Is this the beginning of the end for Photoshop, stock photography, and traditional ad shoots?
In this episode of Leveraging AI, Isar Meitis dives headfirst into the wildly capable new Gemini 2.5 Flash image generation tool, also called “Nano Banana.”
Here’s the twist: It's not just about creating images. It's about building scalable, repeatable, hyper-custom content pipelines that save weeks of work and thousands in budget with zero design skills required.
In this session, you'll discover:
- What exactly is Gemini 2.5 Flash (Nano Banana) and how it differs from ChatGPT or Midjourney
- How to create realistic, brand-consistent images from nothing more than a sketch and a sentence
- Why image consistency, layering, and template reusability are a game-changer for marketing teams
- How to transform product shots, team headshots, and social ad campaigns in minutes
- The “dangerously easy” way anyone can now create deepfakes and why that’s a double-edged sword
- Real-world, business-relevant use cases: from eCommerce to real estate, design, ad testing, and product mockups
- How to scale image creation with workflows using tools like Weavy and upscaling solutions
- Why this is the beginning of a major disruption in content production and how you can stay ahead
- What’s still missing: resolution limits, lack of layers, and where AI tooling must evolve next
About Leveraging AI
- The Ultimate AI Course for Business People: https://multiplai.ai/ai-course/
- YouTube Full Episodes: https://www.youtube.com/@Multiplai_AI/
- Connect with Isar Meitis: https://www.linkedin.com/in/isarmeitis/
- Join our Live Sessions, AI Hangouts and newsletter: https://services.multiplai.ai/events
If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
GMT20250901-193429_Record (00:00):
Hello and welcome to the Leveraging AI
(00:01):
Podcast, a podcast that sharespractical, ethical ways to
leverage AI to improveefficiency, grow your business,
and advance your career.
This is Isar Meitis, your host,and in the past week and a half
the world, or at least the AIworld, has gone bananas.
And the reason it has gonebananas it's because Google has
introduced Gemini 2.5 flashimage generation tool, which
they initially launched on theLM CS arena under the brand name
(00:25):
nano banana.
So everybody was going crazywith what this model can do.
Now we know it's Gemini 2.5flash image generation
capability, and this model cando incredible things.
Now, after doing a very shortreview of what it can do on this
past weekend's news episode, Idecided I have to dive in and
actually show you different usecases that are absolutely mind
(00:47):
blowing.
So let's start with what are thecapabilities of this model and
why is it so exciting?
And then we're gonna dive intothe actual details and examples
that I, some of them are minethat I'm gonna show you exactly
how they're done.
Some of them are just outputs ofmultiple people from the web
that are sharing their results.
mAny of them are way higher thanwhat I know how to do, or I had
the time to do in thisparticular case.
(01:08):
But let's get started withcapabilities.
So the first and most importantaspect of this model is that
it's incredible at keepingconsistency of everything on the
image.
So it can keep the consistencyof the background, it can keep
the consistency of object, itcan keep the consistency of
faces, people, modes,interactions, themes, styles,
(01:29):
literally anything you want, youcan take from an image and apply
to another, merge, two imagestogether and so on.
The other thing, talking aboutmerging things together, it is
very good at multi-image fusion.
I will show you multipleexamples of this.
You can combine differentobjects, you can combine objects
and scenes and background.
You can combine literallyanything you can imagine.
(01:49):
Now, the other thing that youcan do is you can do local
edits.
Either adding, removing,changing, more or less anything
in an existing image.
This image could be generated byAI or generated by an actual
image, meaning some picture youtook of something, it doesn't
really matter.
You can make changes to it.
When I say make changes, you canremove people from an image that
somebody photobombed yourpicture or you have a picture of
(02:11):
yourself with the alpha tower inthe background, and there's a
lot of other people there, butyou don't want them in the
image.
You can have them all removed.
You can, by the way, do theother way around.
You can take a picture of yourfamily and then add yourself to
the image that even thoughyou're the one that worked
behind the camera, taking thepicture, you can upload a
picture of yourself with orwithout the same exact stuff
that you were wearing that day,and then ask Gemini to add you
(02:33):
to that image.
The other thing that you can dois you can create a template and
then use it consistently acrossmultiple examples.
So this is good for creatingcards, badges, mockups of ads,
et cetera.
Literally any template that youwant, you can use as a template
and then reuse it again andagain and again in multiple
options and differentimplementations.
Again, I'll show you severaldifferent examples of that.
(02:54):
World knowledge, image reasoningis huge.
There's two aspects of that.
One, it is because it's athinking multimodal AI model and
not just an image generator.
It understands the world, whichmeans on one hand it understands
physics and how things interactwith one another.
So when you ask it to create animage of something, it'll make
sense in the physical worldunless you ask it not to.
(03:15):
And the other aspect of this isthat it understands business and
flow charts and so on.
So if you are now in a room, ina conversation and you have a
whiteboard and you're sketchingand drawing things, you can take
a picture of that.
Let's say there's a flow chartdiagram and you can ask it to
design the business flow andshow it in a graphical way.
(03:35):
It can do that because itunderstands everything that you
have created and you canrecreate it in a more accurate
way.
You can also, because it's alarge language model, you can
continue to have a conversationwith it to continue to edit and
manipulate the image step bystep by step, which you
obviously cannot do if you areusing a tool like Me Journey or
just a standard image generator.
(03:56):
Now, in theory, you could dothat because you can take a
screenshot of the image, enterit as a starting point, and then
try to edit from there.
But it's clunky and it'sdefinitely not as coherent and
cohesive as it is doing it innano banana.
From this perspective, it's verysimilar to using the new-ish
image generation model fromChatGPT that was released
earlier this year and again tookeverybody by a storm.
(04:18):
This model is also really goodat generating text on images,
but because it's also very goodin consistency, it can apply
your logo or anybody else's logofor that matter, to either a
banner that you're creating orput it on top of whatever object
that is in the image because itunderstands, again, physics.
So if there's a cup and you putyour logo on the cup, it will
(04:38):
render it as if it's in acircular way, wrapping around
the cup, et cetera, et cetera.
T-shirts, glasses, anything youwanna put this on will be
generated in the correctgraphics of the logo, but also
in the correct 3D applicationwhere you want to apply this.
Now, as I mentioned, you canalso transfer style.
So you can take a style from animage, a style from a type of
(04:58):
cloth, a style from literallyanything and apply to anything
else.
It is very good at maintainingpose and applying wardrobe
changes.
So if you want to see howsomething fits on you, you can
take a picture of you, take apicture of that piece of
clothing or accessory and youcan apply it to you and see how
exactly it is going to looklike.
I will show you multipleexamples of this as well, and
(05:21):
you can play around with thiswith different things.
You can re colorize and improveany kind of image.
So let's say you have abeautiful neighborhood photo and
you want to change the color ofthe buildings from blue to red.
You can do that, but moreimportantly, you can take old
photos that are either black andwhite and grainy, or not even
(05:41):
black or white, but just are inbad shape and use nano bananas
to make it significantly better.
There are multiple examplesonline right now for that, and
it's absolutely mind blowing.
It is on one hand, not as fastas some of the really fast image
generation models, like thefaster mode in mid journey, but
it is significantly faster thancreating images in chat pt,
(06:02):
which is the other option wehave if you want to have a
conversation and have a realworld understanding and include
that as part of a biggerprocess.
What you do need to be aware ofis that when you're creating
images with.
Nano bananas.
It will put two kind ofwatermarks on the image.
One is a Gemini logo on thebottom right corner, and the
other one is a synth id,invisible watermark, which is
(06:24):
basically a hidden pattern intothe actual pixels that Google
tools know how identify as toolsthat were created by ai.
I don't necessarily agree withthe visual watermark.
I think it's kind of stupid,especially that you can crop it
out in about two seconds.
But I love the fact that theyfinally, including synth ID in
the image generation aspect ofit.
(06:44):
I really hope that the worldwill come to an agreement on a
standard.
I don't care if it's synth ID oranything else of image and video
generation tools that generatestuff with AI that you'll be
able to, without any effort,hopefully with just a plugin on
your computer to know everyimage on every video, whether
they're generated with AI ornot.
So I really hope we're gonna goin that direction.
(07:05):
Now, let's start looking atmultiple examples and use cases,
and I'm going to share my screenand I'm going to explain
everything that is on the screenbecause I know you're right now
either walking your dog ordriving a car or jogging on a
treadmill or whatever it is thatyou're doing right now,
listening to this podcast.
So I will explain everythingthat is on the screen, but
because this particular episodeis extremely visual, I suggest
(07:27):
that afterwards when you havetime, click on the link in the
show notes of this episode, goto our YouTube channel and watch
the YouTube segments at leastjump to the stuff that you're
really interested in so youdon't have to watch the whole
thing again.
That doesn't make any sense, butat least you can jump to the
parts that you wanna see exactlywhat I was demonstrating.
Okay, so let me share my screenand let's dive into a bunch of
(07:49):
examples.
GMT20250901-210806_Recording (07:50):
So let's start by looking where to
get to this new magical modelthat you can use with inside of
Gemini.
So first of all, you just loginto your Gemini account and you
can see I have 2.5 pro up here,in the model selector.
I actually don't need that.
All you have to do is go towhere your prompt is.
There's a section for tools.
It has all the previous optionsthat were there, like deep
research, create videos withvideo canvas and guided
(08:10):
learning.
But you can see there's a newoption called Create Images with
a little banana image next toit.
You click on that.
And then you can start creatingnew images.
To be fair, you don't even haveto do that if you ask it to
create an image or if you dragan image in there for editing.
We're gonna talk about thatlater on.
It will know what it needs todo, and it will know how to run
this model either way.
But if you wanna be safe, youcan choose it from here.
The other option to get to it isthrough google AI Studio.
(08:32):
So if you are in ai,studio.Google.ai, there's all
these amazing options likestreaming live and so on.
But if you go to the regularchat, there's a button here that
says Try Nano banana.
And you click on that and nowyou can literally write your
prompt, edit your images, uploadstuff, whatever you want do, and
run it from here.
The biggest difference is ifyou're running it in Google AI
Studio, you are paying fortokens.
So you're paying a really smallamount of money every time you
(08:53):
run any command in here in AIStudio in general, including
creating and editing images.
And it is included with yourlicense if you're running it in
your regular Gemini account.
So my suggestion to you is startin the regular Gemini account
and if you run out of creditsand you need more, then go to
the air studio and just keep onrunning it over there and over
there you can run it foreverbecause you're paying for every
(09:13):
token.
GMT20250901-193429_Recordin (09:14):
The first few examples I wanna show
you come from Google themselves.
It's showing differentimplementation that they have
shared with the world.
The reason it's cool to look atis because first of all, it's
Google themselves.
They know exactly what the modelcan do.
But the other thing is it has alot of explanations of how to
actually use the modelseffectively.
So the first one is from theirdevelopers.google blog.com, and
we will share the results tothat.
(09:35):
And this article is called Howto Prompt Gemini 2.5 slash Image
Generation for the Best Results.
And if I scroll down here.
You'll see.
How does this look like?
So you see a beautiful image ofan all Chinese guy holding a
piece of pottery that he hascreated.
The image looks completelyphotorealistic.
He's in his studio.
The background looks completelyreal.
The lighting is perfect, andthey're showing you a few
(09:57):
things.
They're showing you what is theprompt, and they're showing you
what is the template of theprompt.
So I will give you an exampleabout this one.
The prompt itself is aphotorealistic closeup portrait
of an elderly Japaneseceramicist with deep sun edged
wrinkles and a warm knowingsmile.
He's carefully inspecting afreshly glazed T-ball.\ The
(10:18):
setting is his rustic sun drenchworkshop.
The scene is illuminated by softgolden hour light streaming
through the window, highlightingthe fine texture of the clay
captured with an 85 millimeterportrait lens, resulting in a
soft, blurred background inparenthesis, bouquet.
The overall mood is serene andmasterful.
(10:38):
Vertical portrait orientation.
And again, the overall image isperfectly coherent with a very
long and highly detailed prompt,and they're giving it as a
template.
So the template is a photorealistic.
And then what is the shot typeof, what is the subject next
action or expression.
Next set in whatever environmentthe scene is illuminated by
(11:02):
liking, description, creating amood.
Atmosphere capturing with a whattype of lens details.
And you can skip that if youdon't know that.
You can just explain what kindof look you want for the image
emphasizing, and then keytexture and details.
The image should be in whateveraspect ratio you want, so you
(11:23):
can apply this to literallyanything you need.
So this is just one example.
A very different example isthere's a cool red panda sticker
that looks perfect as a sticker.
It's 2D.
It's very, very different thanthe previous image.
And here the example prompt is akawai style sticker, a happy red
panda, wearing a tiny bamboohat.
(11:43):
It is munching on a green bambooleaf.
The design feature, bold cleanoutlines, simple cell shading
and vibrant color palette.
The background must be white.
And it looks like a perfectsticker that is aligned exactly
with that.
And again, they have thetemplate, which I'm not going to
read on how to create stickers,and then they give even more and
more examples on how to createlogos and how to create a
(12:05):
perfect image of a product photoshoot.
In this particular case, it's ablack mug on a granite style
kitchen countertop with fumescoming out of it.
And then they have aminimalistic view of a white
background with a delicate redmaple leaf in the corner.
And again, the example prompt isa minimalist composition
featuring a single delicate redmaple leaf positioned in the
(12:27):
bottom right of the frame, etcetera, et cetera.
So very different from all theother ones.
Another great example is a verydark noir art style comic.
Panel out of a comic sketch andit is perfect.
It has this guy in a long, darkcoat with a hat.
It looks very shady in like analley.
(12:48):
There's a place in the back thatis called the Black cat.
It is raining, but it's all inlike these black and white
sketches.
It looks absolutely incredible.
And again, same kind of thing.
They're giving you the exactprompt and the exact template to
create the same kind of thing.
Then they're showing you thatyou can edit images.
So the first example they giveis there's an image of a cat on
the left, and the prompt isvery, very simple.
Using the provided image of mycat, please add a small knitted
(13:10):
wizard hat on its head.
Make it look like it's sittingcomfortably and matches the soft
lighting of the photo.
And the second image looksexactly like the first image,
only the cat has a littleneeded.
Wizard hat on its head.
His ears are seeking out andit's really, really cute.
But everything else in the imagestays exactly the same.
(13:31):
The next example is they'reshowing a beautiful contemporary
living room with a really bigblue sofa.
And in this particular case,you're using the prompt using
the provided image of a livingroom change only the blue sofa
to be a vintage brown leatherChesterfield sofa.
Keep the rest of the room,including the pillows on the
sofa and the lighting unchangedand it works.
(13:52):
Now it's.
Again, absolutely amazing.
Everything in the room staysexactly the same.
There are multiple details inthis room.
There are lights in the ceiling.
There are shelves on the rightwith multiple Tcho keys on them
in this contemporary room, andnothing else changes, including
the view from the window otherthan the solf itself.
This is nothing short ofmagical, and they're showing
multiple examples.
Another really interesting onethat I actually did an
(14:13):
experiment of my own, andactually took it a little
further, is what I told youbefore, is the ability to dress
somebody.
So they took two inputs.
They took an input of a dress,just a flat image of the dress
without anybody.
It's just the dress itself andan image of a model that is
standing in a studio wearingblack pants and a white sweater.
(14:33):
And here the prompt is create aprofessional e-commerce fashion
photo.
Take the blue floral dress fromthe first image Let the woman
from the second image wear it,generate a realistic, full body
shot of the woman wearing thedress with a lighting and a
shadow adjusted to match theoutdoor environment, and the
image is absolutely perfect.
You can see the same Molly, it'sthe same face, same hairstyle, a
(14:55):
different pose.
It looks as if she's walkingquietly in a little alley in
Europe.
The lighting is perfect.
It's absolutely amazing.
And she's wearing the exactdress from the flat image, so
you can use this either ifyou're selling products or if
you want to try on products, youcan do that as well.
Now I wanna show you my littleattempt on the same kind of
thing.
And I took it to anotherextreme.
(15:16):
I thought about, okay, if Iwanna sell a dress and I wanna
do a cool campaign with it, howabout dressing famous women's
statues with that dress?
So I have two pictures here.
One is a picture of the statueof the woman in front of the
ferry building in San Francisco.
And I have a picture of a dressand the picture of that statue
of the woman in San Francisco.
(15:37):
And literally all I wrote isdress the statue with the dress
in the other picture really,really short.
And I got the exact output thatI wanted.
You can see the statue, theferry building stays the same.
The people walking around andsitting next to the statue is
the same.
Everything is the same, but thestatue is wearing the dress.
Then I wanted to get fancy.
I uploaded a picture of theStatue of Liberty, which is kind
of strange because it's alreadydressed in a gown or a dress or
(15:58):
a toga.
I'm not sure exactly what she'swearing.
But I asked the same question.
I uploaded the dress again.
I said, make the Statue ofLiberty.
Wear the attached dress insteadof its current cloth.
What it did is actually really,really cool.
It wrapped the Statue of Libertywith the same fabric and kinda
like applied it to the currentthing the Statue of Liberty is
wearing.
This is really, really cleverbecause I wasn't exactly sure
how this is gonna work out.
And it actually worked out veryinterestingly.
(16:20):
So this is just my example ofthat.
So let's look at a different setof Google examples.
So in this particular case,they're showing different sets
of double photos and how theycan be edited in order to apply
new aspects or new anything tothe same images.
They're all really, reallystunning.
So let's look at them one byone.
The first one has a input imageof a girl with her face really
(16:43):
dirty in a spacesuit, and thesecond image is a closer closeup
of her without the helmet.
So just taking the helmet off.
The next one is taking a sideshot of a redhead girl in a
truck, driving in an openterrain space somewhere, and
then it applies differentbackgrounds to the same picture,
so snowy mountains, et cetera,whatever you want.
(17:03):
The girl stays the same, hergaze, the lighting, everything
stays the same.
The scenery outside changes.
Another example is that theyhave.
Is taking two little funnylittle cartoonish characters and
then placing them in a recordingstudio and they're actually
operating the controls asthey're recording somebody else.
One of them have headphones onand so on, and they have the
prompts to each and every one ofthose.
(17:24):
I'm just sharing with you, youwhat you can do.
And there's really amazingimages of different kinds of
photos.
They have a photo of a girl onthe left and then multiple
images of her in Polaroidwearing different things in
different hail styles on theright.
Again, absolutely accurate andamazing.
And then an underwater sceneand.
Literally anything you canimagine.
Here's another example ofmerging two images, so a
(17:46):
professional photography imageof a woman and then a guy, and
then the joint photo.
They are hugging and lookingvery tenderly at each other in a
street in the snow with bouquetlighting in the background.
It's a perfect image, but it'sexactly the same people from the
two other images.
So these are just differentexamples of what you can do with
this.
The next one is really, reallycool.
They have a picture of a youngguy, kind like looking towards
(18:10):
the camera, and then they havedifferent professions for the
same guy all created just withproms.
So in the first one, he'swearing doctor scrubs and he has
stethoscope on his neck.
In the second one, he's ateacher in a classroom.
In the third one, he's workingon a clay statue, and in the
fourth one, he's preparing doughin a bakery.
All of them look exactly likethe guy only in completely
(18:30):
different settings withcompletely different poses, but
it is very clearly the same guyagain, using consistency.
So these are these kind ofexamples.
Then they have an example of howto turn a dog that is an actual
real dog into a game characterin a game style that looks like
Mario.
Then they're showing anothercartoon character.
In this particular case, it's alittle dinosaur that looks like
it's made out of plastic, andthen they're dressing it up in
(18:51):
multiple different kind ofuniforms as a superhero, as a
pirate, as a cheek and so on.
Absolutely incredible.
Then they're showing how you cancombine different styles and
aspects of different photos.
So they're taking a photo of awoman wearing an orange piece of
clothing and a photo of a pieceof glass floating in midair, and
they're showing the woman theoutput is the woman floating in
(19:12):
this bubble of glass in midair,and it's very impressive.
They took a picture of whalesjumping in the water and a
picture of a snowy mountain withthe lighting on top in the
sunset and combine the two whereyou can see the mountain in the
background as the whales arejumping in the ocean.
They took a picture of a forkand a picture of spaghetti and
created a fork made out ofspaghetti just with a single
(19:33):
prompt and so on and so forth.
Really, really amazing.
And then they're showing how youcan clean images and make
changes to images.
So there's an image of an imageof a building with very distinct
color patterns on it.
In several different segments ofthe pattern are in blue, and the
prompt is literally change allthe blue to green and you get
the same exact image, with thesame model standing in the
(19:54):
middle, yet wearing her yellowcloth just with a green
background instead of blue.
They have an image of a guywalking towards a truck,
probably somewhere in theMidwest.
There's a crazy tornado in thebackground.
And in the second image, it's abeautiful evening with no bad
weather whatsoever.
But everything else is exactlythe same.
They have an image of a housewhere the next four photos are
(20:14):
the same exact house, but infour different seasons.
And you can very clearly see howthe house looks like when it's
covered in snow, how it is whenthe leaves are changing, and so
on and so forth.
The next example that they'regiving is how to push design and
use existing ideas in order toapply them to something else.
So I have a picture of abutterfly, and they're creating
a picture of a model with abutterfly dress that is
completely inspired by thepattern on the butterfly's
(20:37):
wings.
They have a picture of a livingroom that is very nicely done,
and then they're adding a colorpalette, literally just five
different colors, and they'reapplying it to the cushions on
the sofa to give some colors tothe room.
Again, everything else stays thesame.
Then they're taking a face of aperson and applying it as a
cartoonish style on the side ofa cereal box and so on and so
(20:58):
forth.
Taking something from one imageand applying it to a different
image in multiple aspects.
Let's look at a few moreexamples of things that I did.
So I found this thing on x.com.
And I hope this is the originalI was really trying to find,
because these examples now areeverywhere.
I was trying to find theoriginal creator so I can give
them credit.
So in this particular case, it'sZoe, I hope I'm pronouncing his
(21:20):
name correctly.
And his handle in X is Zoe.
Z.
Z Zo.
So Zo, Z, Z.
And what he did is he took asketch of a figure.
The way you start, when youstart creating a painting of
something, especially whenyou're trying to create an image
of a superhero or something likethat.
sO just the outlines of the bodyIn a very specific pose, and he
(21:40):
took an image of a Japanesemodel and he asked to apply the
model to the pose in thesketched character, and he did a
perfect job.
It is absolutely mind blowingbecause the girl is the same
exact girl.
She's wearing the same exactstuff she's wearing in the image
and she's standing or kind oflike crouching in the same exact
(22:00):
pose as the sketched figure issitting at.
The other really incrediblething is that the, the original
image of the model, you cannotsee the bottom of her feet, so
it had to guess what she'swearing.
Give it the same kind of styleof the other stuff that she's
wearing.
And it's actually done a verygood job of doing this.
And I was interested was that,how many attempts it took to
(22:20):
create that?
So what I did is I took theimage on the left, basically the
sketched outline of thesuperhero pose and together with
the image of the girl on the topright.
And I asked Gemini and I droppedit into Gemini and I wrote the
following prompt, remove theimage of the girl from the top
right corner and then.
We already learned that that'sdoable.
I can remove stuff from theimage.
So I got a image of just thesketched outline of the
(22:43):
superhero pose, and then Iuploaded a selfie image of more
or less just my head andshoulders.
That's it.
And I said, now create a photorealistic studio photo shoot of
the guy in the image, in thesame exact pose as the action
figure sketch white background.
And I got a perfect pose of me.
And again, you had to guesseverything that I'm wearing
(23:04):
because all you see is myshoulders with an off white
sweater.
So it applied the sweater as, asif I'm wearing it tied around my
neck.
And it gave me very stylishclothing that I will never
actually wear because I'm not astylish kind of guy.
I wear t-shirts and shorts mostof the time, but the image is
perfect and it's totally me,like the face is me.
The pose could have been me ifwould've been that flexible.
(23:24):
Uh, but it's very, veryimpressive.
And then I said, okay, now dresshim up as Spider-Man, but
without the mask.
So now I have the same exactimage because it's a superhero
kind of pose with me instead ofwearing the clothes I was
wearing before or that I'dinvented for me dressed as
Spider-Man, but it's still myface, still the same pose.
And then I said, okay, place himon top of a skyscraper at night
in a medium shot.
And it did that as well.
(23:46):
It Even placed the light fromthe moon in a very specific
angle that drops shadow from thepose that I'm standing on the
rooftop.
And then you can go on and onand on and build an entire
superhero story from that oneoriginal shot.
So it's one line of text andthese magical outputs just come
out.
Let's look at another examplefrom X.
This creator from X created asketch on a piece of paper and
(24:09):
took a picture of it.
The sketch is of a modelstanding and stretching up.
It is the kind of sketch youwould see from clothing
designers create when they startthinking and sketching the ideas
for their next clothes.
So it's this girl who'sstretching up and you can see
that she's wearing Calvin Kleinunderwear and you can see what
kind of shirt she's wearing, butyou can't really see the girl.
And with one prompt, he createdan image of an actual model in
(24:32):
the same exact pose, wearing theexact clothe that he was
imagining in his sketch.
This is absolutely insane.
It's going from an idea topaper, old school way, going
from that to a photo shoot,fully realistic image of the
model in the same exactscenario.
This is absolutely mind blowing.
Now let's look at combiningimages.
(24:52):
I was trying something very coolhere.
So what I did is I looked onlinefor standard images of different
ingredients.
I found most of them on websitesof different grocery shops.
So I have an image of a avocadosliced.
I have an image of corn.
I have an image of sour cream.
I have an image of chickenthighs wrapped in a frozen
chicken thigh box.
(25:12):
I have cilantro, I havetomatoes, and I have tortillas.
Each and every one of them is astandalone image that I got from
the web with no background ofeach and every one of these
ingredients.
And then all I wrote is create apromotional social media image
of a dish I can make from allthese ingredients.
And it created this beautiful,incredible image of a modern
(25:32):
kitchen that is blurry in thebackground.
There's a gray countertop on topof that, there is a wooden
cutting board.
And on top of that, there isthis perfect deliciously looking
chicken wrap.
And the ad says, dinner solved,exclamation point, delicious
chicken wraps made with freshingredients.
It is absolutely amazing.
Then I ask it to change theaspect ratio of the image, and
(25:54):
it failed.
And I must admit, I triedchanging the aspect ratio of
images on Gemini multiple times,and it fails every single time.
Now it might be me personally,I've seen a lot of other people
do it.
When I asked Gemini, can you dothat?
It said, yes, absolutely.
But every time I tried it, itfailed.
I tried it in multiple scenarioswith multiple images.
It never worked for me.
I assumed this is something theywill solve soon.
(26:14):
Then I was trying to create aneven funnier ad.
So I was playing around withdifferent Yeti images in the
end, what I got to is there's aYeti in a bus with multiple
other people next to the Yeti.
Everybody's eating chickenwraps.
The Yeti is all dirty with thesour cream dripping all over his
fur.
And the ad is sayingunapologetically delicious, and
on the bottom, in smaller frontit says chicken wrap.
(26:36):
So tasty.
You do not care.
Getting messy in public, it is agreat funny ad that I went from
ingredients to an ad I canprobably use in about six to
seven minutes.
This would've taken the old way,months, or at least weeks to
create between coming up withthe ideas, doing the shoot of
the actual wrap, et cetera, etcetera, et cetera.
(26:56):
I can create now a whole seriesof ads based on the same style,
based on the same comedyconcepts, and just keep on
creating them right here in thesame chat.
Another example of usingmultiple images to achieve a
goal is I took a selfie ofmyself with a very colorful
background, really messy, and Itook a screenshot outta my brand
guidelines of a transition fromdark blue to light blue that I
(27:18):
use in different aspects of myproducts and my websites and so
on.
And I took a screenshot of mylogo and I asked the AI to
create a professional headshotof this person.
Keep the image aspect ratio ofthe person as is, change the
background to the colortransition in the same color of
blue as your attachment on thetop left corner, add a small
version of the logo that isattached.
(27:40):
The person should be wearing awhite buttoned up shirt with a
blazer and no tie.
And other than the aspect ratioaspect of it, which again we
already covered it created abeautiful, amazing professional
photo shoot of myself with thetransitioning color behind me
with the logo.
Perfectly done on the top leftcorner.
This is amazing.
This is absolutely incredible.
(28:00):
What I did then is I croppedjust the section of myself,
because again, I tried multipletimes to get the aspect ratio
right.
And it didn't, so I just croppedthe area where I appear and then
I went to the next step.
I decided to use it as atemplate.
So I uploaded the cropped imageto another new chat and I
uploaded an image of my son justbefore getting a haircut.
So his hair is really long andhe is sitting in a barber shop
(28:22):
looking at the camera, kind oflike a before and after image.
And what I asked the AI to do isI said, use the professional
headshot of the bald guy as atemplate, apply the same exact
background and similar clothingstyle to the kid.
And now I have a picture of myson wearing a buttoned up shirt
in a blazer with the same long,funny hair that he had before he
took the haircut, but with thesame exact transitioning blue in
(28:42):
the background.
Why is that important forbusiness?
You can create professionallooking headshots of everybody
in your company, in the sameexact style, with the same exact
background, with your logo in afew minutes.
Just by having images of youremployees, any employee, doesn't
matter what the image is, andyou can recreate it without
having to bring a professionalphotographer.
Set out the background, havecameras, edit that, let people
(29:04):
select for multiple options.
All of that can be done on acomputer in just a few minutes.
Another example, this one isfrom a user called Choi on X.
Again, his handel is ais ai andhe took a surveillance camera
image of a woman that is lookingto the left, and all he asked is
in the prompt to have her lookstraight into the camera, and
(29:26):
she does.
Again, everything else stays thesame.
The corridor, the lighting, theclothing that you, the woman is
wearing, everything is the same.
Why am I showing you this?
I'm showing this to see how easyit becomes to create deep fakes.
You can literally create anyscenario you want with any
person's face in anysurrounding, with just a simple
prompt and images of the personand the surroundings.
(29:48):
Now I want to talk to you aboutthe next level that Google
already provided us.
So in addition to all theseincredible capabilities, I'm
gonna show you some moreadvanced implementations.
Later on, I wanna show youanother amazing tool that they
created.
So Google created a solution inGoogle AI Studio that allows you
to create apps using this newGemini capability.
So you can use existing appsthat they're already sharing or
(30:10):
create your own apps that youwanna create just by explaining
them and prompting exactly whatthe app will do.
So they're shared severaldifferent examples of what these
apps do, and I'm gonna show yousome of them.
The first one allows you to editimages.
It's called AI powered photoediting.
Simplified retouch photo, applycreative filters or make
professional adjustments using asimple text prompts, no complex
tools needed.
(30:31):
So this is a tool that is calledPix Shop that they've built with
their own app building tools soyou can create your own tools
that do different things, thatallows you to edit images in a
professional way.
The next version is actuallyreally, really cool.
It's called Past Forward.
It allows you to upload an imageof you or anybody else and it
creates you in different decadesand it creates this kind of like
as Polaroid images right at thebottom of the image.
(30:53):
You will see which decade it'sfrom and it's absolutely cool.
Like my first top left image isfrom the 1950s and it has this
SIA color and I'm wearing thesejacket and tie old school,
including the images in thebackground.
There's like an old sofa andwood plank panels on the wall it
looks like from the fifties, butit definitely looks like me.
I actually really like myseventies version of me.
(31:15):
It has really long lush hair,kind of like Miami Vice style.
And I'm wearing a patternedshirt with a really large collar
and I look really, really cool.
The original image comes fromNASDAQ and it has several
different company logos in thebackground.
So what it did, which is really,really incredible, it understood
where I am.
It understood what's in thebackground and in the seventies
(31:35):
image it has companies from thatera.
So it has IBM, and Coca-Cola andIntel versus the companies that
are there right now in theoriginal image that I uploaded
and so on and so forth.
It's really cool.
It's really fun to play with andthe way it works, you can
actually click and see theprompting that it's doing in
order to create all theseimages.
So.
It's just very cool example.
The next one is a lot morepractical.
(31:55):
It's called Home Canvas.
And what they're showing in homecanvas is the ability to take a
image of a room, an actual roomthat you have in your office, in
your home, or wherever else, andtake a product image.
In this particular case, it isshowing a light stand, light
source.
And you can literally drag itand tell it where to place it.
I'm gonna place it on the littlenightstand next to the bed and
(32:15):
it's gonna think about it andit's gonna create that exact
light on the bedstand.
And you'll see in a secondexactly what it does.
So it's thinking about it, it'stelling you what it's doing, and
now it's placed it there and youcan actually even see the light.
So it's like a yellowish light,and you can see exactly how it's
gonna look like, how it reflectson the bed, and the bed stand,
and the floor and the shadows.
Everything is realistic,literally by just dragging and
dropping the thing in.
(32:37):
Now, I must admit, I try this onmultiple different scenarios
with multiple differentproducts, and it works about 25%
of the time and not a hundredpercent of the time, but every
time that it does work, it'sabsolutely magical.
The other cool thing is that itknows how to change the
orientation of the object to berelevant to where you placed it
in the room.
So again, it understands the 3Dnature of things.
(32:58):
It can anticipate how they'regonna look like from different
directions, and it can applythat in the new scenario.
More on that in the end.
Don't leave before we get to theend, because if you're selling
any kind of products.
Or you're doing promotion forany kind of products, you want
to see how you can do this atscale for your products.
And we're gonna show that as thegrand finale of this episode.
THe next few examples are peopletaking it to a completely
(33:19):
different level.
The first one that really blewmy mind comes from Alex Rasco.
His handle is at at x.
And what he created is hecreated a video of famous
pictures like Van Gogh, selfportrait, and the Mona Lisa and
so on.
And each and every one of themon his turn is turning into an
actual person that is walking inCentral Park in New York, and
(33:40):
they're meeting each other andbecoming very friendly and
taking selfies.
And it's nothing short of magic,both in means of creativity, but
also the actual implementationof this.
And then he adds more and morephotos of more and more famous
pictures, and they're turning.
Into actual humans and they'rein Central Park and they're
meeting all the other people andthey keep on engaging with one
(34:00):
another.
Again, brilliant, brilliantcreativity.
Absolutely stunning.
And it's showing you the abilityto take the fact that it knows
how to keep consistentcharacters and turn this into an
actual video with a storylinewith really unique, amazing
capabilities.
This is obviously not somebodywho's doing this the first time,
but the fact that it's possibleis absolutely mind blowing.
(34:22):
Let's look at another examplefrom a different creator.
This creator is called Framerwith the handle at zero X
framer, and he created thisunbelievable anime, multiple
scenes, about 30 to 45.
Second video of a chase of thisyounger kid driving his souped
up, upgraded motorcycle with thepolice, chasing him around, and
(34:43):
the police car gets into anaccident on the bridge, almost
falling off the bridge.
And it looks like aprofessionally done manga video.
It is incredible.
Like I'm sure my kids, if therewas a full episode, would watch
this thing.
How has this achieved the sameexact thing?
You can create the same exactcharacter with complete
consistency, the same scene, thesame street, the same police
(35:03):
car, the same police officer,the same bridge in multiple
angles using nano banana.
Then drop it into your favoritevideo generation tool and then
create this.
In this particular case, hecreated it with C Dance 1.0,
which is a video generation toolthat is getting a lot of love
online right now.
But he's saying he created thisentire video in less than two
hours, which is mind blowingagain.
(35:24):
The old way would've taken ateam of multiple people several
days to create this short film.
The next example that I wannashow you that goes into how
crazy the deep fakes of this is.
This comes from a guy's blog inIndia and he's showing how he's
adding different people, famouspeople into his selfies.
In this particular one, it's aselfie of him together with Brad
(35:45):
Pitt.
It is completely realistic.
There is no way of telling thatit's not actually Brad Pitt next
to him, and what you can see healso have a selfie with a known
Indian celebrity, which I don'tknow, but the original photo is
just him, obviously in theselfie.
The background is the samebackground.
The lighting is the samelighting, and you can add
whoever you want next to you.
(36:05):
This is really, really scary,but this is reality.
This is what we have right nowin the hands of anybody who has
access to Gemini, which isanybody who has access to the
internet.
And again, to make this crazier,you can drop this into any video
generation tool and have themhave a conversation in front of
the camera as if someone isfilming them or it's like a
selfie film of them talking.
This is totally doable in amatter of minutes to anybody you
(36:27):
want in the world.
The next one is a really simpleand yet really cool and helpful
use case that I actually usedthis past week for a
presentation, so one of thethings I was teaching in one of
the workshops last week is howto do workflow automation
together with ai.
And I showed examples in Make,and I showed examples in N8N.
Now, when I create mypresentations, all my
presentation slides have a blackbackground.
(36:48):
This is just a style that I likeand this shows up in all the
different presentations that Igive.
The problem is that Make hasthis kind of off-white
background.
So you have one option is toactually screenshot the thing
out of make, drop it into yourslide, but then you have a ugly
off-white rectangle in themiddle of your slide.
That's the top thing that you'reseeing on the screen right now.
The other option is to take itto a background remover.
(37:09):
Some of the background removerstoday are actually not bad.
I think the best one right nowis built into Canva.
So I used, for the secondoption, I used the Canva
background remover.
There are two problems withthat.
One is that the text is almostblack.
It's like a very dark gray.
So when you remove thebackground, the text that is
again, very dark gray, almostdoesn't show up against the
black background of my slides.
The other problem is for most ofthe different nodes in clay, the
(37:33):
inner image, basically that'stelling you what kind of node it
is, is in the same color as thebackground.
So when you remove thebackground, it turns black and
it's sometimes very hard to seewhat it is.
So then I just took thescreenshot that I took
originally that has the whitebackground and the dark text,
and I dropped it into Gemini andI literally wrote a prompt that
says, please change thebackground off this screenshot
to black and change the font.
(37:54):
Off the text to white.
That's it.
And I got exactly the outcomethat I wanted.
The image stayed exactly thesame, the text stayed exactly
the same, and I could now pasteit into my slides.
So this is a very practicalday-to-day kind of use case.
That was very hard to do beforeunless you have access to
Photoshop and you're a Photoshopwhiz.
And then still, this is a fasterway to do this.
The next example is an exteriordesign exercise that I've done.
(38:18):
So I took an image of thisreally old falling apart
balcony, and you can see theyard and you can see, kind of
like, see it's a weird angle,but you can kind of see the
doors, and the house.
You can see the light next tothe door, but it's an old porch
that is falling apart.
Then you can see old woodenplanks on the floor of the
porch, and they're all.
And they're all very, very old.
The color is mostly not there.
(38:40):
The, some of the pieces of thefence are missing or not
connected well.
And all I did is I wrote thefollowing prompt, create an
image suggesting how an upgradedversion of this porch can look
like.
Do not add furniture, but makeit look new, inviting and
contemporary style.
And then it created thisbeautiful version of the same
exact porch.
Now what you can see again, thebackground stays the same.
(39:02):
The trees stay the same, thepool in the backyard stays the
same.
Everything stays the same.
Just looks like a brand newporch that if I was a designer
or a general contractor, I couldhave probably sold this to the
owner of that porch.
But then I did something else.
I asked it to change the angleso I can see more of the porch,
and it was able to change theangle of the camera.
Now think about how crazy thisis.
It literally has to inventthings that are not in the
(39:23):
image, but understand exactlyhow it's gonna look like.
I'm gonna mention a few veryspecific details that are very
interesting.
It took the light that in theoriginal image shows up on one
side of the doors.
It cannot see the rest of thedoors, so it added a, another
one on the right side of thedoors because it makes sense.
It saw that in the middle of thewall in a specific distance,
there is a outdoor power plugwith a cover on it.
(39:45):
It added another one on this, onthe opposite side of the other
door.
It also, because of the newangle and also, which is
absolutely incredible because itcreated glass panels in the
railing.
Because we now have a new angle,it created reflections on the
glass of how the rest of theporch looks like.
This is really, really nextlevel because you can see the
(40:07):
yard through the reflection.
This looks completely realistic,and yet it made a lot of that
up.
Which takes us to the incredibleability of this tool to change
angles on things that it'sseeing and create new variations
from different aspects of thesame scene.
Now, this could be this porch.
This could be a product thatyou're selling.
So if you're selling a productand you wanna have multiple
(40:28):
shots of the same product frommultiple angles, you don't need
a professional photo shoot, youcan upload one or two versions
of the product.
It'll remove the background onits own, and then you can
generate any angles of theproduct as you wish.
But to put this on steroids, Iwill go to an incredible example
from my good friend, anincredible visual AI creator,
Rory Flynn.
So Rory took this new tool andused it in Weavy.
(40:51):
Those of you who don't knowWeavy, Weavy is an incredible,
incredibly powerful visualgeneration tool that allows you
to create images, combine themwith prompts, combine them with
different templates, and createan entire workflow out of that
template.
So what he did here is he tookan image of a Mercedes racing
car, but that's it.
That's the only visual input.
And then he added differentprompts and different steps that
(41:13):
basically tell it to createdifferent angles of the image.
So what you can see here is thatit's taking the array of
different angles.
You can see an array here andyou can see that it's suggesting
different angles.
It takes all of that in.
It takes the input of the car,and it is creating all these
angles.
So now, instead of just adiagonal 45 degrees from the
front view of the car, we have arear view of the car.
(41:34):
We have a side view of the car,and we have a 45 degree view of
the car as well.
Now to make this even crazier,one of the biggest still
existing disadvantages of nanobanana, also known as Gemini 2.5
flash image generation tool, isthat it, the quality of its
images is still low resolution.
So if we click on this image andmake it bigger.
(41:55):
You can see that the angle isperfect.
The details of the car isabsolutely perfect, but the
quality of the image is a littlegrainy.
However, the next step, becausethis is, allows you to create
multiple steps, the next step isusing, recraft upscale.
So the next image, which is thesame exact image only upscaled,
I can zoom in and zoom in andzoom in, and zoom in and zoom
(42:15):
in, and it still stays perfectlyclear.
So using a tool like Wvi or justusing any external upscale or
allows you to solve the scalingproblem, but then he added
something else for each andevery one of the different
angles of the results, he addedadditional steps that provide
different levels of zoom anddifferent, more ad like angles
(42:36):
of the car.
So now you can see multipleimages of the car in different
angles with closeups fromdifferent directions that.
Each and every one of them lookslike out of a professional shoot
off the car for ads, forwhatever you want.
And all of this is created withnow a single template.
What does that mean?
A single template?
It means that if you drop adifferent kind of image into the
(43:00):
same thing in this particularcase, an old red BMW, you get
all these assets, all theseoutputs, including the high-res
images, including all the zoomedin ones, including every one of
the things that we've seenbefore with the Mercedes racing
car for this new car.
And you can do this as manytimes as you want with other
different images, like thissemi-transparent view of this
futuristic vehicle.
(43:21):
Same exact result.
Why am I showing you all of thisif you are manufacturing or
promoting and selling any kindof physical goods, you can use a
tool like VY together, combinedwith Gemini 2.5 flash combined
with an upscale to create anyasset you want from any angle
with zero time.
(43:41):
Because once you create thetemplate, you can drop any new
image in there and create allthese assets.
So what does all of this mean?
Well, some of these capabilitiesexisted when the new Chate image
generation model came out wheneverybody was starting to create
manga images of themselves,Ghibli style kind of images, and
so on and so forth, myselfincluded.
And that was the first time wereally had a multimodal AI model
(44:05):
that you can have a conversationwith that understands what
you're asking, that understandscontext, that can have a
step-by-step workflow with AIimages.
And it took that to a completelydifferent level in realism,
completely different level inits ability to change angles and
directions, keep consistency ofcharacters and objects and
backgrounds and styles and soon.
And it is right now just anincredible tool to use for more
(44:27):
or less any image generation andor image editing that you want.
So I will ask a few questionsalready asked in the previous
news episode, but I think if youhaven't listened to that, and
even if you have it will be agood reminder.
So is this the death ofPhotoshop in marketing or in any
other aspect, right?
Is this the death of otherediting tools?
Is this the death of TV andmovie studios, et cetera, et
(44:50):
cetera.
Again, I showed you multipleexamples on what this tool can
do, and this is just the firsttwo weeks that it's been out.
Think about what happens when itgets upgraded.
So my answer is not yet, and Ithink the focus here is on yet,
but it's a big step in thedirection of putting all the
things that I mentioned earlierof very serious risk.
(45:11):
So what are the two main issuesthat I see for this tool right
now?
What I believe and hope, atleast from my personal
perspective, would be the nextsteps.
The first one is resolution.
This tool and these tools ingeneral generate very low
resolution results.
I don't think it's atechnological problem.
I think it's a computelimitation problem.
I think if they would alloweverybody to create images that
(45:31):
are eight K in resolution, thenthey will run out of GPUs very,
very quickly because justprocessing that just takes
significantly more compute,which means once they have more
bandwidth and or improve thealgorithms that do this, you'll
be able to generate more or lessany resolution you want.
Now, as I mentioned, there'salready upscales that exist
today that can solve thisproblem, at least to an extent,
(45:54):
which means you can take theoutput of nano bananas, put it
in an upscale, and then get amuch higher resolution image of
the same image that you hadbefore.
The next aspect, which I thinkis the biggest thing, comes back
to tooling.
And as we've seen with therecent six months of releases
from all the large labs, a bigpart of what they released was
new tooling, new ways to engagewith the models, whether it's
(46:15):
canvas for text and codeediting, et cetera.
It's Projects in ChatGPT whereyou can create these mini
universes of context, et cetera,et cetera.
Multiple tools that we can useto just have a better
interaction with the ai.
So I think what it's missing onthe visual side is layers,
right?
Think about even very basic.
Even very basic image editingtools, whether you go to Canva
(46:38):
or even PowerPoint has layers.
You can move things in front orbehind different other objects,
and then you can grab a layerand move it around and resize it
and rotate it and tilt it insome more advanced tools.
And you can do all of that, inall these tools.
I think once Cha GPT and Geminiadds layer control to their
tools, and I'll be reallysurprised if it's not happening
(47:00):
sometime soon.
That is game over to mostprofessional to most image
editing platforms out therebecause it's just better and
it's free or almost free becauseit is included with your
subscription.
If you get to the limits of whatyou're doing, you can always use
it through the API and pay fortokens, which is still gonna be
significantly cheaper than anyother way of doing this.
(47:23):
So bottom line, nano bananas,also known as GPT 2.5.
Flash is a very powerful,capable tool for image
generation, image editing, andany kind of creative work that
you want to do.
What I recommend is, as Irecommend for everything,
instead of just playing with itjust for fun, think about actual
use cases.
These use cases could bepersonal use cases.
You wanna renovate your house,you wanna see how a sofa will
(47:45):
look in your living room.
You wanna create a really coolsticker for your kids, for their
scouts.
You want to create a night storyfor your kids, whatever the case
may be, you can do that withthis model.
This obviously can apply tobusiness needs that we mentioned
a lot in this episode, so findan actual use case, business or
personal, and then just goexperiment.
Go get some inspiration online.
Again, there's no millions ofexamples of how people have used
(48:07):
it, and just try to do your ownthing, just like I showed you
today, and then take it fromthere and decide how that
applies and how it can help youin your day to day life.
I hope you found this episodehelpful.
As I mentioned, you can go checkout just the components or
everything on our YouTubechannel.
There's gonna be a link in theshow notes.
I'm also going to put a link.
I'm also gonna put the creditsin the show notes to every
(48:28):
single person that we have usedtheir work in this episode.
So I want to thank them eventhough I don't know them.
And as I mentioned in theprevious episode, we just
relaunched the self-pacedversion of the AI Business
Transformation course.
We're finalizing the edits bymid-September, so if you join
the course, you will get thelatest and greatest of the
(48:48):
course that I am still teaching.
That started in August and isfinishing in September.
You'll be able to take it as aself pace lesson by lesson at
your own time.
I don't think it's as good astaking the course with me.
I think having a humaninstructor and other people to
engage with during the course isbetter, but the next live course
might happen only in 2026because I'm booked solid with
(49:10):
company specific trainingbetween now and then.
So if you wanna take the course,that is a great way to do this
right now, and then later on, ifyou want to join the updated
live version sometime in 2026,you are more than welcome to do
that.
That's it for today.
Have an amazing rest of yourday.
Keep experimenting, share withthe world what you find so we
can all learn and have a greatrest of your day.
Hello and welcome to the Leveraging AI
(00:01):
Podcast, a podcast that sharespractical, ethical ways to
leverage AI to improveefficiency, grow your business,
and advance your career.
This is Isar Meitis, your host,and in the past week and a half
the world, or at least the AIworld, has gone bananas.
And the reason it has gonebananas it's because Google has
introduced Gemini 2.5 flashimage generation tool, which
they initially launched on theLM CS arena under the brand name
(00:25):
nano banana.
So everybody was going crazywith what this model can do.
Now we know it's Gemini 2.5flash image generation
capability, and this model cando incredible things.
Now, after doing a very shortreview of what it can do on this
past weekend's news episode, Idecided I have to dive in and
actually show you different usecases that are absolutely mind
(00:47):
blowing.
So let's start with what are thecapabilities of this model and
why is it so exciting?
And then we're gonna dive intothe actual details and examples
that I, some of them are minethat I'm gonna show you exactly
how they're done.
Some of them are just outputs ofmultiple people from the web
that are sharing their results.
mAny of them are way higher thanwhat I know how to do, or I had
the time to do in thisparticular case.
(01:08):
But let's get started withcapabilities.
So the first and most importantaspect of this model is that
it's incredible at keepingconsistency of everything on the
image.
So it can keep the consistencyof the background, it can keep
the consistency of object, itcan keep the consistency of
faces, people, modes,interactions, themes, styles,
(01:29):
literally anything you want, youcan take from an image and apply
to another, merge, two imagestogether and so on.
The other thing, talking aboutmerging things together, it is
very good at multi-image fusion.
I will show you multipleexamples of this.
You can combine differentobjects, you can combine objects
and scenes and background.
You can combine literallyanything you can imagine.
(01:49):
Now, the other thing that youcan do is you can do local
edits.
Either adding, removing,changing, more or less anything
in an existing image.
This image could be generated byAI or generated by an actual
image, meaning some picture youtook of something, it doesn't
really matter.
You can make changes to it.
When I say make changes, you canremove people from an image that
somebody photobombed yourpicture or you have a picture of
(02:11):
yourself with the alpha tower inthe background, and there's a
lot of other people there, butyou don't want them in the
image.
You can have them all removed.
You can, by the way, do theother way around.
You can take a picture of yourfamily and then add yourself to
the image that even thoughyou're the one that worked
behind the camera, taking thepicture, you can upload a
picture of yourself with orwithout the same exact stuff
that you were wearing that day,and then ask Gemini to add you
(02:33):
to that image.
The other thing that you can dois you can create a template and
then use it consistently acrossmultiple examples.
So this is good for creatingcards, badges, mockups of ads,
et cetera.
Literally any template that youwant, you can use as a template
and then reuse it again andagain and again in multiple
options and differentimplementations.
Again, I'll show you severaldifferent examples of that.
(02:54):
World knowledge, image reasoningis huge.
There's two aspects of that.
One, it is because it's athinking multimodal AI model and
not just an image generator.
It understands the world, whichmeans on one hand it understands
physics and how things interactwith one another.
So when you ask it to create animage of something, it'll make
sense in the physical worldunless you ask it not to.
(03:15):
And the other aspect of this isthat it understands business and
flow charts and so on.
So if you are now in a room, ina conversation and you have a
whiteboard and you're sketchingand drawing things, you can take
a picture of that.
Let's say there's a flow chartdiagram and you can ask it to
design the business flow andshow it in a graphical way.
(03:35):
It can do that because itunderstands everything that you
have created and you canrecreate it in a more accurate
way.
You can also, because it's alarge language model, you can
continue to have a conversationwith it to continue to edit and
manipulate the image step bystep by step, which you
obviously cannot do if you areusing a tool like Me Journey or
just a standard image generator.
(03:56):
Now, in theory, you could dothat because you can take a
screenshot of the image, enterit as a starting point, and then
try to edit from there.
But it's clunky and it'sdefinitely not as coherent and
cohesive as it is doing it innano banana.
From this perspective, it's verysimilar to using the new-ish
image generation model fromChatGPT that was released
earlier this year and again tookeverybody by a storm.
(04:18):
This model is also really goodat generating text on images,
but because it's also very goodin consistency, it can apply
your logo or anybody else's logofor that matter, to either a
banner that you're creating orput it on top of whatever object
that is in the image because itunderstands, again, physics.
So if there's a cup and you putyour logo on the cup, it will
(04:38):
render it as if it's in acircular way, wrapping around
the cup, et cetera, et cetera.
T-shirts, glasses, anything youwanna put this on will be
generated in the correctgraphics of the logo, but also
in the correct 3D applicationwhere you want to apply this.
Now, as I mentioned, you canalso transfer style.
So you can take a style from animage, a style from a type of
(04:58):
cloth, a style from literallyanything and apply to anything
else.
It is very good at maintainingpose and applying wardrobe
changes.
So if you want to see howsomething fits on you, you can
take a picture of you, take apicture of that piece of
clothing or accessory and youcan apply it to you and see how
exactly it is going to looklike.
I will show you multipleexamples of this as well, and
(05:21):
you can play around with thiswith different things.
You can re colorize and improveany kind of image.
So let's say you have abeautiful neighborhood photo and
you want to change the color ofthe buildings from blue to red.
You can do that, but moreimportantly, you can take old
photos that are either black andwhite and grainy, or not even
(05:41):
black or white, but just are inbad shape and use nano bananas
to make it significantly better.
There are multiple examplesonline right now for that, and
it's absolutely mind blowing.
It is on one hand, not as fastas some of the really fast image
generation models, like thefaster mode in mid journey, but
it is significantly faster thancreating images in chat pt,
(06:02):
which is the other option wehave if you want to have a
conversation and have a realworld understanding and include
that as part of a biggerprocess.
What you do need to be aware ofis that when you're creating
images with.
Nano bananas.
It will put two kind ofwatermarks on the image.
One is a Gemini logo on thebottom right corner, and the
other one is a synth id,invisible watermark, which is
(06:24):
basically a hidden pattern intothe actual pixels that Google
tools know how identify as toolsthat were created by ai.
I don't necessarily agree withthe visual watermark.
I think it's kind of stupid,especially that you can crop it
out in about two seconds.
But I love the fact that theyfinally, including synth ID in
the image generation aspect ofit.
(06:44):
I really hope that the worldwill come to an agreement on a
standard.
I don't care if it's synth ID oranything else of image and video
generation tools that generatestuff with AI that you'll be
able to, without any effort,hopefully with just a plugin on
your computer to know everyimage on every video, whether
they're generated with AI ornot.
So I really hope we're gonna goin that direction.
(07:05):
Now, let's start looking atmultiple examples and use cases,
and I'm going to share my screenand I'm going to explain
everything that is on the screenbecause I know you're right now
either walking your dog ordriving a car or jogging on a
treadmill or whatever it is thatyou're doing right now,
listening to this podcast.
So I will explain everythingthat is on the screen, but
because this particular episodeis extremely visual, I suggest
(07:27):
that afterwards when you havetime, click on the link in the
show notes of this episode, goto our YouTube channel and watch
the YouTube segments at leastjump to the stuff that you're
really interested in so youdon't have to watch the whole
thing again.
That doesn't make any sense, butat least you can jump to the
parts that you wanna see exactlywhat I was demonstrating.
Okay, so let me share my screenand let's dive into a bunch of
(07:49):
examples.
GMT20250901-210806_Recording (07:50):
So let's start by looking where to
get to this new magical modelthat you can use with inside of
Gemini.
So first of all, you just loginto your Gemini account and you
can see I have 2.5 pro up here,in the model selector.
I actually don't need that.
All you have to do is go towhere your prompt is.
There's a section for tools.
It has all the previous optionsthat were there, like deep
research, create videos withvideo canvas and guided
(08:10):
learning.
But you can see there's a newoption called Create Images with
a little banana image next toit.
You click on that.
And then you can start creatingnew images.
To be fair, you don't even haveto do that if you ask it to
create an image or if you dragan image in there for editing.
We're gonna talk about thatlater on.
It will know what it needs todo, and it will know how to run
this model either way.
But if you wanna be safe, youcan choose it from here.
The other option to get to it isthrough google AI Studio.
(08:32):
So if you are in ai,studio.Google.ai, there's all
these amazing options likestreaming live and so on.
But if you go to the regularchat, there's a button here that
says Try Nano banana.
And you click on that and nowyou can literally write your
prompt, edit your images, uploadstuff, whatever you want do, and
run it from here.
The biggest difference is ifyou're running it in Google AI
Studio, you are paying fortokens.
So you're paying a really smallamount of money every time you
(08:53):
run any command in here in AIStudio in general, including
creating and editing images.
And it is included with yourlicense if you're running it in
your regular Gemini account.
So my suggestion to you is startin the regular Gemini account
and if you run out of creditsand you need more, then go to
the air studio and just keep onrunning it over there and over
there you can run it foreverbecause you're paying for every
(09:13):
token.
GMT20250901-193429_Recordin (09:14):
The first few examples I wanna show
you come from Google themselves.
It's showing differentimplementation that they have
shared with the world.
The reason it's cool to look atis because first of all, it's
Google themselves.
They know exactly what the modelcan do.
But the other thing is it has alot of explanations of how to
actually use the modelseffectively.
So the first one is from theirdevelopers.google blog.com, and
we will share the results tothat.
(09:35):
And this article is called Howto Prompt Gemini 2.5 slash Image
Generation for the Best Results.
And if I scroll down here.
You'll see.
How does this look like?
So you see a beautiful image ofan all Chinese guy holding a
piece of pottery that he hascreated.
The image looks completelyphotorealistic.
He's in his studio.
The background looks completelyreal.
The lighting is perfect, andthey're showing you a few
(09:57):
things.
They're showing you what is theprompt, and they're showing you
what is the template of theprompt.
So I will give you an exampleabout this one.
The prompt itself is aphotorealistic closeup portrait
of an elderly Japaneseceramicist with deep sun edged
wrinkles and a warm knowingsmile.
He's carefully inspecting afreshly glazed T-ball.\ The
(10:18):
setting is his rustic sun drenchworkshop.
The scene is illuminated by softgolden hour light streaming
through the window, highlightingthe fine texture of the clay
captured with an 85 millimeterportrait lens, resulting in a
soft, blurred background inparenthesis, bouquet.
The overall mood is serene andmasterful.
(10:38):
Vertical portrait orientation.
And again, the overall image isperfectly coherent with a very
long and highly detailed prompt,and they're giving it as a
template.
So the template is a photorealistic.
And then what is the shot typeof, what is the subject next
action or expression.
Next set in whatever environmentthe scene is illuminated by
(11:02):
liking, description, creating amood.
Atmosphere capturing with a whattype of lens details.
And you can skip that if youdon't know that.
You can just explain what kindof look you want for the image
emphasizing, and then keytexture and details.
The image should be in whateveraspect ratio you want, so you
(11:23):
can apply this to literallyanything you need.
So this is just one example.
A very different example isthere's a cool red panda sticker
that looks perfect as a sticker.
It's 2D.
It's very, very different thanthe previous image.
And here the example prompt is akawai style sticker, a happy red
panda, wearing a tiny bamboohat.
(11:43):
It is munching on a green bambooleaf.
The design feature, bold cleanoutlines, simple cell shading
and vibrant color palette.
The background must be white.
And it looks like a perfectsticker that is aligned exactly
with that.
And again, they have thetemplate, which I'm not going to
read on how to create stickers,and then they give even more and
more examples on how to createlogos and how to create a
(12:05):
perfect image of a product photoshoot.
In this particular case, it's ablack mug on a granite style
kitchen countertop with fumescoming out of it.
And then they have aminimalistic view of a white
background with a delicate redmaple leaf in the corner.
And again, the example prompt isa minimalist composition
featuring a single delicate redmaple leaf positioned in the
(12:27):
bottom right of the frame, etcetera, et cetera.
So very different from all theother ones.
Another great example is a verydark noir art style comic.
Panel out of a comic sketch andit is perfect.
It has this guy in a long, darkcoat with a hat.
It looks very shady in like analley.
(12:48):
There's a place in the back thatis called the Black cat.
It is raining, but it's all inlike these black and white
sketches.
It looks absolutely incredible.
And again, same kind of thing.
They're giving you the exactprompt and the exact template to
create the same kind of thing.
Then they're showing you thatyou can edit images.
So the first example they giveis there's an image of a cat on
the left, and the prompt isvery, very simple.
Using the provided image of mycat, please add a small knitted
(13:10):
wizard hat on its head.
Make it look like it's sittingcomfortably and matches the soft
lighting of the photo.
And the second image looksexactly like the first image,
only the cat has a littleneeded.
Wizard hat on its head.
His ears are seeking out andit's really, really cute.
But everything else in the imagestays exactly the same.
(13:31):
The next example is they'reshowing a beautiful contemporary
living room with a really bigblue sofa.
And in this particular case,you're using the prompt using
the provided image of a livingroom change only the blue sofa
to be a vintage brown leatherChesterfield sofa.
Keep the rest of the room,including the pillows on the
sofa and the lighting unchangedand it works.
(13:52):
Now it's.
Again, absolutely amazing.
Everything in the room staysexactly the same.
There are multiple details inthis room.
There are lights in the ceiling.
There are shelves on the rightwith multiple Tcho keys on them
in this contemporary room, andnothing else changes, including
the view from the window otherthan the solf itself.
This is nothing short ofmagical, and they're showing
multiple examples.
Another really interesting onethat I actually did an
(14:13):
experiment of my own, andactually took it a little
further, is what I told youbefore, is the ability to dress
somebody.
So they took two inputs.
They took an input of a dress,just a flat image of the dress
without anybody.
It's just the dress itself andan image of a model that is
standing in a studio wearingblack pants and a white sweater.
(14:33):
And here the prompt is create aprofessional e-commerce fashion
photo.
Take the blue floral dress fromthe first image Let the woman
from the second image wear it,generate a realistic, full body
shot of the woman wearing thedress with a lighting and a
shadow adjusted to match theoutdoor environment, and the
image is absolutely perfect.
You can see the same Molly, it'sthe same face, same hairstyle, a
(14:55):
different pose.
It looks as if she's walkingquietly in a little alley in
Europe.
The lighting is perfect.
It's absolutely amazing.
And she's wearing the exactdress from the flat image, so
you can use this either ifyou're selling products or if
you want to try on products, youcan do that as well.
Now I wanna show you my littleattempt on the same kind of
thing.
And I took it to anotherextreme.
(15:16):
I thought about, okay, if Iwanna sell a dress and I wanna
do a cool campaign with it, howabout dressing famous women's
statues with that dress?
So I have two pictures here.
One is a picture of the statueof the woman in front of the
ferry building in San Francisco.
And I have a picture of a dressand the picture of that statue
of the woman in San Francisco.
(15:37):
And literally all I wrote isdress the statue with the dress
in the other picture really,really short.
And I got the exact output thatI wanted.
You can see the statue, theferry building stays the same.
The people walking around andsitting next to the statue is
the same.
Everything is the same, but thestatue is wearing the dress.
Then I wanted to get fancy.
I uploaded a picture of theStatue of Liberty, which is kind
of strange because it's alreadydressed in a gown or a dress or
(15:58):
a toga.
I'm not sure exactly what she'swearing.
But I asked the same question.
I uploaded the dress again.
I said, make the Statue ofLiberty.
Wear the attached dress insteadof its current cloth.
What it did is actually really,really cool.
It wrapped the Statue of Libertywith the same fabric and kinda
like applied it to the currentthing the Statue of Liberty is
wearing.
This is really, really cleverbecause I wasn't exactly sure
how this is gonna work out.
And it actually worked out veryinterestingly.
(16:20):
So this is just my example ofthat.
So let's look at a different setof Google examples.
So in this particular case,they're showing different sets
of double photos and how theycan be edited in order to apply
new aspects or new anything tothe same images.
They're all really, reallystunning.
So let's look at them one byone.
The first one has a input imageof a girl with her face really
(16:43):
dirty in a spacesuit, and thesecond image is a closer closeup
of her without the helmet.
So just taking the helmet off.
The next one is taking a sideshot of a redhead girl in a
truck, driving in an openterrain space somewhere, and
then it applies differentbackgrounds to the same picture,
so snowy mountains, et cetera,whatever you want.
(17:03):
The girl stays the same, hergaze, the lighting, everything
stays the same.
The scenery outside changes.
Another example is that theyhave.
Is taking two little funnylittle cartoonish characters and
then placing them in a recordingstudio and they're actually
operating the controls asthey're recording somebody else.
One of them have headphones onand so on, and they have the
prompts to each and every one ofthose.
(17:24):
I'm just sharing with you, youwhat you can do.
And there's really amazingimages of different kinds of
photos.
They have a photo of a girl onthe left and then multiple
images of her in Polaroidwearing different things in
different hail styles on theright.
Again, absolutely accurate andamazing.
And then an underwater sceneand.
Literally anything you canimagine.
Here's another example ofmerging two images, so a
(17:46):
professional photography imageof a woman and then a guy, and
then the joint photo.
They are hugging and lookingvery tenderly at each other in a
street in the snow with bouquetlighting in the background.
It's a perfect image, but it'sexactly the same people from the
two other images.
So these are just differentexamples of what you can do with
this.
The next one is really, reallycool.
They have a picture of a youngguy, kind like looking towards
(18:10):
the camera, and then they havedifferent professions for the
same guy all created just withproms.
So in the first one, he'swearing doctor scrubs and he has
stethoscope on his neck.
In the second one, he's ateacher in a classroom.
In the third one, he's workingon a clay statue, and in the
fourth one, he's preparing doughin a bakery.
All of them look exactly likethe guy only in completely
(18:30):
different settings withcompletely different poses, but
it is very clearly the same guyagain, using consistency.
So these are these kind ofexamples.
Then they have an example of howto turn a dog that is an actual
real dog into a game characterin a game style that looks like
Mario.
Then they're showing anothercartoon character.
In this particular case, it's alittle dinosaur that looks like
it's made out of plastic, andthen they're dressing it up in
(18:51):
multiple different kind ofuniforms as a superhero, as a
pirate, as a cheek and so on.
Absolutely incredible.
Then they're showing how you cancombine different styles and
aspects of different photos.
So they're taking a photo of awoman wearing an orange piece of
clothing and a photo of a pieceof glass floating in midair, and
they're showing the woman theoutput is the woman floating in
(19:12):
this bubble of glass in midair,and it's very impressive.
They took a picture of whalesjumping in the water and a
picture of a snowy mountain withthe lighting on top in the
sunset and combine the two whereyou can see the mountain in the
background as the whales arejumping in the ocean.
They took a picture of a forkand a picture of spaghetti and
created a fork made out ofspaghetti just with a single
(19:33):
prompt and so on and so forth.
Really, really amazing.
And then they're showing how youcan clean images and make
changes to images.
So there's an image of an imageof a building with very distinct
color patterns on it.
In several different segments ofthe pattern are in blue, and the
prompt is literally change allthe blue to green and you get
the same exact image, with thesame model standing in the
(19:54):
middle, yet wearing her yellowcloth just with a green
background instead of blue.
They have an image of a guywalking towards a truck,
probably somewhere in theMidwest.
There's a crazy tornado in thebackground.
And in the second image, it's abeautiful evening with no bad
weather whatsoever.
But everything else is exactlythe same.
They have an image of a housewhere the next four photos are
(20:14):
the same exact house, but infour different seasons.
And you can very clearly see howthe house looks like when it's
covered in snow, how it is whenthe leaves are changing, and so
on and so forth.
The next example that they'regiving is how to push design and
use existing ideas in order toapply them to something else.
So I have a picture of abutterfly, and they're creating
a picture of a model with abutterfly dress that is
completely inspired by thepattern on the butterfly's
(20:37):
wings.
They have a picture of a livingroom that is very nicely done,
and then they're adding a colorpalette, literally just five
different colors, and they'reapplying it to the cushions on
the sofa to give some colors tothe room.
Again, everything else stays thesame.
Then they're taking a face of aperson and applying it as a
cartoonish style on the side ofa cereal box and so on and so
(20:58):
forth.
Taking something from one imageand applying it to a different
image in multiple aspects.
Let's look at a few moreexamples of things that I did.
So I found this thing on x.com.
And I hope this is the originalI was really trying to find,
because these examples now areeverywhere.
I was trying to find theoriginal creator so I can give
them credit.
So in this particular case, it'sZoe, I hope I'm pronouncing his
(21:20):
name correctly.
And his handle in X is Zoe.
Z.
Z Zo.
So Zo, Z, Z.
And what he did is he took asketch of a figure.
The way you start, when youstart creating a painting of
something, especially whenyou're trying to create an image
of a superhero or something likethat.
sO just the outlines of the bodyIn a very specific pose, and he
(21:40):
took an image of a Japanesemodel and he asked to apply the
model to the pose in thesketched character, and he did a
perfect job.
It is absolutely mind blowingbecause the girl is the same
exact girl.
She's wearing the same exactstuff she's wearing in the image
and she's standing or kind oflike crouching in the same exact
(22:00):
pose as the sketched figure issitting at.
The other really incrediblething is that the, the original
image of the model, you cannotsee the bottom of her feet, so
it had to guess what she'swearing.
Give it the same kind of styleof the other stuff that she's
wearing.
And it's actually done a verygood job of doing this.
And I was interested was that,how many attempts it took to
(22:20):
create that?
So what I did is I took theimage on the left, basically the
sketched outline of thesuperhero pose and together with
the image of the girl on the topright.
And I asked Gemini and I droppedit into Gemini and I wrote the
following prompt, remove theimage of the girl from the top
right corner and then.
We already learned that that'sdoable.
I can remove stuff from theimage.
So I got a image of just thesketched outline of the
(22:43):
superhero pose, and then Iuploaded a selfie image of more
or less just my head andshoulders.
That's it.
And I said, now create a photorealistic studio photo shoot of
the guy in the image, in thesame exact pose as the action
figure sketch white background.
And I got a perfect pose of me.
And again, you had to guesseverything that I'm wearing
(23:04):
because all you see is myshoulders with an off white
sweater.
So it applied the sweater as, asif I'm wearing it tied around my
neck.
And it gave me very stylishclothing that I will never
actually wear because I'm not astylish kind of guy.
I wear t-shirts and shorts mostof the time, but the image is
perfect and it's totally me,like the face is me.
The pose could have been me ifwould've been that flexible.
(23:24):
Uh, but it's very, veryimpressive.
And then I said, okay, now dresshim up as Spider-Man, but
without the mask.
So now I have the same exactimage because it's a superhero
kind of pose with me instead ofwearing the clothes I was
wearing before or that I'dinvented for me dressed as
Spider-Man, but it's still myface, still the same pose.
And then I said, okay, place himon top of a skyscraper at night
in a medium shot.
And it did that as well.
(23:46):
It Even placed the light fromthe moon in a very specific
angle that drops shadow from thepose that I'm standing on the
rooftop.
And then you can go on and onand on and build an entire
superhero story from that oneoriginal shot.
So it's one line of text andthese magical outputs just come
out.
Let's look at another examplefrom X.
This creator from X created asketch on a piece of paper and
(24:09):
took a picture of it.
The sketch is of a modelstanding and stretching up.
It is the kind of sketch youwould see from clothing
designers create when they startthinking and sketching the ideas
for their next clothes.
So it's this girl who'sstretching up and you can see
that she's wearing Calvin Kleinunderwear and you can see what
kind of shirt she's wearing, butyou can't really see the girl.
And with one prompt, he createdan image of an actual model in
(24:32):
the same exact pose, wearing theexact clothe that he was
imagining in his sketch.
This is absolutely insane.
It's going from an idea topaper, old school way, going
from that to a photo shoot,fully realistic image of the
model in the same exactscenario.
This is absolutely mind blowing.
Now let's look at combiningimages.
(24:52):
I was trying something very coolhere.
So what I did is I looked onlinefor standard images of different
ingredients.
I found most of them on websitesof different grocery shops.
So I have an image of a avocadosliced.
I have an image of corn.
I have an image of sour cream.
I have an image of chickenthighs wrapped in a frozen
chicken thigh box.
(25:12):
I have cilantro, I havetomatoes, and I have tortillas.
Each and every one of them is astandalone image that I got from
the web with no background ofeach and every one of these
ingredients.
And then all I wrote is create apromotional social media image
of a dish I can make from allthese ingredients.
And it created this beautiful,incredible image of a modern
(25:32):
kitchen that is blurry in thebackground.
There's a gray countertop on topof that, there is a wooden
cutting board.
And on top of that, there isthis perfect deliciously looking
chicken wrap.
And the ad says, dinner solved,exclamation point, delicious
chicken wraps made with freshingredients.
It is absolutely amazing.
Then I ask it to change theaspect ratio of the image, and
(25:54):
it failed.
And I must admit, I triedchanging the aspect ratio of
images on Gemini multiple times,and it fails every single time.
Now it might be me personally,I've seen a lot of other people
do it.
When I asked Gemini, can you dothat?
It said, yes, absolutely.
But every time I tried it, itfailed.
I tried it in multiple scenarioswith multiple images.
It never worked for me.
I assumed this is something theywill solve soon.
(26:14):
Then I was trying to create aneven funnier ad.
So I was playing around withdifferent Yeti images in the
end, what I got to is there's aYeti in a bus with multiple
other people next to the Yeti.
Everybody's eating chickenwraps.
The Yeti is all dirty with thesour cream dripping all over his
fur.
And the ad is sayingunapologetically delicious, and
on the bottom, in smaller frontit says chicken wrap.
(26:36):
So tasty.
You do not care.
Getting messy in public, it is agreat funny ad that I went from
ingredients to an ad I canprobably use in about six to
seven minutes.
This would've taken the old way,months, or at least weeks to
create between coming up withthe ideas, doing the shoot of
the actual wrap, et cetera, etcetera, et cetera.
(26:56):
I can create now a whole seriesof ads based on the same style,
based on the same comedyconcepts, and just keep on
creating them right here in thesame chat.
Another example of usingmultiple images to achieve a
goal is I took a selfie ofmyself with a very colorful
background, really messy, and Itook a screenshot outta my brand
guidelines of a transition fromdark blue to light blue that I
(27:18):
use in different aspects of myproducts and my websites and so
on.
And I took a screenshot of mylogo and I asked the AI to
create a professional headshotof this person.
Keep the image aspect ratio ofthe person as is, change the
background to the colortransition in the same color of
blue as your attachment on thetop left corner, add a small
version of the logo that isattached.
(27:40):
The person should be wearing awhite buttoned up shirt with a
blazer and no tie.
And other than the aspect ratioaspect of it, which again we
already covered it created abeautiful, amazing professional
photo shoot of myself with thetransitioning color behind me
with the logo.
Perfectly done on the top leftcorner.
This is amazing.
This is absolutely incredible.
(28:00):
What I did then is I croppedjust the section of myself,
because again, I tried multipletimes to get the aspect ratio
right.
And it didn't, so I just croppedthe area where I appear and then
I went to the next step.
I decided to use it as atemplate.
So I uploaded the cropped imageto another new chat and I
uploaded an image of my son justbefore getting a haircut.
So his hair is really long andhe is sitting in a barber shop
(28:22):
looking at the camera, kind oflike a before and after image.
And what I asked the AI to do isI said, use the professional
headshot of the bald guy as atemplate, apply the same exact
background and similar clothingstyle to the kid.
And now I have a picture of myson wearing a buttoned up shirt
in a blazer with the same long,funny hair that he had before he
took the haircut, but with thesame exact transitioning blue in
(28:42):
the background.
Why is that important forbusiness?
You can create professionallooking headshots of everybody
in your company, in the sameexact style, with the same exact
background, with your logo in afew minutes.
Just by having images of youremployees, any employee, doesn't
matter what the image is, andyou can recreate it without
having to bring a professionalphotographer.
Set out the background, havecameras, edit that, let people
(29:04):
select for multiple options.
All of that can be done on acomputer in just a few minutes.
Another example, this one isfrom a user called Choi on X.
Again, his handel is ais ai andhe took a surveillance camera
image of a woman that is lookingto the left, and all he asked is
in the prompt to have her lookstraight into the camera, and
(29:26):
she does.
Again, everything else stays thesame.
The corridor, the lighting, theclothing that you, the woman is
wearing, everything is the same.
Why am I showing you this?
I'm showing this to see how easyit becomes to create deep fakes.
You can literally create anyscenario you want with any
person's face in anysurrounding, with just a simple
prompt and images of the personand the surroundings.
(29:48):
Now I want to talk to you aboutthe next level that Google
already provided us.
So in addition to all theseincredible capabilities, I'm
gonna show you some moreadvanced implementations.
Later on, I wanna show youanother amazing tool that they
created.
So Google created a solution inGoogle AI Studio that allows you
to create apps using this newGemini capability.
So you can use existing appsthat they're already sharing or
(30:10):
create your own apps that youwanna create just by explaining
them and prompting exactly whatthe app will do.
So they're shared severaldifferent examples of what these
apps do, and I'm gonna show yousome of them.
The first one allows you to editimages.
It's called AI powered photoediting.
Simplified retouch photo, applycreative filters or make
professional adjustments using asimple text prompts, no complex
tools needed.
(30:31):
So this is a tool that is calledPix Shop that they've built with
their own app building tools soyou can create your own tools
that do different things, thatallows you to edit images in a
professional way.
The next version is actuallyreally, really cool.
It's called Past Forward.
It allows you to upload an imageof you or anybody else and it
creates you in different decadesand it creates this kind of like
as Polaroid images right at thebottom of the image.
(30:53):
You will see which decade it'sfrom and it's absolutely cool.
Like my first top left image isfrom the 1950s and it has this
SIA color and I'm wearing thesejacket and tie old school,
including the images in thebackground.
There's like an old sofa andwood plank panels on the wall it
looks like from the fifties, butit definitely looks like me.
I actually really like myseventies version of me.
(31:15):
It has really long lush hair,kind of like Miami Vice style.
And I'm wearing a patternedshirt with a really large collar
and I look really, really cool.
The original image comes fromNASDAQ and it has several
different company logos in thebackground.
So what it did, which is really,really incredible, it understood
where I am.
It understood what's in thebackground and in the seventies
(31:35):
image it has companies from thatera.
So it has IBM, and Coca-Cola andIntel versus the companies that
are there right now in theoriginal image that I uploaded
and so on and so forth.
It's really cool.
It's really fun to play with andthe way it works, you can
actually click and see theprompting that it's doing in
order to create all theseimages.
So.
It's just very cool example.
The next one is a lot morepractical.
(31:55):
It's called Home Canvas.
And what they're showing in homecanvas is the ability to take a
image of a room, an actual roomthat you have in your office, in
your home, or wherever else, andtake a product image.
In this particular case, it isshowing a light stand, light
source.
And you can literally drag itand tell it where to place it.
I'm gonna place it on the littlenightstand next to the bed and
(32:15):
it's gonna think about it andit's gonna create that exact
light on the bedstand.
And you'll see in a secondexactly what it does.
So it's thinking about it, it'stelling you what it's doing, and
now it's placed it there and youcan actually even see the light.
So it's like a yellowish light,and you can see exactly how it's
gonna look like, how it reflectson the bed, and the bed stand,
and the floor and the shadows.
Everything is realistic,literally by just dragging and
dropping the thing in.
(32:37):
Now, I must admit, I try this onmultiple different scenarios
with multiple differentproducts, and it works about 25%
of the time and not a hundredpercent of the time, but every
time that it does work, it'sabsolutely magical.
The other cool thing is that itknows how to change the
orientation of the object to berelevant to where you placed it
in the room.
So again, it understands the 3Dnature of things.
(32:58):
It can anticipate how they'regonna look like from different
directions, and it can applythat in the new scenario.
More on that in the end.
Don't leave before we get to theend, because if you're selling
any kind of products.
Or you're doing promotion forany kind of products, you want
to see how you can do this atscale for your products.
And we're gonna show that as thegrand finale of this episode.
THe next few examples are peopletaking it to a completely
(33:19):
different level.
The first one that really blewmy mind comes from Alex Rasco.
His handle is at at x.
And what he created is hecreated a video of famous
pictures like Van Gogh, selfportrait, and the Mona Lisa and
so on.
And each and every one of themon his turn is turning into an
actual person that is walking inCentral Park in New York, and
(33:40):
they're meeting each other andbecoming very friendly and
taking selfies.
And it's nothing short of magic,both in means of creativity, but
also the actual implementationof this.
And then he adds more and morephotos of more and more famous
pictures, and they're turning.
Into actual humans and they'rein Central Park and they're
meeting all the other people andthey keep on engaging with one
(34:00):
another.
Again, brilliant, brilliantcreativity.
Absolutely stunning.
And it's showing you the abilityto take the fact that it knows
how to keep consistentcharacters and turn this into an
actual video with a storylinewith really unique, amazing
capabilities.
This is obviously not somebodywho's doing this the first time,
but the fact that it's possibleis absolutely mind blowing.
(34:22):
Let's look at another examplefrom a different creator.
This creator is called Framerwith the handle at zero X
framer, and he created thisunbelievable anime, multiple
scenes, about 30 to 45.
Second video of a chase of thisyounger kid driving his souped
up, upgraded motorcycle with thepolice, chasing him around, and
(34:43):
the police car gets into anaccident on the bridge, almost
falling off the bridge.
And it looks like aprofessionally done manga video.
It is incredible.
Like I'm sure my kids, if therewas a full episode, would watch
this thing.
How has this achieved the sameexact thing?
You can create the same exactcharacter with complete
consistency, the same scene, thesame street, the same police
(35:03):
car, the same police officer,the same bridge in multiple
angles using nano banana.
Then drop it into your favoritevideo generation tool and then
create this.
In this particular case, hecreated it with C Dance 1.0,
which is a video generation toolthat is getting a lot of love
online right now.
But he's saying he created thisentire video in less than two
hours, which is mind blowingagain.
(35:24):
The old way would've taken ateam of multiple people several
days to create this short film.
The next example that I wannashow you that goes into how
crazy the deep fakes of this is.
This comes from a guy's blog inIndia and he's showing how he's
adding different people, famouspeople into his selfies.
In this particular one, it's aselfie of him together with Brad
(35:45):
Pitt.
It is completely realistic.
There is no way of telling thatit's not actually Brad Pitt next
to him, and what you can see healso have a selfie with a known
Indian celebrity, which I don'tknow, but the original photo is
just him, obviously in theselfie.
The background is the samebackground.
The lighting is the samelighting, and you can add
whoever you want next to you.
(36:05):
This is really, really scary,but this is reality.
This is what we have right nowin the hands of anybody who has
access to Gemini, which isanybody who has access to the
internet.
And again, to make this crazier,you can drop this into any video
generation tool and have themhave a conversation in front of
the camera as if someone isfilming them or it's like a
selfie film of them talking.
This is totally doable in amatter of minutes to anybody you
(36:27):
want in the world.
The next one is a really simpleand yet really cool and helpful
use case that I actually usedthis past week for a
presentation, so one of thethings I was teaching in one of
the workshops last week is howto do workflow automation
together with ai.
And I showed examples in Make,and I showed examples in N8N.
Now, when I create mypresentations, all my
presentation slides have a blackbackground.
(36:48):
This is just a style that I likeand this shows up in all the
different presentations that Igive.
The problem is that Make hasthis kind of off-white
background.
So you have one option is toactually screenshot the thing
out of make, drop it into yourslide, but then you have a ugly
off-white rectangle in themiddle of your slide.
That's the top thing that you'reseeing on the screen right now.
The other option is to take itto a background remover.
(37:09):
Some of the background removerstoday are actually not bad.
I think the best one right nowis built into Canva.
So I used, for the secondoption, I used the Canva
background remover.
There are two problems withthat.
One is that the text is almostblack.
It's like a very dark gray.
So when you remove thebackground, the text that is
again, very dark gray, almostdoesn't show up against the
black background of my slides.
The other problem is for most ofthe different nodes in clay, the
(37:33):
inner image, basically that'stelling you what kind of node it
is, is in the same color as thebackground.
So when you remove thebackground, it turns black and
it's sometimes very hard to seewhat it is.
So then I just took thescreenshot that I took
originally that has the whitebackground and the dark text,
and I dropped it into Gemini andI literally wrote a prompt that
says, please change thebackground off this screenshot
to black and change the font.
(37:54):
Off the text to white.
That's it.
And I got exactly the outcomethat I wanted.
The image stayed exactly thesame, the text stayed exactly
the same, and I could now pasteit into my slides.
So this is a very practicalday-to-day kind of use case.
That was very hard to do beforeunless you have access to
Photoshop and you're a Photoshopwhiz.
And then still, this is a fasterway to do this.
The next example is an exteriordesign exercise that I've done.
(38:18):
So I took an image of thisreally old falling apart
balcony, and you can see theyard and you can see, kind of
like, see it's a weird angle,but you can kind of see the
doors, and the house.
You can see the light next tothe door, but it's an old porch
that is falling apart.
Then you can see old woodenplanks on the floor of the
porch, and they're all.
And they're all very, very old.
The color is mostly not there.
(38:40):
The, some of the pieces of thefence are missing or not
connected well.
And all I did is I wrote thefollowing prompt, create an
image suggesting how an upgradedversion of this porch can look
like.
Do not add furniture, but makeit look new, inviting and
contemporary style.
And then it created thisbeautiful version of the same
exact porch.
Now what you can see again, thebackground stays the same.
(39:02):
The trees stay the same, thepool in the backyard stays the
same.
Everything stays the same.
Just looks like a brand newporch that if I was a designer
or a general contractor, I couldhave probably sold this to the
owner of that porch.
But then I did something else.
I asked it to change the angleso I can see more of the porch,
and it was able to change theangle of the camera.
Now think about how crazy thisis.
It literally has to inventthings that are not in the
(39:23):
image, but understand exactlyhow it's gonna look like.
I'm gonna mention a few veryspecific details that are very
interesting.
It took the light that in theoriginal image shows up on one
side of the doors.
It cannot see the rest of thedoors, so it added a, another
one on the right side of thedoors because it makes sense.
It saw that in the middle of thewall in a specific distance,
there is a outdoor power plugwith a cover on it.
(39:45):
It added another one on this, onthe opposite side of the other
door.
It also, because of the newangle and also, which is
absolutely incredible because itcreated glass panels in the
railing.
Because we now have a new angle,it created reflections on the
glass of how the rest of theporch looks like.
This is really, really nextlevel because you can see the
(40:07):
yard through the reflection.
This looks completely realistic,and yet it made a lot of that
up.
Which takes us to the incredibleability of this tool to change
angles on things that it'sseeing and create new variations
from different aspects of thesame scene.
Now, this could be this porch.
This could be a product thatyou're selling.
So if you're selling a productand you wanna have multiple
(40:28):
shots of the same product frommultiple angles, you don't need
a professional photo shoot, youcan upload one or two versions
of the product.
It'll remove the background onits own, and then you can
generate any angles of theproduct as you wish.
But to put this on steroids, Iwill go to an incredible example
from my good friend, anincredible visual AI creator,
Rory Flynn.
So Rory took this new tool andused it in Weavy.
(40:51):
Those of you who don't knowWeavy, Weavy is an incredible,
incredibly powerful visualgeneration tool that allows you
to create images, combine themwith prompts, combine them with
different templates, and createan entire workflow out of that
template.
So what he did here is he tookan image of a Mercedes racing
car, but that's it.
That's the only visual input.
And then he added differentprompts and different steps that
(41:13):
basically tell it to createdifferent angles of the image.
So what you can see here is thatit's taking the array of
different angles.
You can see an array here andyou can see that it's suggesting
different angles.
It takes all of that in.
It takes the input of the car,and it is creating all these
angles.
So now, instead of just adiagonal 45 degrees from the
front view of the car, we have arear view of the car.
(41:34):
We have a side view of the car,and we have a 45 degree view of
the car as well.
Now to make this even crazier,one of the biggest still
existing disadvantages of nanobanana, also known as Gemini 2.5
flash image generation tool, isthat it, the quality of its
images is still low resolution.
So if we click on this image andmake it bigger.
(41:55):
You can see that the angle isperfect.
The details of the car isabsolutely perfect, but the
quality of the image is a littlegrainy.
However, the next step, becausethis is, allows you to create
multiple steps, the next step isusing, recraft upscale.
So the next image, which is thesame exact image only upscaled,
I can zoom in and zoom in andzoom in, and zoom in and zoom
(42:15):
in, and it still stays perfectlyclear.
So using a tool like Wvi or justusing any external upscale or
allows you to solve the scalingproblem, but then he added
something else for each andevery one of the different
angles of the results, he addedadditional steps that provide
different levels of zoom anddifferent, more ad like angles
(42:36):
of the car.
So now you can see multipleimages of the car in different
angles with closeups fromdifferent directions that.
Each and every one of them lookslike out of a professional shoot
off the car for ads, forwhatever you want.
And all of this is created withnow a single template.
What does that mean?
A single template?
It means that if you drop adifferent kind of image into the
(43:00):
same thing in this particularcase, an old red BMW, you get
all these assets, all theseoutputs, including the high-res
images, including all the zoomedin ones, including every one of
the things that we've seenbefore with the Mercedes racing
car for this new car.
And you can do this as manytimes as you want with other
different images, like thissemi-transparent view of this
futuristic vehicle.
(43:21):
Same exact result.
Why am I showing you all of thisif you are manufacturing or
promoting and selling any kindof physical goods, you can use a
tool like VY together, combinedwith Gemini 2.5 flash combined
with an upscale to create anyasset you want from any angle
with zero time.
(43:41):
Because once you create thetemplate, you can drop any new
image in there and create allthese assets.
So what does all of this mean?
Well, some of these capabilitiesexisted when the new Chate image
generation model came out wheneverybody was starting to create
manga images of themselves,Ghibli style kind of images, and
so on and so forth, myselfincluded.
And that was the first time wereally had a multimodal AI model
(44:05):
that you can have a conversationwith that understands what
you're asking, that understandscontext, that can have a
step-by-step workflow with AIimages.
And it took that to a completelydifferent level in realism,
completely different level inits ability to change angles and
directions, keep consistency ofcharacters and objects and
backgrounds and styles and soon.
And it is right now just anincredible tool to use for more
(44:27):
or less any image generation andor image editing that you want.
So I will ask a few questionsalready asked in the previous
news episode, but I think if youhaven't listened to that, and
even if you have it will be agood reminder.
So is this the death ofPhotoshop in marketing or in any
other aspect, right?
Is this the death of otherediting tools?
Is this the death of TV andmovie studios, et cetera, et
(44:50):
cetera.
Again, I showed you multipleexamples on what this tool can
do, and this is just the firsttwo weeks that it's been out.
Think about what happens when itgets upgraded.
So my answer is not yet, and Ithink the focus here is on yet,
but it's a big step in thedirection of putting all the
things that I mentioned earlierof very serious risk.
(45:11):
So what are the two main issuesthat I see for this tool right
now?
What I believe and hope, atleast from my personal
perspective, would be the nextsteps.
The first one is resolution.
This tool and these tools ingeneral generate very low
resolution results.
I don't think it's atechnological problem.
I think it's a computelimitation problem.
I think if they would alloweverybody to create images that
(45:31):
are eight K in resolution, thenthey will run out of GPUs very,
very quickly because justprocessing that just takes
significantly more compute,which means once they have more
bandwidth and or improve thealgorithms that do this, you'll
be able to generate more or lessany resolution you want.
Now, as I mentioned, there'salready upscales that exist
today that can solve thisproblem, at least to an extent,
(45:54):
which means you can take theoutput of nano bananas, put it
in an upscale, and then get amuch higher resolution image of
the same image that you hadbefore.
The next aspect, which I thinkis the biggest thing, comes back
to tooling.
And as we've seen with therecent six months of releases
from all the large labs, a bigpart of what they released was
new tooling, new ways to engagewith the models, whether it's
(46:15):
canvas for text and codeediting, et cetera.
It's Projects in ChatGPT whereyou can create these mini
universes of context, et cetera,et cetera.
Multiple tools that we can useto just have a better
interaction with the ai.
So I think what it's missing onthe visual side is layers,
right?
Think about even very basic.
Even very basic image editingtools, whether you go to Canva
(46:38):
or even PowerPoint has layers.
You can move things in front orbehind different other objects,
and then you can grab a layerand move it around and resize it
and rotate it and tilt it insome more advanced tools.
And you can do all of that, inall these tools.
I think once Cha GPT and Geminiadds layer control to their
tools, and I'll be reallysurprised if it's not happening
(47:00):
sometime soon.
That is game over to mostprofessional to most image
editing platforms out therebecause it's just better and
it's free or almost free becauseit is included with your
subscription.
If you get to the limits of whatyou're doing, you can always use
it through the API and pay fortokens, which is still gonna be
significantly cheaper than anyother way of doing this.
(47:23):
So bottom line, nano bananas,also known as GPT 2.5.
Flash is a very powerful,capable tool for image
generation, image editing, andany kind of creative work that
you want to do.
What I recommend is, as Irecommend for everything,
instead of just playing with itjust for fun, think about actual
use cases.
These use cases could bepersonal use cases.
You wanna renovate your house,you wanna see how a sofa will
(47:45):
look in your living room.
You wanna create a really coolsticker for your kids, for their
scouts.
You want to create a night storyfor your kids, whatever the case
may be, you can do that withthis model.
This obviously can apply tobusiness needs that we mentioned
a lot in this episode, so findan actual use case, business or
personal, and then just goexperiment.
Go get some inspiration online.
Again, there's no millions ofexamples of how people have used
(48:07):
it, and just try to do your ownthing, just like I showed you
today, and then take it fromthere and decide how that
applies and how it can help youin your day to day life.
I hope you found this episodehelpful.
As I mentioned, you can go checkout just the components or
everything on our YouTubechannel.
There's gonna be a link in theshow notes.
I'm also going to put a link.
I'm also gonna put the creditsin the show notes to every
(48:28):
single person that we have usedtheir work in this episode.
So I want to thank them eventhough I don't know them.
And as I mentioned in theprevious episode, we just
relaunched the self-pacedversion of the AI Business
Transformation course.
We're finalizing the edits bymid-September, so if you join
the course, you will get thelatest and greatest of the
(48:48):
course that I am still teaching.
That started in August and isfinishing in September.
You'll be able to take it as aself pace lesson by lesson at
your own time.
I don't think it's as good astaking the course with me.
I think having a humaninstructor and other people to
engage with during the course isbetter, but the next live course
might happen only in 2026because I'm booked solid with
(49:10):
company specific trainingbetween now and then.
So if you wanna take the course,that is a great way to do this
right now, and then later on, ifyou want to join the updated
live version sometime in 2026,you are more than welcome to do
that.
That's it for today.
Have an amazing rest of yourday.
Keep experimenting, share withthe world what you find so we
can all learn and have a greatrest of your day.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
The Herd with Colin Cowherd
The Herd with Colin Cowherd is a thought-provoking, opinionated, and topic-driven journey through the top sports stories of the day.