Published: Oct 7, 2025 by Isaac Johnson
I started with this photo
I asked it to “restore this photograph, remove scratches, scuffs and marks. keep the composition and subjects the same. do not modify the subjects in the picture except to correct missing details. correct the yellowing on left of photo. enhance any blurry details”
The Nano Banana model did a decent job but did not really enhance the photo at all
Perhaps I can ask it to just enhance that above photo “This image is still blurry. Can you enhance it while maintaining subject consistency. add sharpness and make it as if taken with a DSLR camera”
It’s minor, but I do see some updates. I’m noticing my Schnauzer on the left is looking a bit strange though.
One thing I like to do is use MidJourney to then take the image and make a small video
We can see it ignored me about the existing dog and added a new one
For instance, that photo from last week:
I took and had it turn into a video of me stirring the pot
Here is an old blurry photo of me with my great grandfather Edwin
It was able to update it a bit
I’m not sure if the “reading a story to his grandchild” prompt worked perfectly. But it is interesting to see him in motion (so long ago I don’t have great memories)
I came back to Nano Banana to try one more. Edwin was a railroad conductor and I thought it might be nice to see a photo of that.
It doesn’t quite look like him though
Rob the Red
I have this wonderful AI generated photo of my former boss. I, personally, think it’s a majestic painting of someone who is a genuinely nice guy.
How might the new AI take the art from the older AI
The first draft looked nothing like my former boss. I asked the AI to keep the face and proportions consistent and I got a slightly better version
The second is a closer match
I pivoted and took his very professional LinkedIn photo and tried that. I had to go many iterations because words like “splashes of Red” (he is “Rob the Red”) and “battle axe” triggered safety warnings.
I didn’t like how his garments were too “RenFest”, not the real viking.
I asked for a refinement:
This is far more appropriate
I punched this into Midjourney, with the prompt: “he looks sternly at the camera. His eyes tell a story of heroic battles and long evenings battle the infamous “code red”. He is a warrior of a stock rarely seen.”
I also came back and asked for an animation of the modern take “he is ready to to chop some wood and you better get out of his way”
Headshots
We can also use this for headshots.
For instance, I can ask it to take a selfie from work and “Edit the man in this image: change his outfit to a fitted, professional suit. keep the original face details and hair stands to maintain realism and avoid an AI-generated look. Make his face look forward, body slightly straightened. Crop it as a half-body portrait. Neutral background with soft natural lighting. Photo-realistic Linkedin profile photo”
Funny enough it came back sideways
But I rotated it locally
and then used Gimp to get it completely straight
I also thought, perhaps I can just ask the model to fix. This looked worse - way more like someone is screwing with zoom
Maybe MidJourney can help. This turned more into nightmare fuel, IMHO
This might fit from when I was a bit younger and thinner
I’ll try once more, with weirdness turned all the way off and adding a second photo of myself for reference
Much better
This might even work - it’s close at least
Just for fun, I asked it to animate that image
Animations
Perhaps we want to make a nice Pixar character
Or Studio Ghibli…. seriously, why does Vertex keep making me sideways
I thought it might be fun to animate my Pixar self trying to take a selfie. Other than it suggesting i use an iPhone (wrong), it looked fun
or working on my laptop
I asked it to do the same thing without talking
extending it gets a bit odd as the second time around it added keys to the bottom of the laptop
Compared to MJ
Let’s compare some of these prompts we sent to Nano to what we get from MidJourney
First, me as a Pixar character
Perhaps in another 10 years
A re-run didn’t generate that much better results
I tried the headshot prompt next with the same “Edit the man in this image: change his outfit to a fitted, professional suit. keep the original face details and hair stands to maintain realism and avoid an AI-generated look. Make his face look forward, body slightly straightened. Crop it as a half-body portrait. Neutral background with soft natural lighting. Photo-realistic Linkedin profile photo”
None of them looked like me
Sometimes an “omni” reference helps anchor it, so I tried with another photo of myself as the Omni reference
Much better
This looks close
I’ll try an “auto” low-motion animation on it
Cost
What did all of this cost?
I intentionally avoided doing video in Vertex as I find it is really expensive. Sometime back I did a coffee shop intro video with a barista pouring some coffee using Vertex AI and i got to nearly $20 in spend.
So, we used lots of Nano Banana yesterday, but technically that is a Gemini Flash model:
so it might be cheaper:
Roughly $0.85 for everything and $0.68 for just yesterday. That’s a good deal.
Summary
We talked about using Nano Banana already with an MCP server last week.
Today, I really just wanted to explore what the model could do in a variety of situations: restoring photos, creating fun versions of people, updating prior AI art and doing studio portraits.