grabs a webcam frame.
  2. Use the chat in Google AI Studio for one-off images. Pick Nano Banana as the model and prompt in plain words. In Episode 38 (at 11:59) Jordan asked it for a clay figure of a black lab eating a hot dog at the park, from the same screen where he uses Gemini and Lyria 3.
  3. Build the scene in steps. First "add these charms to a yellow hoodie", then a second prompt that puts it on an 11-to-13-year-old girl with black pants and white shoes. Sam says he very rarely had to re-prompt.
  4. Clean up for the page. Remove the background if the rest of the site uses cut-outs. Sam left the charms oversized in his banner on purpose, because it makes it obvious what the store sells.
  5. For an app, call the API. Jordan served Nano Banana through Replicate, with his API keys stored in Replit. Expect about 20 to 40 seconds per image.
  6. Chain it into video. Use the Nano Banana image as the first frame of a Veo clip. That is the whole costume builder: one image call, one video call.
  7. Let another AI write the prompt. For ads, Sam asks Claude Design for the image prompt, pastes it into Nano Banana, then gives the image back to Claude Design.
And the nano banana generation is pretty fast. It's usually about like twenty to forty seconds or so, and I should become a ninja.

— Jordan Metzner, Episode 18

Where Nano Banana falls short

  • Likeness. In the costume app Jordan said of his ninja, "I don't know if those are exactly my eyes", even though the body and office were clearly his. On a later run the hand position was almost identical.
  • The gap has closed. Jordan made an April 2026 background with ChatGPT Images 2.0 and said he did not know if there was a significant difference from his Nano Banana backgrounds. In July 2026 Sam said Meta Muse Image felt pretty similar to Nano Banana.
  • Misuse. When Nano Banana launched, Jordan showed an edited photo of himself drinking a beer on a Zoom call and warned that this kind of continuity could be used dangerously.
  • What we did not test. We never covered pricing, rate limits or resolution on the show, so we can't speak to them.
Nano Banana has, like, the best image generation as far as the largest models on the market.

— Jordan Metzner, Episode 23

Nano Banana compared

  • ChatGPT vs Nano Banana: Jordan tried Images 2.0 for an episode background and saw no clear difference; he was underwhelmed by it.
  • Meta Muse Image vs Nano Banana: Sam found Muse pretty similar, with the twist that it can pull public Instagram photos of a person.
  • Before Nano Banana, Jordan's AI image editing chatbot chained edits on FLUX. Nano Banana's pitch was doing that same subject continuity inside Google's model.

Episodes featuring Nano Banana

  • Episode 10: recorded the day after launch; Jordan explains the continuity hype at 08:59.
  • Episode 12: Sam's hoodie charm product photos, demoed live at 13:27.
  • Episode 18: the Halloween costume builder with Nano Banana and Veo 3.1 Fast.
  • Episode 19: Nano Banana as one of the starter templates in AI Studio Build.
  • Episode 23: Jordan calls it the best image model of the big labs.
  • Episode 38: a quick Nano Banana generation inside Google AI Studio.
  • Episode 40: the HipTrain ad image, and Images 2.0 vs Nano Banana.
  • Episode 50: Sam compares Meta Muse Image.

FAQ

How do you use Nano Banana?

The easiest way we found is the Nano Banana chat inside Google AI Studio: upload a photo, give a short prompt like 'add these charms to a yellow hoodie', then prompt again to put the result on a model. For apps, Jordan called the Nano Banana API through Replicate from a Replit app.

Is Nano Banana good for e-commerce product photos?

In our experience, yes. Sam turned photos of 3D-printed charms into lifestyle product images for his daughter's store in minutes, and said he very rarely had to re-prompt.

How fast is Nano Banana?

In Jordan's Halloween costume app, each Nano Banana generation took about 20 to 40 seconds through the API.

Is Nano Banana better than ChatGPT Images 2.0?

Jordan generated an episode background with Images 2.0 in April 2026 and said he could not see a significant difference from the backgrounds he used to make with Nano Banana. He was a little underwhelmed by Images 2.0.

What is Nano Banana bad at?

Likeness. In the costume app it kept body pose and hands almost identical, but the face was not always Jordan's.