#nano-banana
1 messages ¡ Page 1 of 1 (latest)
I have got it all the time in image editing
maybe I am just so frequent as a user
lol
This model is my favorite thus far
whats your go to image prompt?
Hmmm...good question... turning my images into a funny comic panel is one..
I went with the option that was more full 
well.. prompt failed.
but reflection of light in the glass is extremely realisitc
yeah very true
I don't know what to think about Nano
It's very good sometimes and very bad some other times
Iâm just amazed at what it generates, without getting blocked. Plenty of NSFW. Wasnât sure if that is purely due to testing or if it will be that way once released.
It actually has the legs of wine, impressive
How so?
Hmm I haven't seen or heard reports of that yet
Pretty good so far vs all the other image models
The editing blows my mind.
I know it's test model with some rough edges... but is there any possibility that there is multiple models under the same anonymous alias- just like Google did with gemini-test?
is it confirmed that it's from Google?
I guess no. For prompts like "create art of lab who created you", it outputs Google, but that could be just false info in its system prompt / learning data.
no
Well flux max is also very good at this...
It's good at editing but doesn't understand complex prompts
I got very bad result for complex prompts
Nope. Not nearly as good in the testing.
Like two images, replace one character in one image with the other
For me at least
look
"Replace the text, use the exact same style and font, with "different day same sh1t"
if only the letter wasnt missing lol
- it added "the"
bad result
much better
it struggles so much lol
it's a super easy task, every other model get it right
asked it to replace it with a cig
it's good
sometimes it's good but sometimes its very bad
i add the same image with the part that needs to be edited highlighted, that seems to help
where does one access the banana?
How does one try out nano banana? I'm new to LM Arena
hold up is this what I think it is?
Ususally a skill issue with prompting
It's only in battle mode and is randomly assigned, you can't choose it. Only find out after you submit a prompt and choose whioch image you like the most.
Got to get lucky. Itâs in battle mode so youâll be sampled two random models
i like how it recreates the bad photoshop look perfectly
I love how nano-banana has its own dedicated channel now
I also low how the model's name is only a few steps removed from tiny peâ user was banned for this message
yea pretty cool
a joint? yea
Nano banana is good but I find myself liking the Imagen outputs a lot more
Nano banana sadly also doesnât know half the pop culture stuff that GPT1 does
LikeâŚGPT1 can create all kinds of anime, movie, game, tv show, etc. characters
Nah if you know you know
The models that should be removed from the site are DALL-E 3 and Gemini Flash
They have the worst outputs
They shouldnât even be in the running
Nano Banana is the best ai. đ
cute windows 2.0 kek
what is nano banana
an image model
nano banana
prompt was Cute pastel colored Windows 2.0 inspired OS with a drawing program open and a cute drawing of a satanic kitten holding a knife in it
I love it!
really? Source?
"source"
??
A nano banana chat channel?
It's clearly a Google model, usually OpenAI's image models (GPT-Image and Dall-e) take forever to generate the image, Google's image models are faster. And another thing, some DeepMind employees liked posts on X about nano-banana
is nano banana region limited? nver comes to me in battle
Nah. The model is probably refusing your prompts or your input images
It is hidden from leaderboard too?
thanks
Yeah, its anon model
Nano banana has a very similar style to Imagen, too

Impressive
Cc go đ
damn nano banana losing to flux
I love how flux left him in the crater xD
tbh thats what I wanted haha
How do you guys use nano banana?
Cheesy 90s Geocities page dedicated to Tsuchinoko with colorful word art "TSUCHINOKO REAL" and photos of Tsuchinoko and 90s clipart
I feel like that was supposed to say "you bet your beanie babies it is!" Which is awesome
nano banana is removed?
No it should still be there - only accessible through Battle mode though
my bad
hey why don't you guys also upload prompts when u upload the images - it would be really helpful ... i guess?
Can you please share some prompts?
What is nano-banana? Why does it get its own channel?
Wow it's so good at UI and text
Nano-banana is the first model after GPT to know this character
Lies, Imagen knows it too
Sigh... Google still ahead of everyone
this is the nano banana cult, please move to the google cult
ben 10 fan spotted
please share prompts if you can
hmm would you believe if I told you this was nano banana?
Prompts:
"Turn this guy into a mascular buff, flexing, taking a selfie, in photorealistic gym but keep him in cartoon artstyle"
"Turn this guy into a drug addict, in a messy house"
For 2d to 3d:
"Turn this into a figurine placed at the office desk, with office supplies and computer in the background, blurred, closeup view of figurine"
"Turn this guy named Overflow taking a selfie in Wales, with photorealistic backgroun of wales"
"Overflow from the image depicted as dead, with his body dead on the ground, unconcious, with his translucent ghost form appears and visible"
Taking out the design printed on the T-shirt, show me the print
I really canât wait for nano-banana to be a selectable model in direct chat
It sucks having to roll randomly, sometimes many times, to get it
Draw a bounding box showing where the omnitrix symbol is located
can we agree that nano banana is an insane model
it was once a humble model with humble name
until we saw its true colors
Animate me an image of Tokyo's city skyline. Give a neon aesthetic, add futurist advancements.
can we use 2 image at same time for nano banana ??
since im not getting any result with that from nano banana
yes
like i was telling if u can share one eg so it will help me
like i told him to replce
or such but it dont give accurate result and use gemini for tht
but if use just a single image it will give result with nano banana
I think I told banana to draw their first meeting.
isn't it the image of guy i know
Varun Mayya
i mean you don't have his permission for it ..
it paid off
Anyone with knowledge would mind telling me how could I create image/videos similar to these?
/list
internet explorer 5 splash in skeuomorphism
prompt?
is https://nano-banana.org/ scam?
Nano Banana â preview AI for inpainting, outpainting and background replacement. Mask-free, text-guided edits for product and creative workflows.
create a random video and it must have sound music or anything goes it should be go viral
Please read #1397655624103493813 for instructions.
Just made these ones. Itâs really good at details especially of famous brands. Google is using itâs database pretty good đ
kk tysm!!
I did this with flux kontext (test)
agreed
btw what are your thoughts on where nano banana came from? which company made it or which image model is it similar to?
looks like a google model
yeah it seemed that way to me too
though the imagen 4 available on google apis doesn't seem to be up to par with it
maybe they made some advancements, otherwise this quality is incredible
2 strong evidences
- it labeled Swampfire as HeatBlast incorrectly which is common in Flash model
- asking who it is would say "I'm a large language model by Google"
hmm
also, adding anonymous models to lmarena is making people drive up the votes right? otherwise people might just keep using direct chat
smart of them to do that
Also Gemini 2.0 Flash is the only model that can do rudimentary visual reasoning, seems to inherited with nano banana
Try like doodling something then put a text inside the image "Make us do this", it will extract the task
hey @flint anvil did u get any better results while using 2 characters ??
Depends, if its too complex, it may not correctly generate
because im getting near accurate result in one character
both man sitting on an elephant in a forest
like this was the result with the prompt
2.0 Flash was severely flawed and broken, it feels comparable to like 2b or 4b models when it comes to image generation and editing especially prompt adherence
Original vs generated
wuts dis channel for
Talking and posting generated images of nano banana
It's only available in Battle mode at https://lmarena.ai/. This channel is meant just for discussion related to the model.
Okaay tyysm!
Ralistic is scary
yes hahahaha
Realistic is hilarious
I've been tryna use nano, like 20 tries on image to image with no luck. Any suggestions?
I keep getting it
Just luck
This AI makes stuffed tooys
Make it look like a casual mirror selfie of a random young person wearing this T-shirt. The style should be Y2K streetwear with baggy jeans. It should feel natural and unpolished, like a photo someone would post on Vinted to sell their clothes.
Amazing result
I thought that was a real person
I thought you extracted the tshirt lol
oh it's also possible
Yeah
it's even very good at it
It looks a lot like the one in the picture
it's the same
Yeah
LMAO
to export the tshirt
ok I'll do it
smaller print but pretty good output
yes it's pretty
Tried that with some random pants from pinterest.
Looks real realistic
if it looks very ralistic
It's good at showing the rest of the person's body.
What I did was ask him toI want these baggy jeans to be black and to be worn by a 19-year-old girl.
Bro
lmaooo i luanched a trend
better than Aleph
yess
Inprove Quality vs original
broooo đ
Hello
Which model did you use?
I would like to have old black and white photos made in color, which model is good?
vote the one and it will pop which model it gave
Okay thanks
Nano banana implies the existence of a regular sized banana
Imagine, that'd be an awesome model
You know why I like this model because it has good image quality to improve the quality of the photos
Lmarena quietly adds mega-banana whats the first thing you do with it?
Ayeee, Paulie
First I ask you to correct the color and improve the quality and then color the image.
Although this is current, I asked him to improve the quality.
The first imagen is better Quality
Point where is the fracture based on this imaging, generate a bounding box
this can revolutionize healthcare đ¤Ż
and accessibility applications đ¤Ż
@rare trout it's new to me, you turn on the battle and then on the logo image and send the photo and ask that the photo can be in color?
Well, the box should be smaller
Nano Banan can't unblur images
T shirt generated
I'm pretty sure it can, sometimes it just bugged
why did you select nano as better if it could not unblur the image?
But the quality of the model is good, better than other xomo topaz models.
or was that just outline
I really hope that the released model isnât censored and crippled beyond belief
Thatâs the problem with these closed source models
That may happen since it's google
They lure you in with amazing and fun things and then censor heavily when the fun gets started
What happens is that very poor quality images deform them.
Look what OpenAi did
All that fun stuff with GPT1, blew up on X with all the memes, then got neutered beyond belief
This is why Iâll stick with open source models
No censorship, guardrails or corporate handholding
The bad thing about chatgpt is that it changes faces
I wonât pay to use Googleâs model if it refuses even the most mundane prompts
Thatâs a waste of my money
ChatGPr changes its face for security reasons
oooooh interesting okok
Yeah
seems crazy how a diffusion model that's built to denoise can't denoise but generates pretty good images
And now OpenAI has made gpt5 more sycophantic like 4o because of all the complainments of it being too robotic
yay.
Yep
People ruin everything
Youâre better off sticking to open source stuff
If youâre gonna pay for something it should be able to do what you want it to
But I digress
personally, I like ai that doesn't mince words and gets to the point
Speaking ofâŚ
but all the LLMs these days add too much fluff to their replies, makes me want to spend some time making a reliable system prompt
One of my GPT1 things. Nano-banana canât do this
same, i need it to be a slave, not a friend. it's sad to say, but it's just a machine
true, i couldnt see the fracture myself
Is this good?
Yeah
Okay will try tonight
the results are good?
I suppose Iâll show my nano-banana stuff
It does give me good results.
Some testes with Nano Banana
Original is on the left, My nano-banana version is the right. Itâs Fei Long from Street Fighter.
Good
is this model consistent
Realistic pommi
Generate bounding boxes pointing to objects
no doubt this is gemini 2.5
it was able to reason 3d
has realism
nano banana
đ
Sometimes it does some weird sht. Did you try it again?
Hmm, few details are missing, incorrect eyes, wrong hairstyle and isn't she Japanese?
She lost her distinct hairstyle
Well
The model doesnât know sheâs Japanese
Itâs going based off the image itself
The anime one is real bruh
Ah I see.
@crisp rain Uchiha madari
Add a blood stain on the ground like someone was dragged
Yeah
I keep getting blurred backgrounds in portrait images. I've tried "background not blurred", "clearly visible background" and "clean background" but no luck. Any suggestions?
Huh
Iâll tell you another reason
I think how people write English affects outputs as well
Iâve noticed a lot of odd things as a native English speaker
Thatâs SĹsuke Aizen
Yeah, he's the author of the manga bleach
Interesting. So if I write the prompt in Japanese, I might have a chance of getting Japanese originated images?
yaer I made this with flux
You should definitely try that but I think so đ
Kyoka Suigetsu lol
Alright. Thanks for the ideas
Heâs not Tite Kubo!
You need to be very specific with flux
Flux kontext
First I wrote to take off her hair
Yeah he's not tite Kubo, tite Kubo is under his genjutsu since if he writes the story himself people would be freaked out seeing a pen move on it's own
second that he gave him a white beard
and put him in a suit and tie
I want to change my hair color to blue
Original vs flux kontext
My god, this is the best result i ever seen! "Recreate a cat with the mouse cursors
"
"Zoom out this picture like it was taken 10 meters away"
wtf is that
this model looks crazy good... kudos to whoever cooked it,
it has some limitations but i think it's new sota on image edits
99% chance this is from google
I dont see the fracture lol
Why I can't edit images, it only gives me text
Make sure this box is checked
Yeah, I use that.
It's telling me to try again later T_T
I think the Taiwanese broke it
does it work with 2 input images? Like put X in the scene in Y?
It keeps saying "Something went wrong while generating the response. Please try again." when using image to image
Hmm looks like we may have an outage going on
Team is aware of the issue and working on a fix!
Someone has time to do a little tuto thank you
rematerise cette image et ajoute des dophins au fond
Why is there a channel for nano banana
To discuss nano-banana!
But there is no channel for any other image modelđ¤
Is this model so good that it deserves its own channel?
It's been getting a lot of chatter so we wanted to make a temporary channel for it. Just helps the flow of conversations a bit better.
also the fact that its a stealth model invites a lot of speculation
City skyline of Bangkok Thailand but neon and futuristic
howcome I keep gen in battle but can't see nano banana model
It's there, I just got it 
same!
: O
but the worst thing about nano
is that it's not stable
like sometimes it's super good
and sometimes it's the worst
any other site can directly get nano banna
don't think so
nano-banana's generations?
yeah
like sometimes i ask for simple stuff
in one run it does it
and then it does some random shiot
sometimes it even gives me the same image
with no edit
yeah wait
Edit this picture so like it was taken 10 meters away, the person remains the same and all, just like take 10 meters back and take the photo.
"Zoom out this picture like it was taken 10 meters away; like we can see the person fully standing."
other models do it
but nano struggles
he changes the guy
if you get any good result, send
look at this
wtf is that lmao
it didnt change anything xd
even gemini gets a better understanding
so yeah, it's good but clearly has limitations
Can I have the original image?
(also is that a real place?)
it seems like image larger then 2024 won't get nano banana
nano gave me this for "Make a drone shot of this picture.
"
it's much better
but i had to remove the "30 meters away"
it has problems with distances
comfirmed
this is the result I got
Yeah this was the result but with closer instead of further away. Interesting
(side note I still don't think this is a real place ^)
thats great
it's so beautiful :p

daily motivation to get rich
okay i discovered nano
is more accurate with "feets"
instead of meters
It's currently accessible through Battle mode on our site.
You ask what you want and if you're lucky, you'll have it
thanks
but how will i know which one is Nano Banana?
or by default it is nano banana>
?
okay
and you keep doing it until one of your votes is nano banana
ah okay got it.
sure i am trying, thank you so much eikichi for your help. Appreciate it
no problem!!
Thank you for helping explain @rocky grove !
got this,
is it good?
nano banana is cool really
its awesome
Yeah that's great looking.
i mean i would love this tool to have it
this gonna make visualization so much easy
yes!
do you think the code name
has actually a meaning?
like "nano"
or is it random?
đ
it's random
i think it have much more potential than nano
"edit this like it was taken in a very sunny day and the sun light is coming through the window."
cool result!!
yeah... it's really better the runway aleph
runway is very good at making good demos
but their product is always bad tbh
it always seems unfinished
i really think it's the worst company in the field
true
but i think... they try to bring ideas and other ai make it perfect
i mean.. maybe for some users it's good who are just using it for fun
yes, i am a graphic designer for past 10 years, till now flux is the best editor for me... but now i think things will change
"Generate another picture in this exact style, like another place."
yah for fun and experiment its good
still good overall
yah
okay
but when will google release it?
google must create a entire new plan for Creators separately
i too think so
and can they add it to their google studio ?
when it will be out
yes they will
for some reason banana struggles at zooming out or drawing the rest of cropped images without dramatically changing them
well... they are working on whisk too right?
yes
but sometimes it works
that's what i don't understand about it
it's not stable enough
if you put them on a white background and ask it to fill in the blank space it sometimes does better
its just mind blowing
really preserved the logo style
it worked really well
yes
its awesome
kinda messed up the face actually
yeah
anyone knows any image creative upscaler other than Topaz bloom and magnific?
easily photoshoppable but yes
runway upscale is good
Astra?
like if you want higher resolution
okay, is it better than topaz?
wow
ok... here i should say... it did great
i'll rerun it
the generated image is crazy, looks more real
yeah hahaha
here when I ask it to zoom out and do the rest
when I ask it to fill in the blank space
turned it into illustration .. .far away from original style?
it messed up the butt/hips but that's because of my wacky proportions
yeah, bottom one did better
it just does poorly at zooming out
guess the real image?
looks more real than the real one
xdddd
damnnn
the best way to test if it is good at keeping face consistency
is to ask for edits with celebrities pictures
since we know how they look
yah
we will be able to tell if the new facial expression is accurate
lmaooo
"Make the person brush their teeth"
it looks like him
left one is generated?
clean face
yea holding brush differently i guess
nah
the face
isn't the same
zoom in ull see
eyes
hair
yeah
but once it's out
i believe it will be much better
cuz well be able to have settings
etc
more runs too
waiting to try it out.....
well... waiting to roll a perfect dice if you know
which is generated?
suit one
đą đ
making me nervous....
the seminude one is generated
nooo
gg
omg
suit one was generated
yo boy.... now hand me my prize
original image
lol.... it's sweet tho
cool,
woo... i love that.... i mean i can use this idea
Nano banana seems to struggle with fingers, blurring and proportional zooms. Anyone knows why?
my pinterest is full with weird but cool ideas
no one knows why sadly
really.... i wanna see
đ i was thinking the same, how did you come up with such an idea
its seriously crazy
what's your profession?
So the word âflirtingâ violates TOS? Get outta here with that nonsense đĽ´
a lot of random stuff
i'm an ethical hacker lol
This is exactly what I mean about censorship that doesnât even make any sense
đ now i can understand
make it like mars... or kinda something strange stuff
turn the mountain into volcano
my prompt is very hard tho
"edit this photo like it was taking during the night, we can see a lightning strike the mountain, make the ligthning realistic, subtle, a little blurry. we need it to look like a normal real photo."
it's super difficult
to make lightning strike
only god can XD
a 30 year old gym enthusiast beautiful indian women eating chyawanprash from this bottle in a spoon and promoting it
its mindblowing
yeah but text isnt super good
i can create months of realistic social media posts in a day
amazing result in my opinion
no other model come close
lightning looks very realistic
cool, but look at those three UFO lights
it's reflection
yah the cloud lighting
reflection of?
good job
probably something inside the house
model hallucinated it
look at flux lol
nano knows more than we know
its outdated soo soon
đ
so how do you re do images? with the redo button in "assistance A or assistance b", or copy the prompt and the image and start creating again?
until you get nano banana
redo again
copy prompt
and image
okay,
thanks
for how long LM Arena can let us do things continously?
an who funds it to be free?
ai companies
cuz then we know which model is better
it's also good for them
for monitoring
Do the LMarena devs/staff know who owns the Nano-Banana model?
pretty sure they know
@agile oxide Do you know what the limit is for nano banana, battle to edit photos,? the video arena is 8
which is better
9
10
2
nano banana
Make this design with the same colors, text and visual identity on a large carton of the product.
Prompt?
just type like do better in skeuomorphism it always turns out cooler
Make a 3d model of a guy whose passionate for gym and vlogging
You can't use nano here lol
Go here, send your prompt
Vote for the best image
And if you're lucky, you'll get Nano Banana
sorry
xddd
best model ever
"Recreate this image but replace the laptop with an iphone, like it was found in stone. Make sure some parts of the phone are still in the stone, put some dust on it etc so it's realistic like it has been dug out the ground."
park this cadillac escalade next to a low income house in a poor neighborhood
this was great
lmao
crashed b2 spirit
nano banana da đ no cap
How to used nano-banana?
On battle mode on lmarena.ai . It will appear in battles randomly and you cannot use it directly as it is a pre-release model.
its a test.. I want to use for my patients.. I do nose jobs
like mass simulating leads
oh got ya, good luck with that dude
Uhh, just know that every prompt and conversation you have on LMArena are stored by them and with a probability to be put in a public dataset.
Tks!! Good to know that
left image initial image . prompt " put the name of the company who created your model on the sign "
holy nano banana
I prefer human artist
I guess i figured out how nano gives the output from its style and i can just regenerate the image until i find the perfect one
is that lapwing?
looks like so
I feel like when I upload more than 1 image of myself, nano just selects one and use only that image
it has a preference for the last image sent from what i've tested
wtf ?
what? you don't like peak?
That is actually w probably
nsfw?
Not so w anymore
Hey
Was nano banana taken down ?!? I have not got it after like 40 tries now, what the...
I just made these images 5 minutes ago, so it was up then atleast
đ
It's the only model I care about out of the bunch
Oh... I see what the devs of this site did there
đ¤¨
try using a new chat everytime you generate an image
The nano-banana model won't even appear at all if the system detects and suspects that the prompt might be nsfw.
I did
But now I experimented a bit and found the actual root of the issue
It's not that the site does that, it's more likely that the model provider itself detects it as nsfw, rejects it and lmarena just routes to another model
if it was the site it would just reject the request itself
But there was this time I generated where specifically a model detected nsfw, rejected, but it didn't route to another model, instead that ouput panel on the left straight up showed the failure
what's that
that's what I mean
The nano-banana model won't even appear at all if the system detects and suspects that the prompt might be nsfw
if it detects NSFW it errors out
ah ok
Yeah but in my case for nano-banana, it didn't do that
like this
So the nano-banana doesn't even appear error like this -- it simply straight up never appear at all
if nano-banana showed this similar error, I would have known about the existence of the issue, but it didn't, I had to experiment to figure out why
@agile oxide check this mate
Am I the only one who almost never gets nano banana? Is there any way to increase the chances or something?
In my case I notice the nano banana is much less likely to appear if the prompt might be detected/ considered as nsfw
Just something I figured after experimenting around
It specifically won't give error like other models - it simply won't appear at all, which is the tricky and confusing part
my prompt doesn't even have any kind of problem, because I'm asking to change the color of my iphone, I think it might just be bad luck
It's a shame that it doesn't support multiple images input...
For example, I want output of character in image A, but in artstyle of image B
the artstyle is very specific, not something common like "pixel art", so it can't just be prompted by text... the character is also very specific aswell and the training data doesn't know it
yeah
it technically does support multiple images, it's just not very good at that
like this:
I may have went a little overboard with the amount of images
the less images, the better it performs, but tbf like 70% of the time it just doesn't work when you use multiple images
it's clearly undertrained on that
Flux kontext test
Look amazing
What AI is this
Thanks
Also why is there no login ? If we clear site cache data everything will be gone ?
We don't currently have a login system, but this is very much on our radar. This would be a big improvement to have this feature.
Yes, I would like to sync my prompts-outputs across different devices, and like sometime in the future I come back and wanna see which prompts I used for which outputs, they would still be available
Yep - we're very much on the same page!
đ
Btw, I'm new to this, but are you guys running this site at a loss or how does it sustain ? This site seems to have tons of models, and somehow doesn't seem to have a payment/ subscription option ? Normally most AI sites I see else where charge very expensively after petty few limited free tries
I'd check out this blog post: - https://news.lmarena.ai/new-lmarena/
That was truly an inspiring read
I personally like âLMArena will stay open and accessible to everyoneâ and âLMArena will always be a neutral, open, community-driven space for evaluating and improving AI reliability.â
đ¤¨
Yeah after you voted the name didn't appear? After refresh what happens?
Good! Glad to hear it!
This site is honestly one of my favorites
Please donât change
Ok after refresh I see it now
I only regret one thing
that I didn't discover this site sooner
Making it free and open for everyone really helps out a lot
đĽ˛
I only know about it when people tell me this goated model nano-banana is only available here
Super appreciate all the kind words! The team thanks you both!
Youâre very welcome đ
Keep up the hard work guys!
I didnât mean to sound ungrateful about Legacy earlier đ
I just loved using it and used it almost every day
It's super understandable to see why not having access to it all of a sudden (esp without some kind of heads up) wouldn't be a fun experience.
It's a bit ironic: sites that advertise aggressively to make people know them are most of the time đŽ
But actual great sites like this are like hidden gems among rocks, they don't advertise themselves because they don't waste their fund, so it's also kinda harder to know their existence to begin with
Just like how uBlock Origin is the GOAT of all ad blocks, this site is of that equivalent for the AI-enthusiasts in open source community that don't have strong local hardware to run stuffs... like me (only a weak laptop with 8GB RAM)
Btw
in battle, is there any quick rerun button ?
Cant be more true.
Click on that cycle button.
But not possible if you have already voted.
yeah
Like I input an image and a prompt, and both results are terrible, I wanna rerun that same input img and prompt
after voting ofc
After voting? You cant rerun the same model.
Try pasting the same prompt + image again
Yeah I wish for example pressing up arrow button is a hotkey to do that
getting the input prompt & img right above previously quickly
Good idea. Share this in #1372230675914031105
Pineapple will most likely take notice.
Done â
Btw does this rerun the same model under the hood ? Or a random different one ?
Because if same then that's useless for me
Before voting, same model. After voting, different model.
I thought after voting the rerun button disappears ?
By that, I meant pasting the exact same prompt
After voting.
Btw, nano-banana is the GOAT, but why does it fail to understand such a simple prompt to use a different artstyle ? Like all other models understand when I simply say to change to anime artstyle, 16-bit pixel artstyle, etc... but nano-banana would just output the exact same input image as output ???
We dont know. Maybe a simple bug, or a technical error. But thats why we are here.
Its a pre release model after all.
Many issues.
Do we know which company rleeased it ?
If you ask it to generate a image of which lab created you, it generates an image of Google. So Google most likely.
@humble verge
Fr man, they are at the top of everything
They got imagen, lyria, Gemini, genie and what not
Each of them excel in their field
Wow, looking back
There was a time I used to think the tech giant Google was underwhelming for losing to a new underdog AI startup OpenAI (when GPT-4 was much better than Bard/ Gemini)
And as time went on, it became apparent Google inevitably wins the long run with their practically unlimited resources at disposal
OpenAI has the tech and short term head-start momentum but not the resources for the marathon Lol
Exactly.
No matter what field, those with huge funding, data and workforce win at the last. History has proven this.
Yeah itâs great until you realize itâs going to be heavily censored, guardrailed, handheld and neutered upon release đĽ´
Nano banana?
Itâs Google
Itâs also not an open source model, so yes
Theyâll control what you can prompt
In the Future? Maybe
Yes, I get it we all stand with the open source homies, but the more competition between those companies the better for consumers, those giants are the ones pushing the maximum limit boundaries, which the open source community eventually try to catch up with
Like imagine if OpenAI never showed ChatGPT to the world
True.
And yet ChatGPT is hella censored
If it isnât making money it isnât being made
I think GPT 5 has weak guardrails. Thinking models are the ones censored heavily. Their A/B testing provided fruitful results.
Well
Peak AI behavior.
Are you serious đ¤Łđ¤Łđ¤Łđ¤Ł
What lol
Lmfao
Ayo mods be watching us
Tell me how any of that makes sense
đ¤đđ
Is this a Chinese run server or something?
Welcome to the corporate world
Guess we are also not allowed to mention đ
That is ridiculous.
Let me say something
Image gen is literally glorified fan art
Thatâs all it really is
Or really amazing Photoshop
It works like both
Yeah, that is its main usecase, I make use of it to create various images of characters from anime I like. But with the ongoing situation, this might become restricted in the near future.
Which is why I make stuffâŚon my local hardware
Which is why open source models will be the way to go
Ah I see, Stable Diffusion?
Donât let corporations control what you can make
I actually used to be part of the group that established Stable Diffusion from the beginning before it was even called that
The group was originally called PANCE
Emad was also there
We trained the first Stable Diffusion model for them on Discord
And got paid absolutely nothing
This was years ago
I have old messages from those people
I had no idea Ai would blow up as big as it did
I thought it was a novelty thing at first
I was also a mod briefly on the Stable Diffusion Discord
Was he a fraud ?
They made me a mod when they started the site, and then moved in their cronies and kicked us all off
Emad?
Yeah
He had good intentions at first
I think the money got to him
But he was also heavily in debt from other investments
I don't know all the dark stuffs behind the scenes, but man, SD model really laid the foundation for the open source community of AI art
Guess what it used to be?
Disco Diffusion
I just read my old messages lol
Theyâre from 2022
If you could rewind time to back then, would you do anything differently ?
Yep, and because of the little people that got paid nothing, like me
I couldâve been rich and successful myself
If they hadnât screwed me or others over
đ
They filled our head with lies
About how we would make a difference
And how weâd be successful
I think they just wanted us to test the models

