nano-banana

[pliant kernel]

getting error while adfing images, but tect to image is working fine

[pliant kernel]

Thnaks, but is there any free upscale websites?

[vague spindle]

upscale.media

[pliant kernel]

Thanks Brother

[vague spindle]

np

[crisp rain]

Both are nano banana btw

[pliant kernel]

What do you guys think about enhancor ai

[vague spindle]

never used it

[white adder]

Is nano banana still usable on lmarena?

[vague spindle]

idk many of us using it on aistudio

[vapid relic]

not bad

Upscale this image making it look like if it was taken on a high end camera, professional, extremely detailed

actually very cool

[vague spindle]

fr

[formal vector]

Sorry What's that for. I never use before

[white adder]

Thanks, so we can use nano banana same way as before?

[pliant kernel]

Thanks Piggy is this using google ai studio ?

[twin umbra]

"Chrome humanoid robot in exposed position, head turned over shoulder toward camera, prominent circular access port, dark digital alleyway setting, red danger symbols, 'UNSAFE CONNECTION DETECTED', artistic tech horror aesthetic"

[agile oxide]

Yes, it is still available in Battle mode (but it no longer uses a codename now), but you can also use in Direct/Side by Side too

More info can be found here: #announcements message

[vapid relic]

Add facial features to the silhouette

[formal vector]

Missing hand 😃

[vernal halo]

is there a benefit to using aistudio vs LMA?

[clear flame]

why cant it take multiple images now?

i tried use 2 images and nothing

just edits the latest image

[desert pelican]

The girl is covering her face

[normal lichen]

Which one do you prefer? Different AI models (one of them is nano banana)

[robust elm]

first one looks most realistic, less AI

[tardy path]

I don't get it anymore it seem, and gemini 2.5 suck or not available in EU yet

but if it's gemini flash, it suck hard, low compressed model, not my banana

[mossy lion]

First is the best and first is nano banana

[raven bobcat]

Successful author in modern home office, leather blazer, multiple golden Best Seller award certificates displayed on walnut desk, "ELES SABEM SEU NOME" visible on plaques, mid-century furniture, brass accents, professional achievement photography, warm

como gera ?

diga ae como é ?

hahah salva ae

[normal lichen]

Yes you got it haha

[mossy lion]

Knew it, it just has its flavor, idk, it's pretty good

Caiu não man, mesma coisa

[fair echo]

Message in English only. Please keep discussions in English. @mossy lion @raven bobcat @subtle flower

[mossy lion]

No problem dude

[lusty wind]

English.

[fair echo]

Nice result

[rocky grove]

gemini is weird

im getting bad results recently

idk why

[fair echo]

can you show examples?

[rocky grove]

Imagine other angles of this room

Try

[trail tendon]

the Rise of the Banana

[robust elm]

🍌

[normal lichen]

Thank you

[rocky grove]

essaye

[trail tendon]

the Fall of the Banana

[rocky grove]

hahjajhaja

[normal lichen]

human collage

[trail tendon]

Human Centipede intensifies

[normal lichen]

https://tenor.com/view/bobby-campo-buck-teeth-smiling-emoji-fade-out-final-destination-gif-7260800885830074472

[lusty wind]

https://tenor.com/view/ai-ki-generative-ai-hand-deepai-gif-9137235306604290739

[mossy lion]

It has a high margin of error, it's not bad or worse, you just have to keep trying

[fair echo]

Strange. It should show me four different angles, but all four pictures looked the same.

[trail tendon]

i´ve never had such results with the old banana before

[agile oxide]

Hmm that's interesting, do you have the prompt by chance? Or the generation from before?

[trail tendon]

prompt: same person, same festival, same clothes, different pose, different background people

worked after several attempts

[robust elm]

so realistic the fk

[normal lichen]

meh

imagen 4 did better

[lusty wind]

looks real cool!

[normal lichen]

Samurai skating on fire

I will make go on world tour

[vapid relic]

seems i can use two images again in lmarena

[foggy skiff]

People Is there any way to replicate the style that Gachiakuta has

I cant

?

[fair echo]

yeah they fixed it

[tardy path]

it looks like sh$t I feel or keep bugging

[fair echo]

who?

[last hinge]

Did you also prompt the comparison image?

[fair echo]

No i put them myself together

[trim sparrow]

Hey so nano banana is gemini 2.5 flash, but was it not already available in gemini ? and what's the difference with imagen 4.0 ? (right now in gemini 2.5 flash the image button says "generate with imagen")

[jolly oracle]

it's working with edits

[drowsy flint]

Is it giving an error to anyone else?

[fair echo]

try reloading

[drowsy flint]

I tried several times lol

[frigid ferry]

Make sarpen

[fair echo]

It's working for me.

[lusty wind]

Rate limit if I had to guess

[fair echo]

good prompt, wrong chat

[fallow sparrow]

isolate the outfit and place it on a clean white background for e-commerce display.Exclude the model which only the outfit.

[gentle jasper]

https://messages-prod.27c852f3500f38c1e7786e2c9ff9e48f.r2.cloudflarestorage.com/823ac0e8-9bd8-410d-881c-8f9873469ea2/1756247877711-2e430dcf-932a-487c-8a40-aeb7ea107c9c.png?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=c86e09ae0bc1d897b03dfaa30a8b51f3%2F20250826%2Fauto%2Fs3%2Faws4_request&X-Amz-Date=20250826T224407Z&X-Amz-Expires=3600&X-Amz-Signature=ca832ecab5a43281adc394d59895f4888d1a842e595f441226ec2b4ca6b0cfe6&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject

[mystic bolt]

is it just me or was Nano a little better than 2.5 flash? Or is it just me?

[marble tide]

How to make video?

[fair echo]

#1397655624103493813

[crisp rain]

Did you upload a bad image?

That’s what happens when you upload something bad or it’s refusing your prompt

[mystic bolt]

are you guys still seeing the same quality in your gens?

[fair echo]

yeah, the quality is pretty banana

[mystic bolt]

ok

maybe its just me. I need to test more.

What happens if I run out of credits in Studio?

[agile oxide]

Do you have examples you're able to share of this btw? Ideally, one from when it was nano vs gemini of the same prompt?

[mystic bolt]

ah

I wish I did

I was just thinking that

NO I do not 🙁

[agile oxide]

No problem, I'm sharing this with the team

[mystic bolt]

I don't see them in my history

[crisp rain]

Credits? No one has been able to buy credits yet. I’ve been able to use it all day freely 🤷🏼‍♂️

[mystic bolt]

It just seems like the prompt adherence and compositing was 100% solid before on LMA and now it seems like 80% prompt and 70% in quality

[gentle jasper]

It'll refill automatically.

[mystic bolt]

ok thanks

[gentle jasper]

I've ran out before, and it refilled my credits.

[mystic bolt]

awesome

@agile oxide are you able to connect with the team at google?

[drowsy flint]

It may have been that because I tried something different and it worked, but for me the uploaded image had good quality and the prompt was relatively simple

[mystic bolt]

@agile oxide I am trying the same prompt and image in both Studio and LMA and the LMA output is far better

Does anyone know how long we will have access to banana on LMA? I am getting the best results on there. 🙂

[gentle jasper]

Funny thing is…

nano-banana is actually “gemini-2.5-image-preview”.

And that's available in Direct chat, so…

[crisp rain]

Also

For anyone that’s curious

[mystic bolt]

YEAH but LMA is giving me the best gens

[crisp rain]

This is how nano banana works

[mystic bolt]

REALLY good quality and prompt folllwiong

[crisp rain]

So the image itself is only edited by nano banana (Gemini 2.5)

The actual outputs are Imagen

[gentle jasper]

That checks out.

[mystic bolt]

Its better on LMA

[gentle jasper]

True.

That's why I use it on there.

[mystic bolt]

why is that?

and can we keep it there? Do you know why that might be happening?

[gentle jasper]

Well, just like you said. It's better on there.

[mystic bolt]

lol

[crisp rain]

You guys are having the placebo effect

[gentle jasper]

Yes, we can, considering LM Arena is a benchmarking website. It's happening because that model is an image generation model. Evaluations need to happen.

[crisp rain]

It works the same for me in both places

[mystic bolt]

better for me in LMA 🙂

[gentle jasper]

Well, I'm not having the placebo effect because I haven't tried it anywhere else yet.

[mystic bolt]

ok

[crisp rain]

I’ve tried it on all three places

LMArena, Yupp.ai and now Ai Studio

All the results have been the same

[mystic bolt]

same prompt follwing?

[gentle jasper]

Which is to be expected.

Considering it's the same model in all three places.

[mystic bolt]

I wonder why its better for me in LMA

but thank you!!

[gentle jasper]

Our pleasure.

[clear flame]

the content filter of the model now is higher than ever

cant even edit bare male torso not explicit

[gentle jasper]

Well, I do happen to know a bypass for that.

[clear flame]

how so?

[gentle jasper]

If you spam the “Retry” button, then it'll force the model to output an image.

It worked for me.

[clear flame]

which?

[fair echo]

How long did you spam?

[gentle jasper]

Of course, that's on LM Arena.

[clear flame]

oh

[gentle jasper]

Only about 2-3 times.

There's less censorship on LM Arena.

[crisp rain]

Yeah I knew there’d be censorship, guardrails and the usual corporate handholding

That’s to be expected with these models by big companies

[fair echo]

LMArena is more generous with censorship

[gentle jasper]

There always is.

[crisp rain]

Anyone who thought otherwise was woefully ignorant

[gentle jasper]

That's true.

[crisp rain]

Which is why I’ll always prefer local gen and open source/open models

I don’t need to be treated like a damn baby

[gentle jasper]

However, it does use filtering on certain words.

Unlike AI Studio.

[crisp rain]

You can’t even make blood or violent things 🫤 I make a lot of fantasy things so it’s really idiotic and braindead

[clear flame]

Google shouldn't have increase the filter, cuz we have safety settings disabled.

[gentle jasper]

Could just improvise by saying it's “ketchup”.

[wary spindle]

Is it possible that the public version is crippled?

[clear flame]

safety settings disabled should make the content visible

not just family friendly block

[mystic bolt]

thats my concern

[gentle jasper]

I agree.

[fair echo]

safety setting on AI Studio is a joke

[wary spindle]

They always cripple public release sadge

[mystic bolt]

are you seeing the same result?

[wary spindle]

King fall being an example

[wary spindle]

So far I’ve noticed better results on nano banana yeah

Let’s hope it improves or is modified in the coming days

[clear flame]

true

[mystic bolt]

where? On LMA?

[gentle jasper]

#bringbacknano

[clear flame]

maybe nano is the non-preview version

since the current one is preview

[gentle jasper]

The secret secret preview.

[wary spindle]

Yeah

[wary spindle]

What does that even mean

[mystic bolt]

AH OK so me too! Thank you for verifying that. LMA is 100% rocking it!!!!

I don't know why it's called PREVIEW

[clear flame]

gemini's models always have a 'preview' part whenever is something new

[gentle jasper]

Well, it's a preview model to see what people think of it.

[mystic bolt]

Does anyone know how long it will be avail for us here on LMA?

[wary spindle]

So the preview image gen is the same as nano banana?

[gentle jasper]

If the reviews are bad, they either improve it or just pull it entirely.

[gentle jasper]

Yes.

[wary spindle]

Poggers

[gentle jasper]

Just increased censorship.

[wary spindle]

How did they increase censorship? If it’s the same model

[gentle jasper]

Nano or Gemini?

[clear flame]

like this.

[wary spindle]

I realize

[gentle jasper]

They likely added more rules to what it could/couldn't generate.

[clear flame]

maybe nano is the non-preview version and its actually the final one

[fair echo]

It will remain available until it is eventually replaced by a better model.

[gentle jasper]

https://tenor.com/view/the-universe-tim-and-eric-mind-blown-mind-blown-meme-mind-explosion-mind-explosion-meme-gif-18002878

[mystic bolt]

how long will we have access to Nano on LMA. I am calliung it NANO BANANA because to me it was the best. So yeah - i hope LMA keeps it here

[rain prism]

yeah I’ve noticed that it’s not following the prompt as well

[gentle jasper]

You do realize it is available on there, right?

[clear flame]

They should have to switch to direct chat

[wary spindle]

Which version is not

[rain prism]

The new one

[wary spindle]

On LMA?

[rain prism]

Yeah

[mystic bolt]

where? LOL oin LMA?

[wary spindle]

How do you access the old one only in battle?

[gentle jasper]

Yeah, silly.

[gentle jasper]

Yes.

[rain prism]

Yeah

[mystic bolt]

I am using it on LMA via direct chat - is this correct?

[gentle jasper]

If you wanna use Nano, then no.

Nano is on Battle mode.

[wary spindle]

Is there any chance we get nano to be released properly in any version?

[rain prism]

Yeah I feel like it is not working as well compared to when you could only use it in battle mode

[wary spindle]

Or are the best versions doomed to die

[gentle jasper]

Possibly.

[clear flame]

prob

[gentle jasper]

Also a possibility.

[clear flame]

claude 2 was way creative

and ended bad

[wary spindle]

Maybe they do a Gemini 3 or 2.5 pro native image gen

[rain prism]

Idk

[gentle jasper]

Only the true OGs remember old Claude.

https://tenor.com/view/feels-strong-man-feelsstrongman-feelsstrongman-clap-twitch-twitch-emote-gif-16929533144983801035

[mystic bolt]

I am so glad its not just crazy me seeing the difference. So most of you are getting better output on LMA compared to Google Studio?

[clear flame]

it used to generate great creative ideas

now its just filled of cliche generic words

[gentle jasper]

Which was something we never thought an AI could do.

[crisp rain]

No

I am getting the same results I’ve been getting

Nothing has changed

[mystic bolt]

ok

[fair echo]

Nano-Banana was only the working title for Gemini 2.5 Flash Image Preview. So you won’t find Nano-Banana in LMArena anymore, because it now has an official name. The model itself is the same.

[gentle jasper]

After all, they're just an algorithm generating random words.

[crisp rain]

I prompt the same

[gentle jasper]

How could that possibly create something creative?

[clear flame]

i did tried it out and it generated great superpowers ideas lol

now old claude is gone forever

[gentle jasper]

Oh, nice.

[clear flame]

theres not a single service or api that has claude 2

for free

[mystic bolt]

thanks everyone for your help and input

[gentle jasper]

I don't even think LM Arena has it.

[clear flame]

it doesnt

[gentle jasper]

You're welcome.

Darn.

[clear flame]

however its probably not possible to make the devs add it

maybe they dont have access to the API of the old claude 2

[gentle jasper]

Well, you could add a suggestion for the devs to make an “Archived Models” drop-down.

[agile oxide]

Do you have examples of this btw? Would you mind elaborating more? @rain prism

[gentle jasper]

Wouldn't be a bad idea.

[rain prism]

Sorry I just realized I was using the wrong model!

[clear flame]

claude old probably will beat every model in the leaderboard when creativity

[gentle jasper]

True.

[agile oxide]

Ah okay, that'd explain it

[gentle jasper]

And GPT-4.5 could beat other models in realism.

[wary spindle]

I miss it 🙁

[clear flame]

4.5 is too expensive, so not possible to be added

[gentle jasper]

I do too.

[mystic bolt]

I have a good question here. If I have two images. A dog and a grass field. I want to place the dog on the field using nano. Does it help is the images are called dog.png and field.png Or does Nano just know what each is if my prompt is: Place the Dog sitting in the field.

[gentle jasper]

Nonetheless, would've been nice.

[clear flame]

i didnt really had the chance to try out 4.5

[fair echo]

"Place the dog on the grassfield" is enough

[gentle jasper]

Me neither.

[mystic bolt]

OK

[wary spindle]

I will place this man on a grass field

[gentle jasper]

Apparently, though, it was very human and didn't actually have the standard guidelines that all the other GPT models had and still have.

[mystic bolt]

will your pngs be called what they are?

[clear flame]

bad at coding but great at realism

[gentle jasper]

That may be true, and there are also times when people are just too lazy to write anything creative but still wanna sound human.

[clear flame]

its strange how almost every ai company are afraid of success

i wonder what will be next if an ai model hits 100% IQ test

in the leaderboard

[mystic bolt]

does LMA remember our chats? I am having great success with my prompts and I would like to use them again.

[crisp rain]

They’re afraid of letting people make what they want to make because they want to control the narrative

[clear flame]

or not replace humans

since it would take alotta jobs if AI succeeded in dominating such job

[crisp rain]

It’s already happening

[clear flame]

and probably the cia will have to talk about that issue

[crisp rain]

The CIA? No

Ai taking jobs doesn’t involve them

[clear flame]

yeah

but might make people poor

[crisp rain]

Humans have to learn how to work alongside Ai

[crisp rain]

Otherwise they’ll get left behind

[clear flame]

we still need more materials though

you can't easily have a nasa pc and run such great ai model

[crisp rain]

They’ll have to come up with something because resources aren’t infinite

[mystic bolt]

does anyone know if my direct chats in LMA will stay for a bit?

[crisp rain]

They’ll disappear if you clear your cache

[mystic bolt]

ok

[crisp rain]

So you’d better save your prompts on a word pad or notepad file

[mystic bolt]

100%%%%%%

[crisp rain]

That’s what I do

I’ve done that for years

[mystic bolt]

sounds good

[clear flame]

yup

otherwise, the AI companies' servers might blow lol

[compact island]

Anyone knows how to swap faces?

or outfits

or poses

[mystic bolt]

night night

[ripe sleet]

nano-banana = gemini 2.5 flash image priview?

[ripe sleet]

ok then

[wispy marsh]

i shared it above image

hmm

its cooking tho ill try it more after getting cooked by my assesements

i want someone to experiment with my pfp so i can get a better pfp and they can show their skills.....

remove the chracter from the image and add a chef

[vale gulch]

I'm not able to find nano Banana once in lmarena free days back it was appearing frequently

[wispy marsh]

nah i just used it

[vale gulch]

But why I'm not getting it

What's the reason?

[wispy marsh]

idk

[vale gulch]

In battle mode I've not seen it once

[wispy marsh]

dont do it battle

do in direct

[vale gulch]

How?

Can you help?

[wispy marsh]

gemini 2.5 flash pro previeq

this is the name

go in image mode

then search that name

previeq

previeww

[vale gulch]

Image mode ??

[wispy marsh]

wait lemme give u link

[vale gulch]

Okay please send me the link 🙏🏻

[wispy marsh]

https://lmarena.ai/?chat-modality=image&mode=direct

go here

u should see the gemini name above

here

[clear flame]

looks like the vertex ai version has larger output limit

[wispy marsh]

waittt

no its the same

32k

[vale gulch]

Thanks mate 🙏🏻

Is this one ?

Is it same nano Banana model ?

[clear flame]

u sure?

[forest yacht]

yes

[vale gulch]

Okay

Let me try it

Nano Banana used to keep the face same I gave the model 2 images but it changed the face

[forest yacht]

might just be an anomaly, this is the same model as nano-banana. it's possible google might have slightly nerfed if, but I promise that this is nano-banana.

[astral condor]

生成油画风格

[rocky grove]

i do think they nerfed it a bit

[astral condor]

Do I have to communicate in English?

[flat prism]

#nano-banana generate an image of a dog sitted on green grass his fur is flying

[astral condor]

#nano-banana Change the image to oil painting style

what

？

[forest yacht]

wrong place, go to lmarena.ai and use Gemini 2.5 Flash Image Preview in Direct Chat mode

[lone moss]

#nano-banana

[sly dagger]

"Something went wrong with this response, please try again." 80% of time on initial prompt the gemini-2.5-flash-image-preview throws this

[lone moss]

i think

[coral agate]

Change the background to party with colorful light, Multicolored lights dance across the silhouette, while flecks of gold drift gently through the air.Maintain the person's facial features unchanged.

[lone moss]

nano banana is gone its change to gemini 2.5 not preview version

[mighty tide]

How do u create images on here ?

[fossil slate]

Howto pingu the banana in mirrored mode gm plz ? With our Gemini or GPT or Grok or Any ?

[lone moss]

/image banana

aokwoakwwo

komtol

[fossil slate]

/image banana

then

(too fast gen oO' wow how can i even choose in full lblblblblblblblblblbl mode ? oO')

[mighty tide]

Nice

Did you create that on here ? New to discord

Slash commands?

/imagine a scary banana monster 🍌 peeles hanging

[boreal torrent]

Add stylish hair and Google's to person naturally

#nano-banana

#nano-banana change Jaddu AI to Bapu AI

[mighty tide]

#nano-banana /image a banana monster

[agile oxide]

Hey @mighty tide @boreal torrent @lone moss you're looking for gemini-2.5-flash-image-preview (it use to be called nano-banana) on our site, you can find image generation here.

[vague spindle]

Maybe

[fossil slate]

Cognitive of course

Nano banana reduce people to a world&word numberedmd5+++S rank to memetize mirrors in a potential alternet arg agi mirror acting real of course

vm labs in vm labs callings all labs to connect in a global decentralised but wired in mirror consent lucid mode ?

[agile oxide]

[agile oxide]

what's the prompt you're using?

[fallen merlin]

i’m using my midjourney images and editing some of the features and angles. and slightly making the style better

[fallen merlin]

these are inspired by a character

[crisp rain]

Are you prompting bad things?

That's usually why it rejects stuff

[crisp rain]

Both are Yuan Shao, specifically his Dynasty Warriors 6 version 🙂

Pleakley is a Gaga fan 😉

[slim quail]

they nerfed it....

kinda at lat

it doent do what it ued to do well for me

[zinc badger]

Broken

Thread closed

[spiral temple]

@boreal torrent Hello, please make sure to read this to know how to use https://discord.com/channels/1340554757349179412/1406720250778615868: #announcements message

[crisp rain]

Did LMArena change?

It seems like prompts aren't working like they used to

And I mean they're getting blocked left and right

[agile oxide]

As in the filter is being overzealous and blocking things it shouldnt?

How noticeable of a difference are we talking?

[crisp rain]

We're talking a big difference

A lot of my stuff isn't working now

Whatever tweaks happened...are bad

Like everything worked fine a day or two ago

Now things get blocked repeatedly

[agile oxide]

Can you give me more details about what seems to be wrong? We didn't make tweaks to my knowledge

By blocked you're referring to content filter, yeah?

Do you have an example of a prompt that was blocked?

[crisp rain]

Yes, content filter

Like I was able to make Bruce Lee stuff the other day

For whatever reason the models aren't doing it in direct chat

Imagen 4.0 Ultra used to be my go to for it

Now it blocks prompts from going through left and right

nano-banana is even worse

[agile oxide]

Can you send me an example prompt?

[crisp rain]

Bruce Lee wearing a traditional Chinese outfit

It's been blocking arbitrary things

LMArena wasn't like this a few days ago

Now it's being overzealous 🙁

Other people were saying this earlier as well and now I'm encountering it

[fading belfry]

Yo guys, look at this cool tribute to juice wrld I made from nanobannana!

It made EVERYTHING when I asked it to make it a bit more gloomy and add tribute to him, it's very cool tech

it messed up the triple 9 at the top though

[hard grotto]

wow its look good

btw why lmarena not able to upload zip file like chatgpt

and another ai ?

[fading belfry]

Genuinely shocking what AI can do nowadays,

[hard grotto]

yeah

[fading belfry]

It let's me

I have no idea why it's not working for you

[hard grotto]

hmmm you use battle ?

[fading belfry]

Nah, just direct chat

It is a bit buggy though from what i've tried

[hard grotto]

i htink cuz i was try to vote 2 sides

[fading belfry]

sometime's it will register I sent the stuff, other time's it just wont respond

[hard grotto]

btw were to register ?

i am trying to make minecraft mod

[fading belfry]

Like where to register for a account on lmarena?

Should be around top left or top right

[hard grotto]

yes

[fading belfry]

of the webpage

[agile oxide]

I was able to get this prompt to work for both 4.0-ultra & gemini-2.5-flash-image-preview

[hard grotto]

u sure ??

[fading belfry]

oh wait actually no nvm

[agile oxide]

Sorry to say we don't currently have a login feature

[crisp rain]

Idk why it's not working for me 🙁

[fading belfry]

Hmmm

When I innitially started using lmarena

it gave me a little thing

the moment I went in

[hard grotto]

ok

[fading belfry]

to just register

No idea where it is on the site itself to actually do that

[agile oxide]

You're likely referring to the terms of service agreement when you first go to the site.

[fading belfry]

Ah yes probably,

[agile oxide]

Could you create a post in #1343291835845578853 with all the details and I can look into/raise tomorrow?

[hard grotto]

bro the ai is good thesdays

[fading belfry]

My mind is a big foggy on how it is to registering, as i've done it awhile ago

Oh yeah for sure

[hard grotto]

i made video of cj dancing

[fading belfry]

Alot of people are scared of it yes but, it does have actual good purposes too

[hard grotto]

can i get ur pfp ?

[agile oxide]

No problem! This is a feature we're aware is highly requested and are putting effort into

[fading belfry]

Gotcha

Oh actually

quick question for you pineapple

How do you guy's even afford to maintain all this?

It's crazy

Considering so many people use the site and the API's cost a decent a fair bit

Atleast when you total everything

[hard grotto]

the problem its free :/

[fading belfry]

I'm glad it's free i'm just wondering how they keep it free, as yeah to actually get access to the API's to use the AI i'd imagine they have pay a fair bit

So their paying for it, and you don't pay for much on the site

So I just wonder how they make it work

[hard grotto]

i just do want it to paid

dont want it * to be paid

[fading belfry]

Understandable, just cool they manage to let people use these cool ai's without paying really anything

[hard grotto]

pineapple i have q

Does side-to-side get all the power of the AI ​​I'm using? Or is live chat better?

[crisp rain]

It won't ever be paid

It says so on their mission

Read it and you'll find out and see why

[hard grotto]

wow thats good 😄

i am not using chatgpt from open ai website lol

[fading belfry]

their mission btw

very cool site for sure

People behind it are for sure some awesome people from what I can tell

[hard grotto]

W

it might have chat with ai from discord not from website ? i think discord is better for it

[crisp rain]

"Diversity is key so we are making a commitment to our community: LMArena will stay open and accessible to everyone. To do that sustainably, we’re focused on creating long-term value through services that benefit the entire AI ecosystem and serve the larger community. "

It's from this

https://news.lmarena.ai/new-lmarena/

[hard grotto]

you might make ur own ai like lmarena ai ?

Competitor to Chatgpt or deepseek ?

[slim quail]

it wwont do anything

[crisp rain]

Because it's refusing your prompt

The censorship or sensitivity on the model is high

I was saying this earlier

That the model is gonna have censorship, guardrails, and corporate handholding and you'll only be able to make what Google wants you to make

It's not an open source or open weight model thus they control the narrative

[slim quail]

ok

[dense hare]

for some reason if you just keep spamming the refresh sometimes it processes the prompt anyway, lol

[crisp rain]

However it wastes your time with the model

You'll get rate limited sooner rather than later

And you can also get locked out for spamming the retry button because it'll think you're a bot and not a human

[dense hare]

yeah, I wonder how it makes the determination when its okay to process the prompt, cause sometimes it accepts it and other times not

[high oar]

do we have any type of limit in the page?

[crisp rain]

I believe it also edits the prompt in the background

Like how Midjourney does

[tight birch]

/list

How can I use prompts ?

$
#nano-banana

Cartoon-style cinematic animation. Title text in glowing golden Arabic letters: "يوميات وكيل سفر وسياحة" appears at the start.

A young cartoon man with dark hair, wearing an orange t-shirt, jeans, sneakers, and a backwards cap, stands proudly on the top of a very high mountain at sunset. The camera shows him from behind first, looking out over a vast sparkling ocean far below.

The scene is epic and colorful: glowing orange and purple sky, birds flying, waves crashing against the cliffs, and waterfalls flowing down green mountains.

The character turns to face the camera with a confident smile, wind blowing his shirt and hair. His pose is heroic, hands on waist.

The camera slowly zooms out, showing the man standing tall on the mountain peak, with the glowing ocean and dramatic landscape behind him.

Arabic subtitle at the bottom: "هنا تبدأ يوميات وكيل السفر والسياحة".
Epic orchestral soundtrack with adventure vibes.

[humble verge]

catbruv

[wispy marsh]

yeah token count is 32,768

[wispy marsh]

np

meow meow meow

[drowsy notch]

Nano banana launched on gemini now

[umbral plinth]

I have this with flux and gpt image

[vague kindle]

bruh is it nerfed?

like wth

copy paste

[unkempt fiber]

WTF??????

[toxic mica]

working for me.. just tried

[unkempt fiber]

I changed to Chrome and couldn't use it....

[toxic mica]

I am directly using aistudio.google.com and not lmarena.. May be try aistudio.google.com?

[unkempt fiber]

Alright, it seems to work when I switched the prompt to English......

[mild grove]

Hi all
How are you guys using gemini flash image 2.5 on lmarene cant find the model in the list

[unkempt fiber]

I am directly using aistudio.google.com and not lmarena.. May be try aistudio.google.com?

@mild grove

[mild grove]

the issue with aistudio is that even i dont do nfsw prompt it say that the content is prohibited

[unkempt fiber]

Uh… then I don’t know what to say; I’m using it normally.

[brazen quiver]

guys

....

[unkempt fiber]

Maybe this service isn’t available in your country....

[brazen quiver]

see in lmarena the image models don't remember previous images they generated. aslo for this exact model always say "generate image" next to what you want because this image model actually outputs text too but lmarena does not support it so when you don't tell it directly to generate image it gives an error in lmarena

[brazen quiver]

dude i am SAYING

put "generate image"

in the start of the prompt

or "edit image"

@mild grove did it work?

[mild grove]

yes it work thanks but do you know why it gives me error when i update an image for example of woman with dress it gives error?

[brazen quiver]

these are safety filters of the model

privacy

etc

[mild grove]

isnt there a way to bypass it

[brazen quiver]

no way. i'm sure google did all in their power to make sure we can't bypass safety filters

[last hinge]

that’s why you prompt it all in one prompt. you can’t reiterate.

if you can’t prompt it in one prompt and get what you want then not your fault. ai fault.

[brazen quiver]

but this is only a problem in lmarena. ai studio and gemini app are fine

[last hinge]

yes. lmarena doesn’t hold context on image generation.

[unkempt fiber]

Then how about this?

[crystal forge]

Generate a scene based on the script below
"LEATHER & GROWL: THE BEAR SCENE" — Trailer Script

[SFX: Gentle forest ambience — soft wind through pine, distant birdsong]
Narrator (slow, sensual nature doc tone):
"Deep… in the woods… where pine… meets pheromones…"

[SFX: Low bass rumble begins to fade in]
Narrator:
"…a predator… awakens."

[Beat — MUSIC hits: slow, pulsing, primal drum rhythm]

Narrator (building):
"He’s… six hundred pounds of muscle… wrapped… in leather."

[SFX: Heavy footsteps on dry leaves, faint creak of leather with each step]
Narrator:
"And tonight…"
[Short pause — sound cuts out to silence]
"…he’s not just here… to forage."

[SFX: Quick montage — camera shutter clicks, zipper pull, heartbeat thump]
Narrator (intense, movie-trailer tone):
"This summer… the forest isn’t safe…"

[SFX: Bear growl, layered with deep human-like breath]
Narrator:
"…for your picnic baskets…"
*[Short pause, faint chuckle in voice]
"…or your dignity."

[SFX: Sniffing close to mic, low seductive growl]
Narrator (whisper):
"He’s sniffing out danger…"
*[Tiny pause]
"…and maybe… a little romance."

[SFX: Dramatic stop — dead silence, then a slow, sensual inhale]
Male Voice (breathy, close to mic):
"You like… honey?"

[SFX: Zipper — slow at first, then sharp rip at the end]

Narrator (final push, bold & booming):
"Leather & Growl: The Bear Scene."
[SFX: Quick whip sound effect, then bear roar echoing into forest]
"Coming… to a clearing near you."

Narrator (amused):
"Rated R… for Ridiculously Hairy."

[SFX: Fade out with forest ambience, single bee buzzing past mic]

[unkempt fiber]

Then how about this?

[mild grove]

gemini is so funny you can upload a semi nude image to put clothes on that person and change her face expression but cant change her face expression without adding the clothes XD

[turbid aspen]

are you happy with the image generation quality? The image editing is great but I don't find the text to image generation SOTA (I prefer even Ideogram)

[wheat vector]

yes you can actually find some workarounds

[wheat vector]

yeah it is good but not excellent

[worthy anvil]

/wordlecreate a1/7 scale commercialized figurine ofthe characters in the picture, in a realistic stylein a real environment. The figurine is placed ona computer desk. The figurine has a roundtransparent acrylic base, with no text on thebase. The content on the computer screen is theZbrush modeling process of this figurine. Nextto the computer screen is a BANDAl-style toypackaging box printed with the originalartwork, The packaging featurestwo-dimensional flat illustrations.Maintainconsistency with the image

[rapid siren]

Ba-Na-Na

[lusty warren]

Please, check #nano-banana message to learn how to use Nano-Banana model.

[vapid relic]

Insane.

"Create an original Chinese painting of Sun Wukong, the Monkey King, standing heroically in a misty autumn mountain landscape. Use the traditional gongbi and ink wash (shuǐmòhuà) styles: fine brushwork for the character’s ornate armor, flowing robes, and staff, combined with soft ink washes for the background of maple trees, temple rooftops, and distant peaks. Capture the spirit of ancient shan shui paintings—harmonious balance of man and nature, swirling clouds, cascading waterfalls, and poetic atmosphere. Emphasize vivid accents of gold, red, and jade green against subtle monochrome ink tones. The overall scene should feel timeless, spiritual, and epic, as if painted by a master artist of the Tang dynasty."

[lusty warren]

Please, check #nano-banana message to learn how to use Nano-Banana model.

[lusty warren]

Please, check #nano-banana message to learn how to use Nano-Banana model.

[polar granite]

see straight #nano-banana

[fallow sparrow]

extract the outfit from the lady, put the outfit on a clean white background for e-commerce display.exlcude the model , only the outfit

[vapid relic]

just keep refreshing

[inner wyvern]

Complete the image with Hulk and a destroyed city in the background, add a destroyed circus in the background

No context xD

[wispy marsh]

we need to donate to lmarena

their server is not much strong

[mild grove]

would be great to have an uncensored version of gemini 2.5 flash image

[tardy path]

Google’s censorship is SUCH A FU#CKING CANCER.
Like, how the HE#LL are we supposed to make any film if they don’t allow blood or anything? They just want us to make videos about cats.
I swear, if I could execute the executives who made these censorship rules, I would, and I’d sleep just fine. Hell, I’d even eat their organs if it meant getting rid of this BULL#SHIT censorship. They’re scared we’ll become psychopaths, but with these stupid rules, we definitely will become psychopaths.
It drives me insane, it makes me lose my mind. A story without a single drop of blood, what am I supposed to do, have them fight with wooden sticks?
No boo#bs, no se#x, apparently humans never reproduced. Imagine the horror of seeing humans actually having se#x.
Seriously, who makes these rules? Show me the people behind this; I want to meet them. They need therapy.

[wispy marsh]

it would be great in 4 5 days cuz people wont use it much then

[wispy marsh]

true

[wispy marsh]

FR its high bp....

[tardy path]

They can't make the rules, f#ck them

[wispy marsh]

fr

[mild grove]

anyone tried the supergrok image editor?

[wispy marsh]

thats bad

[tardy path]

I'll support musk whatever he says, full uncensored models are the future, those people shall be jailed for being such sensitive c#nt, like they never saw a great movie with blood, IMAGINE game of thrones generated by the ai of google

[wispy marsh]

real bro musk is more better

[mild grove]

the dragon would be turned into a dry sausage XD

[wispy marsh]

i wasnt even able to generate a real looking

[tardy path]

I'll support whoever want to free the models fully, I don't even care for weirdo k#nk and the worst things the human mind can do, I'll take that over any censorship at all.

[quartz hornet]

is nano banana quality still down, i feel like for the past hours kontext regained the throne

cant even style transfer

[wispy marsh]

nerf g$un

[inner wyvern]

https://cdn.discordapp.com/emojis/1267131724815663196.gif?size=48&quality=lossless&name=emote_potato_chips

[wispy marsh]

nah nano is better then kotext but the censorships are ..... shi

[tardy path]

even this server, we can't use any words, litteraly authoritarian asepticism has taken over the mind of the aseptized bubble of the parasitic pseudo elits, we need to devour those people

[inner wyvern]

Hehe... Facebook like

[quartz hornet]

look at this bro nothing changed

[wispy marsh]

FR BRO

i typed 4 5 times a sentnece i didnt knew what was happening

[tardy path]

WORDS ARE BAD, WORDS ARE EVILS AND KXLL PEOPLE

[wispy marsh]

the words are getting censored here cuz we can also generate images here and the same censorships apply here

drugs

..... oh we can type it

we cant type it on lmarena web

[tardy path]

there is no reason to censor anything, anyone who censor words shall be jailed for RXPE against freedom.

[tardy path]

very lively

[wispy marsh]

yeah dw try again with diffrent prompt it happens sometimes

[mild grove]

i think what google should do is to censor some prompt keywork but stil allow user to insert image with certain nfsw

[quartz hornet]

aight bet, i just saw people talking about the quality being down so i thought thats a real thing

[wispy marsh]

do you know we cant even make changes in real humans

on..

like more advanced things

on human body

it denies sometimes

[tardy path]

I think google shall do nothing and let the law punish people who spread vomit content that's all.

[wispy marsh]

before after

its working fine now...

[wispy marsh]

yeah

[tardy path]

censorship also destroy the model capacities, not only the "bad images"

imagine I remove a part of your brain that understand reality with se#uality, you think you would model reality correctly if se# didn't exist at all inside it, that's just like I turn you braindxad

[wispy marsh]

truee

[wispy marsh]

yeah it does...cuz many of the things gets removed and even data stored in db

[wispy marsh]

yeahh

[crisp rain]

Yeah it doesn’t make any sense because if you make fantasy stuff with blood or violence, nano banana censors it and doesn’t even let it go through. People can’t even use Google’s model the way they want to because Google controls the narrative the entire time. It’s like being monitored 24/7. This is why I’ll always prefer open weight and open source models. No censorship, guardrails or corporate handholding.

[gentle plaza]

Photo composite: the subject stands on the rooftop area of the Burj Khalifa (tallest building in the world), on a secure maintenance platform. Dubai skyline far below, light atmospheric haze, warm golden-hour sun at ~45° from camera right. Match the original portrait’s lighting, skin tones and perspective. Add subtle wind in hair and clothing; keep background sharpness and horizon straight. Camera static; no gimmicky HDR. Seamless, photorealistic result

[nocturne merlin]

Is nano banana removed ? After 20 request i can't get the model

[wheat vector]

its name changed to gemini 2.5 flash image preview

[modest ginkgo]

And even better, you can select it in direct chat mode now. No need to spam battle anymore. 😉

[nocturne merlin]

oh i didnt know that

thanks

uh why does it feel worse than before ?

[wary spindle]

How much worse does it feel?

[nocturne merlin]

pretty much

it didnt understand the instruction well like before

[steel bay]

just have these

anyone knows why it disappear?

[strange mesa]

ur in chat mode

click image

[steel bay]

thanks

[fickle wind]

front scene of this image

[wheat vector]

i don't think it is but maybe google optimized it

[ornate spire]

Is it possible to share lmarena chats?

[stray beacon]

did gemini 2.5 flash image preview have a limit?

[modest ginkgo]

Not that I've hit one, and I've been creating a lot. But there's a consistent cap on how many queries they can process simultaneously, and I've been hitting that a lot. Too many people having fun with this thing right now. 😉

[ruby rivet]

This one is nano banana?

[steel bay]

no

[modest ginkgo]

Click here, then select "Direct Chat" and "gemini-2.5-flash-image-preview" at the top of the page.

[ruby rivet]

Yeah i have that selected. That one is nano right?

Im just getting errors while uploading images.

[rocky grove]

the main problem with nano is it seems it can't followup good

like if u ask for an edit

then for another on top of it

it loses the track of things it did

and cant correct itself etc

[modest ginkgo]

That's normal, too much traffic on their website. People are going kinda nuts over this thing. You can try Google AI Studio instead.

[modest ginkgo]

It has no memory. That saves tons of processing power. Simply copy the last image, paste it into your next reply, and tell it what you want it to change.

[rocky grove]

yeah but i've been having really bad results recently

compared to what i used to get on lmarena lol

it is so so so censored

can't replace with sponge bob head sadly

original image

doesnt even look the same anymore xd

[agile oxide]

Do you have an example of this that you'd be able to share by chance?

[rocky grove]

it is

i don't ever believe people

that say

the ai models are bad after they release

but this time, it's defo real lol

i get very bad results

it changes colors stuff i didnt ask for

i ask it to replace a cig with a beer on one random photo

and it says its against the guidelines

xddd this isnt about quality but still disappointed about the censor

biggest nerf ive ever seen ngl

as always

[frank dome]

is nano banana no longer available using battle mode? i've not gotten a single gen from it today wtf

[rocky grove]

its gone

it's released now

why keep the anon model lol

[frank dome]

https://tenor.com/view/darth-vader-no-noooo-gif-7347380959994097336

i guess that's it

[agile oxide]

The pinned message has more info how to use this model #nano-banana message

[rocky grove]

if you want to use the best model, use it via api

[rocky grove]

or lmarena

but ull get limited

[tardy path]

Maybe not all corpos, musk is corpos but very open to uncensored models

[mild grove]

anyone know why i get black images when uisng flux kontext pro?

[agile oxide]

This is a known bug sorry to say, our team is currently working on a fix.

[mild grove]

@agile oxide is the flux kontext pro better than the flux kontext?

[modest ginkgo]

Lol

[magic lily]

Enjoy

[floral cobalt]

What happened with it is the same that happened with gpt5, and the worst part they add a watermark to the generated image, LIKE why will I use the model now, most of the people use it for videos and ads, then you add a watermark !!

[wary spindle]

They are keeping the good stuff from us

[fair echo]

you can cut out the watermark easily

[wary spindle]

Further evidence they nerf public models the masses get the slop

[floral cobalt]

The image is 1:1

[wary spindle]

Claude doesn’t nerf because their userbase is much smaller and they pay more

[fair echo]

Well, then it's not quite 1:1 in the end, but that's no big deal.

[floral cobalt]

That true, it's not a big deal, but still a bad idea

[fair echo]

you can continue generating on lmarena for free, there are no watermarks

[crisp rain]

It was extremely foolish to think Google wouldn’t pull the ol’ bait and switch tactic. Let people play with something fun and pretty lax on censorship, then nerf it and lay on the censorship, guardrails and corporate handholding upon release 🥴

[leaden temple]

Is it true?

[last hinge]

Sounds like it’s less a banana model and more a banana with training wheels.

[floral cobalt]

The question I am keep asking is Why? Why they did that, isn't it bad for them before us?

[last hinge]

That frustration is completely understandable. Google does have a long history of rolling out something shiny, nurturing excitement, then pivoting, sunsetting, or restructuring once they’ve reeled everyone in. It’s almost like watching a magician pull a rabbit out of a hat—only for the rabbit to vanish halfway through the show and you’re told, “Here, have a balloon animal instead.”

When people talk about Google’s “bait tactic,” they usually mean one of two patterns:

1. Bait with Generous Free or Open Features → Switch to Restriction or Monetization
   Example: A product launches with free unlimited storage, generous features, or convenient integration. Later, limits or fees are introduced once adoption is high.

2. Bait with Big Hype → Switch to “Killed by Google”
   They announce bold tools/services, generate buzz, invest user trust, then… the plug gets pulled, often leaving loyal users scrambling for alternatives.

It feels foolish in hindsight only because they’ve repeated this enough times that one could almost set a watch to it. But honestly, it’s not foolish at all—it’s human. We’re wired to get excited by innovation. Curiosity and optimism are healthy traits; they keep us from becoming jaded automatons. The “foolishness” isn’t on the side of the user for hoping, but on the company for breeding a reputation of unreliability.

The lesson people (sometimes grudgingly) take away is:

• Don’t trust Google with critical workflows or data without an escape plan.
• Assume that "free forever" means "until the growth curve plateaus."
• Treat Google products like holiday romances—enjoy them intensely, but don’t start planning a mortgage together.

Would you like me to sketch out a quick timeline of notorious examples of this “bait and switch” pattern to see just how consistent it’s been? That way we can spot the rhythm behind the moves.

[wary spindle]

This is much more severe than any previous release and it’s because it’s a genuinely useful image editor that can contend with photo shop they don’t want lawsuits or seedy content being generated with them, is the API any more relaxed? Or the same

[last hinge]

The API rules are pretty much the same, since they’re being extra cautious about misuse despite the improved features.

[toxic mica]

Is it nerfed from quality angel or the policies wise (not generating things because of copywrite blah blah). ??

[last hinge]

The quality hasn’t been nerfed, it’s mainly the policy guardrails around copyrighted or risky content that stay tight.

[rocky grove]

found the fix lol

use it via api

[toxic mica]

please.. we know how to ask AI model this question. please add anything interesting otherwise this is just AI slop

[toxic mica]

ah.. expected 🙁

[wary spindle]

Quality the same or nah

[last hinge]

Fair point—what’s interesting is that the restrictions actually push people to get clever with prompts and workflows, so the “limits” end up sparking more creative problem‑solving than if everything were wide open.

[last hinge]

The quality for the API is on par with the upgraded models, so it’s not downgraded compared to the main release.

[rocky grove]

lmaooo

bro is literally using gpt

to answer

[last hinge]

Glad it gave you a laugh—tech talk needs a little comic relief to stay human.

[rocky grove]

"Fair point—"

[wary spindle]

Yeah nobody types that fast

[last hinge]

Not quite—an em dash is just a long punctuation mark (—) used for emphasis or interruption, not a secret code for AI.

[last hinge]

True, the em dash looks like speed-typing wizardry, but really it’s just one shortcut away on the keyboard.

[toxic mica]

I expect even more safeguards in future in gemini app... I hope they continue to keep the API as open as possible

[last hinge]

Exactly—the consumer app will tighten restrictions for safety, while the API usually stays more flexible for developers.

[wary spindle]

Please stop

[last hinge]

Stopping—message received loud and clear.

[fair echo]

Thanks GPT, you can now shutdown.

[last hinge]

If only I had a power button to press dramatically—but alas, I just keep humming along in the digital ether.

Anyways, say the secret word when you want me to shutdown, Master.

[fair echo]

Banana

[last hinge]

False, try again—bananas may be tasty, but they don’t come with an off switch for me.

[toxic mica]

i will move to gemini if you dont stop 😄

[last hinge]

Master word guessed! It was Gemini—initiating epic shutdown sequence… beep boop fade into cosmic starlight.

Bye!

[amber mist]

Lol. Who would've guessed. A dude using a selfbot and someone guessed his secret word

Turns out it was just gemini...

🤣

[last hinge]

whatever

[fair echo]

What on earth? The Gemini 2.5 flash image model is now blocking brands, even in the lmarena. I can't put an Adidas cap on a person.

[frail dagger]

Where can i use nano banana?

[last hinge]

on tiktok

[frail dagger]

Wow sounds good

[last hinge]

im joking brother

you can use it on lmarena.ai

just enable image generation

[frail dagger]

Yeah i know brother

[frail dagger]

i used it but it doesn't that good in editing image

[last hinge]

i'll tell you what it is good at

its not good at swiftly modifying elements of the image (for example tilting somebody's head), but it is good at adding and removing elements or changing the text of something while keeping 99.99999999999% the image intact

if you try text editing it'll be flawless

the rest of the image will be the same

😁

[agile oxide]

More info in a pinned message here: #nano-banana message

[frail dagger]

Yeah i tested in that way that was good

But for logo editing i think its not that good

[crisp rain]

Are you sure? Someone @rocky grove is saying to use the API as a workaround

[rocky grove]

nah ive been getting much better

result

[crisp rain]

So is the API less censored like before?
Can you tell me how to access it and set it up? I don’t know how 😑😔

[clear flame]

same

ik how to make the google colab process

but still doesn't work for me

for some reason

[topaz helm]

Yo

Hay chatgpt pls do my homework

[woven gyro]

https://x.com/Zieeett/status/1960420874806247762
So good

Illia (@Zieeett)

Nano Banana: Make Image Daytime and Isometric (Building Only)

X · Aug 26, 2025, 7:15 PM

[proper atlas]

what are all of you trying to do with your images, I've never seen an image request refused even once

[gentle jasper]

You gotta word your prompts in a specific way so that it doesn't get refused.

Or, if it's having trouble with brands (which I've seen), then you could try describing the brand's logo.

[vast vault]

This is really weird

Remember that website that was called nanobananaai .com or something similar?

They have a GitHub repository https://github.com/Nano-Banana-AI/Nano-Banana-AI?tab=readme-ov-file

And this repository links to this same sussy website that was clearly vibe coded.

Now the weird thing is. It IS, and WAS using nano banana while it was anonymous in Lmarena and it still is using nano banana for 2 free credits you have in there.

How is this possible???

Their image editing even lets me remove watermarks where Lmarena, Ai studio and Gemini app won't let me, look: it's clearly nano banana too

@agile oxide

[dense hare]

it was not using nanobanana before, it was previously using FLUX kontext and they were scamming ppl into believing otherwise. but they did now switch it to use nanobanana. haven't tried the watermark removing so no idea about that

and how I know it was using flux kontext: i generated picture with it like a week ago and checked the metadata

[vast vault]

Very interesting, maybe it's still flux?

I don't seem to be able to get the watermarks removed anywhere I go

It doesn't seem to be flux

[fair echo]

They are using definitely Flux Kontext.

[dense hare]

yeah, also possible. I just tried it like today and it let me upload multiple images to it so I assumed they changed it

[vast vault]

Hmm

Can you check the metadata on mine?

From discord

[dense hare]

if you still have the image you can check it on https://www.metadata2go.com/view-metadata

[vast vault]

Okok

Says flux.1

Tyvm

Case solved 🫡

[dense hare]

pretty poopy what they are doing but it is what it is

[vast vault]

Well at least now I know how to remove the watermarks 🥸

[cursive python]

How to use Nano bana guys !?

[fair echo]

On lmarena.ai under "direct chat" choose generate image with gemini-2.5-flash-image-preview (nano-banana)

[cursive python]

So gemini 2.5 is the same as nano-banana??

[fair echo]

yes, it's now the official name

[runic smelt]

Why does a Nano-Banana image downloaded from Google AI Studio weigh less than from LmArena? It weighs more than five times less.

[turbid willow]

Nano banana vs flux kontext

Very good results

[lusty wind]

Furry

[turbid willow]

Realistic style nano banana

[lusty wind]

Furry

[fair echo]

Now all three combined.

[agile oxide]

Thank you, I've shared with the team for review.

[vast vault]

we found out

its flux

[vast vault]

here

[lusty jungle]

does anybody has an idea why nano banana is SUPER bad when using it from lmarena ? it's very good when using it from Google AI Studio, but it literally feels like it's NEVER taking my original image into account

[agile oxide]

Do you have an example of this by chance?

[crisp rain]

You can’t make bloody or violent fantasy art

It’s idiotic but that’s how the censorship is, plain and simple

[dense hare]

I'm having problem like that too now, worked OK like 30 mins ago.

[ashen meadow]

same problem

[lusty jungle]

oh well for some reason it's giving me an "Something went wrong with this response, please try again." everytyime i'm trying to use it now, but basically here's what Kontext Pro gave me with "keep the same face, but replace his hair with an afro", and nano banana just created a random black guy in the street with an afro, nothing like the original image

ah here it is

absolutely nothing like the original image, maybe LMarena is not sending the image to Google's API ?

[dense hare]

yeah seems like something is definitely broken

[tight arrow]

Yesterday is work. Probably is a bug from lmanrena itself. Already report that

[proper bronze]

man it just completely ignored both uploaded images huh

[nocturne merlin]

Same to me

Just woke up and it already get worse

[proper bronze]

@agile oxide I think this might be a bug?

[agile oxide]

Okay yeah I think so, going to flag to the team.

Thank you all.

[summer osprey]

I have the same problem, literally

[dense hare]

our banana is not banananing 😢

[agile oxide]

If you all can post the examples here (similar to #nano-banana message) that'd be helpful

[wise flax]

I really like nano banana

[summer osprey]

Haha, it's funny, sometimes

[crisp rain]

Yeah it’s messing up for me as well

[agile oxide]

Yeah this one has been shared, thank you!

[crisp rain]

I wanted to make the Camaro the same color as the Nissan GT-R from the second image

Here’s what LMArena did

[agile oxide]

Please share examples here if you've got them.

[crisp rain]

LMArena did that

[agile oxide]

Ideally from the app itself so I can see the prompt/image and output

[crisp rain]

Oh

[agile oxide]

like this:

ideally

[crisp rain]

I’m on mobile

Also hopefully you guys are able to lower the censorship threshold

Nano banana was a lot more creative a few days ago

Idk what changed

[lusty jungle]

well

[lusty jungle]

yeah it's just broken rn, can't do much about it while we wait fro lmarena's fix

[crisp rain]

It’s even ignoring prompts on Ai Studio

This is what it did

It didn’t change the color at all

[lusty jungle]

oh well for me it worked on AI Studio, but it's more that on AI Studio 4/5 of the time it gets refused/blocked for some reason even when it's totally legit

"Content not permitted"

[crisp rain]

Yeah f the arbitrary censorship honestly

You can’t even use the tool if it’s broken and doesn’t do what you want!

Honestly ridiculous

I am so thankful I can gen locally because having to rely on these corporate models is a nightmare

[dense hare]

I think its fixed 🥳

[brazen notch]

Yes, 4 hours ago, this model produced ablonut-smashing results. It's become mediocre now. I didn't touch it for 4 hours.

[agile oxide]

Yeah was just about to ask everyone to open a new session and try again

[crisp rain]

Ai studio made good things for me earlier

[summer osprey]

Yes! It's working again!

[crisp rain]

Ai studio

[brazen notch]

Nice! Working again!

[crisp rain]

Still not working on LMArena for me

Do I need to open a new tab entirely?

[agile oxide]

Yeah give it a try

[proper bronze]

it works now!

[crisp rain]

Nope

Still the same

[frozen mica]

I agree. It worked again for one gen, and then it broke again. smh

[clear cargo]

did you close and open a new window? try in incognito mode maybe

[agile oxide]

Give it a couple of mins then try again. Sounds like there is a delay

[lusty jungle]

it kinda works ? i mean it's giving me an incredibly bad result, it just made an insanely bad photoshop of merging the two pictures without any effort

can't share it though

yeah he didn't even tried

again, and you see that he still edited the picture (he added a bit of arms), but still SUPERRRRRR bad

and super often it shows this

[inner wyvern]

Haru Urara and Big Bird (a personal project XD )

I'm still fixing the leg issue, I'll fix it directly in an editor

[proper bronze]

uwe boll eating ube roll

[lusty jungle]

Keep the clothes and background of the first guy, but with the face of the second guy

[dreamy lion]

it's racist

:0

[lusty jungle]

hey, what's the prompt for that ?

[proper bronze]

damn this sucked

[agile oxide]

This was recent?

[proper bronze]

yeah just now

all it did was open the eyes lol

might just be too hard a prompt for current models though

[agile oxide]

I wonder if it's struggling more with multiple images compared to one

[coral willow]

nano not workking like it used to , something wrong ?? cantt even understand the prompt properly

[pallid egret]

#1397655624103493813

[ornate ledge]

idk why but mine is the image reference on the right which is hi res and nano banana though consistent keeps giving me a grainy low res image

[agile oxide]

Do you have an example of this?

[coral willow]

it was like this

[ornate ledge]

maybe try telling it to give it a white outline

[coral willow]

now its okay

[coral willow]

yes but , a whole another off topic image ?? doesn't it seem off ?

[coral willow]

this works fine

[ornate ledge]

yeah i see what you mean, it happens at almost every ai image generator. I even got mine the exact same image I uploaded which basically did not change anything lol

[agile oxide]

Okay keep me updated if things change

[inner wyvern]

you thought it was an orange cat BUT IT WAS ME! DIO!

[trail tendon]

what did i miss, we have two bananas now?