general

[woven canopy]

Its everyone

[echo aurora]

The team is currently working on a fix.

It's just how our mod team is setup.

[celest orchid]

Guys please rate my game 🙏

https://019be0d2-a3d2-7aa4-990f-4866eb60907d.arena.site/

This for contest

[echo aurora]

We'll be circulating a voting form when the submissions are closed

[celest orchid]

I just want to get an opinion about the game

[celest orchid]

Are you tested game?

[fossil solstice]

thank you so much! :3

[echo aurora]

I haven't had a chance to play yet, but have been meaning to once I saw it!

Thank you for submitting!

[celest orchid]

Thank you! Hope you like it

[peak sapphire]

@echo aurora Nano banana 3 pro has not been working for 4 days in a row; a second after the start of generation, the error "Something went wrong with this response, please try again" appears.

[frigid tusk]

actually?

[echo aurora]

This is being looked into by our team.

[peak sapphire]

Okay

[molten parcel]

GGs gang

[hushed gyro]

Get an announcement out rn!!!!

@echo aurora

[elder solar]

why is it glm 4.7 so high on leaderboards?

i find it way dumber than any claude 4 model or 4.5 sonnet

[echo aurora]

[frigid tusk]

why video generation only work on battle mode tho?

[hushed gyro]

They will implement for SBS and direct chat later

[echo aurora]

Nope, today! Right now!

[hushed gyro]

And soon😍

[obsidian shell]

It has been weeks with video on arena, why announce it now?

[stray aspen]

that crazy

ive had video arena for so long

[frigid tusk]

oh alr

[echo aurora]

It was an experiment, it's fully rolled out to 100% now.

[elder solar]

we already know sora 2 is the current best video model tbh

[echo aurora]

Maybe, current plans are to keep it Battle.

[obsidian shell]

Nothing changed from our side...

[echo aurora]

Refresh the site.

[obsidian shell]

Its still video, still battle, nothing changed, I had it like that for weeks.

[elder solar]

however, only for tiktok resolution size

widescreen idk which model is the best

[gritty arrow]

is the limit of 3 seperate from the discord limit or are they combined?

[obsidian shell]

Seperate.

[gritty arrow]

so 8 total per 24 hours?

[stray aspen]

yes

[echo aurora]

This wasn't the case for everyone, but now is!

[loud verge]

Hello people

[echo aurora]

Hello

[loud verge]

https://tenor.com/view/paimon-waves-her-hand-paimon-genshin-impact-genshin-impact-gif-20084027

[marsh grove]

@echo aurora what's the daily image generation limit?

[echo aurora]

on web = 3 per 24/hr
on Discord = 5 per 24/hr

The help center article was just updated too: https://help.lmarena.ai/articles/1544829667-how-to-create-videos-with-video-arena

[marsh grove]

for images though?

[echo aurora]

Those are going to be different rate limits which we don't have publicly available. Would note overall use has a rate limit, and so do individual model use in things like Direct & Side by Side.

[magic ravine]

Why can't we use direct chat for video arena? 😭

[marsh grove]

why isn't it publicly available? just asking because generated ~5-10 images and it started giving me an error everytime which i'm assuming is a ratelimit

[echo aurora]

It's possible we expand this in the future. We were hearing a lot of feedback to move this to the site, and we wanted to get this out to everyone quickly.

[steady grove]

video arena

[echo aurora]

What link are you seeing that on?

[steady grove]

from annoucement https://lmarena.ai/?chat-modality=video

[magic ravine]

I see! So it's still a bit of a test for now. I was hoping for Direct Chat because it'd be nice to be able to choose which model to generate with rather than relying on chance

[echo aurora]

It's possible it's rate limit, but currently we're seeing a lot of errors with gemini-3-pro-image-preview models, I assume this is what you're referring to?

[marsh grove]

yes, i'm using the 2k resolution one

[echo aurora]

Yeah that makes sense! Direct and Side by Side could come in the future, but I can't say for sure.

[echo aurora]

Okay yeah I think is unrelated to rate limit, it's just an error issue on our end that's being looked into.

[little ginkgo]

.txt soon

[echo aurora]

[little ginkgo]

Anything but .txt file

[ember iris]

@echo aurora hi bro, any idea on fix on Google gemini pro i2i generation we faced issue.

[celest orchid]

nice

sadly no side by side or direct chat

[echo aurora]

Nope, nothing on the user's end that they can do to fix unfortunately. This one is on us to fix.

[frail thorn]

https://tenor.com/view/free-my-boy-hes-innocent-cat-in-jail-gato-atras-das-grades-grato-preso-gato-corrupto-gif-12704193048178390321

[shrewd citrus]

yay videos

[frail thorn]

https://tenor.com/view/cat-eating-fast-eating-eat-gif-25232258

[ember iris]

Noted, sometimes it's works for me as soon as I get the error,
I reload the page and then hit the regenerate 🔃 button in the model.

Worked only few times not regularly.

[frail thorn]

Gulp staff member 😱

https://tenor.com/view/noticing-gif-12761641764464540176

[half mist]

Please make it work with direct chat, and not as limited because I need to have a video generator without a strict rate limit

[vapid crag]

Is there any solution to the "Something went wrong with this response, please try again" problem? This issue keeps recurring in every prompt I send; I can't create or edit any images.

[tight agate]

Does lmarena needs to be logged in? And is there any cd

[half mist]

I honestly think this should be removed, then added back with it being less limited just because it sucks so bad, and is way too limited. Some of the video models aren’t even that expensive

[celest orchid]

If there are no limits, they will become poor.

[sage hinge]

@echo aurora Heyyy, congratulations on the release of the Video Arena on the web!!! I was really looking forward to it, so thank you so much!

[golden ocean]

Language Model Arena

[half mist]

A limit isn’t the issue, it’s that it’s too limited. For Sora 2 and Sora 2 Pro, that limit is reasonable, but other models don’t need that limited of a limit

[hushed gyro]

Why has these two NB models so unstable since introduction? Is it the provider's fault?

Tbh these are the only two models that can generate consistent images of what my prompt instructs, others are still stuck in 2024 vibes 😭

[celest orchid]

I think that API keys are not free, especially for videos

[hushed gyro]

Yo 2k has no errors!

I'll generate more

See what I mean?

Chatgpt image latest vs NB Pro 2K

[sage hinge]

oh brotha, are people already complaining? you get something for free and you're still complaining? why? make do with what you have, they can't just release everything with a high limit, it's too much

[echo aurora]

It's possible we add to Direct and Side by Side at some point. Can't say I'm familiar with planned changes for rate limits.

[echo aurora]

Thanks!! Glad to hear it!

[half mist]

It doesn’t even let you just use side by side or direct chat, and it’s way too limited at least models outside of Sora 2 and Sora 2 Pro.

[hushed gyro]

Nvm... Errors resurfaced

😭

[sage hinge]

well, pineapple just gave you the answer

[tight agate]

Hey @echo aurora can u tell ty

[echo aurora]

To use Video Arena you do need to be logged in, and the current rate limit is 3 generation requests in a 24/hr period.

[tight agate]

And what abt normal chat prompts for projects

[echo aurora]

Hahaha love it!!!

[echo aurora]

For other rate limits we don't currently have that information publicly available.

[echo aurora]

Going to turn into GIF and make pfp, hope you don't mind

[celest orchid]

Ohhh

it's great idea!!

[cedar pagoda]

Hey when is lmarena adding files support for other things. Then images?

Like txt or coding files to test models

Like for input not output

[celest orchid]

They're testing PDF now, and I hope they'll add TXT soon

[echo aurora]

We added a new category for Experiments in our Help Center, we're going to be adding new articles for things currently being experimented with. I'll be putting out a wider announcement when it's fully updated. https://help.lmarena.ai/collections/6672669378-experiments

[celest orchid]

War is over..

Why PDF first and not TXT?

[gritty arrow]

I was making a spy video in the video generator and accidently had something go against the ToS. Does that use up one of my generations?

[echo aurora]

That is going to count towards the rate limit.

[echo aurora]

Just what we decided to prioritize. We do want to expand these to more options, so more to come!

[celest orchid]

Here are 2 videos for an avatar with normal cropping

[echo aurora]

Hahah these are great. Not sure which one I prefer

[harsh wolf]

guyz, i have a question, is there a way to pay and bypass limits on lmarena please 😭

[celest orchid]

nope

[echo hawk]

https://tenor.com/view/yellow-emoji-no-no-emotiguy-no-no-no-gif-gif-9742000569423889376

[harsh wolf]

😭

[mortal coyote]

is there a specific reason why the Lmarena's GEMINI image generator got better understanding and quality then the original on Gemini webssite ???

[tawny kelp]

Random idea: Have an episode of Jeopardy with the same Watson that played over a decade ago versus two locally run LLMs running on consumer-grade hardware. If Watson wins, then do Watson, ChatGPT, and Claude. Just to see how much AI has advanced since then.

[echo aurora]

In February 2013, IBM announced that Watson's first commercial application would be for utilization management decisions in lung cancer treatment
TIL

[tawny kelp]

I think I heard about that.

[astral helm]

that's right, i also noticed that but its annoying lately, it has been bugging alot lately

[astral helm]

is there a reason why it's bugging alot lately

honestly, if there is a way to pay your way out of it i'd have done it, lmarena has the best nb pro understanding for prompts

[echo aurora]

I'm not aware of the specifics but the team is looking into this carefully.

[muted timber]

@echo aurora fix this bug

nvm

its working

[echo aurora]

Are others seeing this too? ^ cc @reef hollow

Can't say I've been able to repro.

[half mist]

@echo aurora I also requested to be in the Experiments a while back, but haven’t gotten a response, or confirmation

[echo aurora]

We don't grant access to specific experiments upon request.

[half mist]

I did fill out the form though

[echo aurora]

What form?

[echo aurora]

Btw @reef hollow did you view both videos?

System won't let you vote until both are viewed.

[hard quiver]

Is nanobana pro still down on lmarena?

[gleaming roost]

for some, yes

[woven canopy]

Is there kling ai on video generation?

[verbal ferry]

@echo aurora how is nano banana pro progressing

[viscid cloak]

It recovered for couple hours, but unfortunately the immediate error thing just comes back.

[verbal ferry]

crazy it's going on for too long

[fossil solstice]

i’m having the same issue

[hard quiver]

It looks like this is an issue on Google’s side, because in other Discord servers from different websites that provide the NanoBanana Pro API as a service, everyone is complaining about the same thing.

[echo aurora]

It's getting better overall, but still more progress to make.

[echo aurora]

For a better understanding of what models are available would recommend to check out the leaderboards - https://lmarena.ai/leaderboard/text-to-video

[fossil solstice]

is there an eta by chance?

[verbal ferry]

does anyone know platform for nswf image and video generation ?

[shrewd citrus]

grok

[tidal grove]

Why can't I submit videos to the video section?

[verbal ferry]

isn't there restriction always when i try it gives me Content Moderated. Try a different idea.

[echo aurora]

Sorry to say there is not

[echo aurora]

Hmm what do you mean?

[tidal grove]

Sorry, it's just me

I don't know where the video and PDF went.

[echo aurora]

Video Arena on the site is fully rolled out to 100%, so you should be seeing it.

PDF file upload is still an experiment - https://help.lmarena.ai/articles/8878654804-lmarena-experiments-pdf-upload

[fossil solstice]

maybe try clearing cookies on your browser?

[tidal grove]

@echo aurora Why did you only add the PDF and not the video?

[echo aurora]

What do you mean? Video Arena has been added

[celest orchid]

this impossible

to upload a video

[echo aurora]

Oh video file uploads

[celest orchid]

if u mean creating video, you need to choose battle mode

[burnt sinew]

@echo aurora are you Anh from the video?

[echo aurora]

I am not

[burnt sinew]

😢

[echo aurora]

He's a super nice dude, glad he's on the team

[paper kite]

How can I stop getting these annoying pop ups? I already purchased a yearly subscription for another tier

It pops up everytime I open higgsfield. It’s extremely irritating

[gritty arrow]

I dont know if Im the only one but im getting this bug that makes me have to reload the page every tome i want to generate something on the website

[fossil solstice]

me too

if i don’t refresh, nothing happens, the models just generates forever

[gritty arrow]

same, sometimes i have to refresh a bunch of times

[fossil solstice]

same omg.

😭

but no ur not alone twin

[gritty arrow]

Ive had to disconnect and reconnect for this image 3 times so far. but its not coming ;.;

[echo aurora]

Are you seeing this will all Arenas or just one in particular?

[gritty arrow]

Its happening with image and video, but those are the only ones ive used so far today

[echo aurora]

On average how long are you waiting? Is this in all modes (Battle/Direct/Side?)

[gritty arrow]

for the images, i was waiting about 10ish mins and for the video, thats been loading for an hour and a half.

[echo aurora]

And refresh always results in the generations being done?

[gritty arrow]

Sometimes, not always.
And definitely not for the video 🙁

[woven canopy]

Can I choose the model by myself?

[echo aurora]

No, it's Battle mode only.

[echo aurora]

Okay, will keep an eye out.

[shrewd citrus]

@echo aurora What’s your role in lmarena?

[echo aurora]

Community Manager

[shrewd citrus]

Is that managing discord? Or more?

[echo aurora]

It's more, support/marketing, that sort of stuff.

[rose linden]

Hello, has anyone else had this problem when coding HTML? I use Claude, and AI tells me that it is executing my requests correctly, but it is not reworking my file, so I see no change.

[shrewd citrus]

Nice, sorry if this offends you but do you also get paid to do it?

[fossil solstice]

direct chat for me

[echo aurora]

Lol no offense taken. Yeah, I'm an employee of LMArena.

[gritty arrow]

so ur job is just to chat with people and help out?
nice

[echo aurora]

Pretty much report on feedback, flag bugs, build help center, all that good stuff.

[novel quiver]

hi

[mortal coyote]

is anyone else facing this issue : when generating images , it automatically shifts to side- by - side mode , and then i have to make a new chat

[ashen mauve]

Yes it's the A/B testing that they have implemented, to a lot of people it's causing problems ranging from Minor to Serious

[wind phoenix]

Hello everyone, it's nice to meet you. 💖 I have a question regarding your website. Is there a daily limit on video generation? Or are you currently experiencing a technical issue or something similar?

[golden pulsar]

love the help i get from LMArena thanks!!

[echo aurora]

Glad to hear it!

[echo aurora]

Hello Yes

Video Arena is currently limited to 3 generations per 24-hour period. This limit may change over time.
More info can be found here: https://help.lmarena.ai/articles/1544829667-how-to-create-videos-with-video-arena

[wind phoenix]

But absolutely understandable.

[echo aurora]

Why 😭

[viral notch]

i love the video arena already lol

[sterile tartan]

Prompt?

[viral notch]

"coca cola rocket ship with soda as the exhaust"

[sterile tartan]

Thx that's very creative

[verbal nimbus]

Is that AI? 🤔

Can't tell anymore

[hollow flicker]

LOL

i kinda may or may not have forced gemini to code my discord bot in Ruby

because im to lazy to learn the language even tho its like the ezaist to learn

[distant idol]

@echo aurora

will u guys make this availabe again ????

[echo aurora]

We may, at the moment we are just planning on Battle for Video Arena.

[distant idol]

ud do a huge favor for us ngl, if yall did ts

[crude karma]

hey

[viral notch]

okay that is better. it has the music, the whole logo, and the emission trail. the exhaust looks like literal feces though lol. i wonder if you used a different prompt?

[distant idol]

its the same prompt dawg 🤙
lowk genius twin

but ts was sora 2 pro tho

[viral notch]

ah i see. the video i saved was from the regular sora 2. no wonder yours is better

i wish we could generate more videos. i would be churning out more meme worthy clips lol

[distant idol]

ngl the direct use of video arena would save us

the battle mode is random ud lose all ur tokens

[random wolf]

why I can't use flux pro guys? always error

[distant idol]

it works just fine

[random wolf]

it's saying "Something went wrong with this response, please try again." always

[distant idol]

it happens most of the time with T2I but not with I2I from what i saw

well , what browser do u use

[random wolf]

what's T2I a d I2I?

[distant idol]

text-to-img and img to img

[random wolf]

ok thanks

[robust sluice]

<@&1349916362595635286>

[quartz kettle]

Anyone looking for skilled full-stack dev?

<@&1372208524230397962>

[robust sluice]

so hardworking, scammers

[crystal agate]

Why does it keep saying failed generation vilolates policy where can i see the policies in place

[green yacht]

see ts everywhere in discord servers, these are not getting anyone 💀

[round oracle]

@echo aurora scam

[hollow imp]

@quick jackal

[left lodge]

@echo aurora
What does the glasses icon represent?

[left lodge]

Scam

Moderators are just online doing nothing rn. 💀

[hollow imp]

Image can be attached

Through + icon

[vivid coral]

holy crap the search arena is major updated, we are soooo back

[ashen mauve]

<@&1349916362595635286> Standard Scam and probably Spam in multiple Channels.

[ashen mauve]

you are welcome

yeah it is in every single channel this spam

[vivid coral]

they are probably asleep, I wish I was

[ashen mauve]

same here i literally have a security gig to do tomorrow from like 6 until 9PM

and i just cant sleep

i wish this was a joke but i actually have to work and i just cant freaking sleep

god help me

[fallen blade]

If I search what year is it in LMArena in Direct Chat, why does it show 2024 ?

[fallen laurel]

Hi everybody

Can someone help me figure out where to start the work?

@ashen mauve

[ashen mauve]

what do you mean start the work?

[fallen laurel]

There is some online work here.

I have a lot of AI-related work to do. Is there anything AI-related that is lacking here?

[ashen mauve]

what

[fallen laurel]

What related work is done here?

[ashen mauve]

I mean anything "work" related here is that LMArena is basically one massive testing platform for a lot of mainstream AI's, hence how the Leaderboards are Influinced by the actual Users.

As for Physical "Work" I've got nothing unless you want to go talk to one of the Moderator+ people, because the if you look at the Fourms, Mainly the Bugs there is a lot of Front and Back End Problems for the entire website.

otherwise i have no idea what work you are talking about ¯_(ツ)_/¯

[fallen laurel]

Thanks

[left lodge]

Do you even understand mt question?

[torpid pasture]

Nano banana crashing?

[astral helm]

ye its annoying

[torpid pasture]

Ikr

I want to swap faces

But in discord I can't do that

[astral helm]

how u do it tho

i never got good result

[torpid pasture]

I give a reference face image with a image i want the face i want to swap with

And i generate a prompt using chatgpt

I get good results

I want it for videos now

Any free ai tool available which could capture motion?

[astral helm]

really ? i get either different looking guy than the original image or i get unmatching lighting

[torpid pasture]

I could show you results

[astral helm]

that would be appreciated

[astral helm]

i'm a designer i don't know anthing about video models

[echo aurora]

That it has Vision capabilities

& that it's a cool model

Also thanks everyone for the spam/scam reports, sorry we were slow to react to that.

[astral helm]

bro please tell your team to do something about nb pro

[echo aurora]

We're very much aware of the problem

I wish I had an update to share, but it's still ongoing

[astral helm]

when is it expected to be fixed, it's been almost a week rn

[echo aurora]

I don't have an ETA sorry to say.

[astral helm]

not trying to hate, you guys are the best but i'm in rush for my work, and if there's a way to buy a priority gen i would do it

[astral helm]

also, freepik got a nice deal apparently to get their users to use nb pro unlimited, so i figured u can do too to lower the cost

[echo aurora]

I understand no worries

buy a priority gen i would do it
This isn't something we offer
freepik got a nice deal apparently to get their users to use nb pro unlimited
I'm not familiar with freepik

[astral helm]

ik i'm just letting u guys know, that might help u

[lime glacier]

Hello, why does this error appear when I try to create a video "Something went wrong. Please try again. ”

[willow mural]

nana banana is so unstable

[cloud oak]

i think lmarena site needs more image ai models that are as good as gemini 3 pro (as in..realistic looking humans) to keep up with the user demands cos honestly..over half of the models on the list are not that good.. 🥲

[willow mural]

unstability

[obtuse smelt]

help nano banana is unstable

[willow mural]

ik gng

[obtuse smelt]

hmm

Nano Banana Pro API Crash Review January 21, 2026: A Complete Analysis of the 4K Resolution Timeout Issue and 3 Emergency Solutions

so yeah nano banana is down

i hope gotta fixed soon

[robust sluice]

try it on direct sometimes it work but still high error rate, just like I got one complete result and other is error and lost all the limit

in battle I got only Flux (s), dont know why

[obtuse smelt]

wait is all models down ?

[robust sluice]

is it ?

[obtuse smelt]

hmm yeah

ah man is down

[robust sluice]

not all, my beloved Flux bros still works

[obtuse smelt]

i see

hmm i try other web

[left lodge]

Lol cool badge, then why is there another icon beside representing image upload ? Thats also means it can accept images hence vision?

[ashen mauve]

lovely my chat is bricked again its stuck in an infinity "This content violates our ToS"

[broken viper]

and my chat with coding a site is bricked too, claude is lying that he is editing the code and viewing it

[ashen mauve]

thats about par for the course here

[robust sluice]

so most models doesnt work?

[ashen mauve]

basically yes

for me they work up to a certain point were you are forced into the A/B testing, then it deteriates from there into a point were it randomly bricks itself totally

[obtuse smelt]

man all model not respond again

[ashen mauve]

again par for the course sadly

[tall tulip]

It's error again right?

Not my damn internet?

[naive phoenix]

how to fix it?

[obtuse smelt]

no internet are fine

but yeah model is error

[naive phoenix]

wha

[obtuse smelt]

waiting admin back online

[naive phoenix]

https://tenor.com/view/shiba-inu-gif-7715104432378153269

[obtuse smelt]

is bad today

[obtuse smelt]

yeah gemini is bad

i look in Nano Banana Pro API Crash Review January 21, 2026

is look unstable

[whole sundial]

don't worry, baidu just released ernie 5.0 which they say is better than nano banana pro in image gen (no it's not)

i honestly think all that baidu cares about is benchmark numbers, the model isn't that good in real life usage

[left lodge]

Lemme try sending this image using pdf to claude.

[hollow imp]

Yes

[left lodge]

Lmarena got pdf support experiment

[obtuse smelt]

well then
me:

https://tenor.com/view/touching-grass-touch-gif-21219969

[left lodge]

Now we can send images to claude through pdf :>

[rotund rivet]

there needs to be a way to regenerate the prompt

because my prompt has been generating for 2 days-

[obtuse smelt]

oh well

[whole sundial]

ernie 5 / nano banana pro / gpt image 1 mini, same prompt, sorted by accuracy (worst to best)
i honestly think baidu's insistence on having the generated prompt be in chinese is to blame, the album's name or artist does not appear anywhere in its prompt, it's honestly garbage
please don't believe benchmark numbers (other than lmarena/aa), especially for image/video gen
use the models and see if you like them, but honestly ernie isn't even worth trying
note: even though gpt image 1 mini is much lower and smaller in size than nano banana pro, it generates album covers much better than it, those must've appeared more in the training data for that model
also i still don't know what ernie was trying to do here, that image and the gpt image 1 mini image look nothing like each other, it doesn't even look like it came from the same prompt

[whole sundial]

they should've added a stop generation button a long time ago

[sour spear]

NBP vs gpt-image may be matter of taste, personally, I don't like the grainy aesthetics and slightly weird anatomy (three fingers for the guy on the left, the guy in the middle slightly squinting), while on the other hand, it looks more "realistic" than NBP for sure.
But I am pretty much shocked that in 2026, a new model like Ernie 5 still can't even spell a big letter title correctly, when NBP and gpt-image are alreay putting out small text without errors.

[left lodge]

Not again wth <@&1349916362595635286>

[naive phoenix]

where is admins😭😭😭

[stoic burrow]

hello

[ebon rover]

Wrong chat for scamming blud, try genshin chats

[left lodge]

Lmarena ia having a lot of experiments , they surely experimenting good.
Around 5+ , multiple can get together.

[rotund rivet]

#1463835779016032316

[whole sundial]

btw that was me trying to get the model to reproduce a real album cover, if you couldn't tell. my rankings were based solely on similarity to the original
but yeah, ernie 5 image gen is honestly just hype, people will say "china made a model that beats nbp", no it can't, probably a good competitor to flux schnell though! (that might be better though)

[wintry jackal]

The max lenght for Video Arena is 10 secs?

[rigid copper]

sup guys

<@&1349916362595635286>

[whole sundial]

i think they already got back access to their account though, look at their name

[rigid copper]

i wondering, why is this thing keep happening in this server

[whole sundial]

do you already have your account back?

[rigid copper]

i see

look at his bio

i see why, people who have their account hacked will send like this lmao

[ocean vortex]

Em... he probably clicked on a link or smth and enabled some webhook or self-host without actually getting the account directly hacked lol

which is why he can still use it

[coral birch]

dumb question, gpt-image-1.5 got added "high-fidelity" at the end of its name, are there any change?

[whole sundial]

my guess is they made a new account to replace the old one, so they changed the details

[rigid copper]

their bio are ridiculous

[whole sundial]

it's the high input fidelity setting from the official openai api

basically means higher input resolution

[coral birch]

I mean, are there any change compared to the previous one on lmarena without the "high-fidelity" or is it just a name change

[coral birch]

okeoke thank you

[ocean vortex]

yeah it's this

[whole sundial]

nothing else is different about the model itself, still the same resolution and there will no be difference in the outputs in text to image

[ocean vortex]

I don't see that option for gpt1.5 personally though

only 1.0

[whole sundial]

must've not made it to their playground yet?

[thin current]

But this isn’t LMarena, right?

[sterile tartan]

Website Name?

[fossil solstice]

are gemini models working for you guys?

[left lodge]

Send the prompt

[long zinc]

??

[swift aspen]

Many something went wrong failures 😑

Just a question. Are bugs taken seriously here, or are they ignored?It's simply impossible to rate the models this way.😑😟

And if my text doesn't make sense, it's Google Translate's fault. 😅

[long zinc]

Fr

[golden ocean]

https://tenor.com/view/cat-sunlight-follow-lie-groom-gif-5081640

[wicked sage]

hi guys

[stray aspen]

Welcome

Lmao

[left lodge]

Hm

[muted timber]

bro

i can t login my self

need help

[round island]

the audio quality for this video is so bad

[mystic raft]

Hey,
has anyone been experiencing weird issues with Nano Banana Pro (+ 2k) lately?

Over the past few hours, it's sorta been in and out. And ive been getting this message occassionally when I generate something

[west fossil]

same here

i cant continue making my video

it's annoying

[hushed gyro]

The errors ffs keep appearing

[obtuse smelt]

yeah i hope fix soon and back to normal again

[mystic raft]

Is Nano Banana down or something?

[bright shard]

The Nano Banana Pro model has many fails, but if you keep trying a few times it eventually produces the image.

[rigid copper]

that error was already a while like couple days, all i say is about 50%-80% chances that the generation will fail.

[neat lagoon]

Yes. I have the same

Its down

[rigid copper]

but at some point it will generate the image successfully.

[crystal agate]

Why do my videos keep saying failed to generate after creating videos

[neat lagoon]

Well, it doesn't happen to me. I try and try, but nothing.

[rigid copper]

i think high demand use cause some model to fail.

[merry wedge]

THIS IS GREAT

[rigid copper]

#january-3rd-contest message @gleaming roost did you delete your LMArena account or what?

that's no way your LMArena index.html got deleted unexpectedly.

[mystic raft]

Nano Banana might be totally down, almost everywhere-ish

I tried replicate.com and OpenArt, still sorta acting up sometimes

[little ginkgo]

Oh no

[hard quiver]

The mods on the Higgsfield Discord must be praying that Google fixes the Nano Banana Pro problem as soon as possible, because in that Discord everyone wants to burn the company down, since Nano Banana Pro doesn’t work there either 🤣

[little ginkgo]

Gpt better

Pls bring Claude back up

Pls bring Claude back up

I need i

[balmy mist]

its back

[little ginkgo]

Its not working for me

I am stuck with inf captcha

[echo aurora]

General reminder for what to do if you come across a Something went wrong error message. This article has some steps that are worth trying along with a way to report further information that'll help our team get a better idea of what is going wrong - https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

Also would like to note that gemini-3-pro-image-preview is still seeing high error rates. If you're having problems with this model I wouldn't recommend reporting further information via the help center linked above. This is already on our radar and actively being worked on.

[olive depot]

Hey, guys! Maybe someone knows how to fix it, i got infinite "Generating..." and can't stop or something else already 3 hours

[little ginkgo]

Ye happens

Csnt be fixed

Delete chat and make new

Rip

[little ginkgo]

Pls fix

[olive depot]

Oh, there is already a very long history of dialogue, it is a pity to lose it, of course

[little ginkgo]

Ye ☹️

Well u can like copy last message

That ai sent

[neat lagoon]

Yes

[long zinc]

Sad

[normal abyss]

sometimes if you wait like hours it will error out and you can refresh, however its not guarentied

[echo aurora]

If you haven't tried this already I'd recommend clearing cache + hard refresh the site. I've seen cases where this helps.

[olive depot]

It's sad that there is no stop request button.

[somber belfry]

hi

[echo aurora]

This is on the team's to-do. It is soemthing we'd like to build.

[hushed gyro]

2k is gone

[neat lagoon]

Is there

[tall tulip]

Yo guys, banana 2k is gone 🙁

Any chance the model will back?

[astral helm]

tf why did u remove the nb 2k guys

[lunar glade]

I hate to say but I honestly think this site may not survive 2026

[astral helm]

i mean it's hard to give free value in 2026

with this automation apps and bots they milk it

[echo aurora]

We've temporarily removed these models until we're able to get the error rate lower. It's not a good user experience to have these errors. cc @hushed gyro

[hushed gyro]

Cuz like remember the 2k was added for the first time? I was like you already have 3.0 image, why a version in 2k? I cannot see the difference

[astral helm]

tf, for better quality duh

[viscid cloak]

srsly? much better quality

[gleaming roost]

[echo aurora]

Because of the errors? I'm going to very much disagree with this, but would like to understand better why you think this.

[hushed gyro]

Hmm maybe Ill seriously test it when it is brought back 🤭

[astral helm]

ye u'll be amazed, also if you u have bunch of faces in ur image, it gets it 2x better than 1K

[hushed gyro]

Hmm maybe it's my prompt wording

Definitely will test it

Also thank LMArena for offering the world's first image models for free on this platform

🫡

[astral helm]

@echo aurorayo bro, just curious do have something added to nb pro that makes it understand prompts better ?

[tall tulip]

Dang,hope u guys can fix it
Appreciate u guys work tho

[viscid cloak]

English plz

[tidal grove]

@echo aurora Why can't I send a message? Grok and opus?

[topaz bay]

What is Baidu smoking

Also, who ever is in charge of theese visuals need to get fired https://x.com/Baidu_Inc/status/2014252300018254054?s=20

Baidu Inc. (@Baidu_Inc)

ERNIE 5.0 is officially live!

As a native omni-modal large model, it is built on end-to-end architecture to enable unified multimodal understanding and generation.

With a 2.4T-parameter MoE architecture and under 3% active parameters per inference, ERNIE 5.0 balances strong

X · Jan 22, 2026, 8:22 AM

[echo aurora]

I do not, but it's worth experimenting with different prompts and seeing what results you like best.

[echo aurora]

Can you elaborate a bit more? Are you getting an error message or just unable to send a prompt at all?

[tidal grove]

No, I just have a question why Gemini is allowed and grok and claude It's not allowed. Why??

[random tendon]

Bangladesh First🇧🇩🔥

[tidal grove]

@echo aurora How do you like the idea of sending a video to ai?

[echo aurora]

why Gemini is allowed and grok and claude It's not allowed
What do you mean is/isn't allowed?

[tidal grove]

I mean that on lmarena You just can't send them messages. Grok Claude opus Is there a reason for this?

[echo aurora]

Can you send me a screenshot of the problem?

[vernal saddle]

@echo aurora Why nano banana 2k was removed from LMArena?

[topaz lichen]

no nano banana 2k, but at least gemini 3 works now

[echo aurora]

We're seeing high error rates with it, and decided to remove it until we can lower this much more.

[echo aurora]

If you type into the chat bot & try to send a prompt, what happens?

[robust sluice]

just find 2k works well after long long time, then it got removed

[echo aurora]

I don't anticipate it being removed for long, this is temporary.

[green yacht]

pineapple, did lmarena update its website security, because I have a website that scraps many model anaylsis websites like Lmarena, LLM-stats, AA, Yupp, etc, and then blends the benchmarks + scores into one score, but recently Lmarena stopped working when fetching the data

[hollow imp]

Hello

[worldly lark]

Hello, I'm having problems with video generation, it's showing an error on the website whenever I'm trying to Generate

[hollow imp]

Can you please ask the team to add pdf support or search to Claude opus 4.5 thinking 32k?

It's 12:42 am and I'm going to sleep and @echo aurora I'm going to wake up with a lot of hopes

Please don't let me down 🙏🙏🙏

[sonic basalt]

Maybe someone nows how use it free and unlimited? VScode or some API keys ???

Cloude Opus 4.5

[echo aurora]

Can't make any promises. It's not an upoad PDF issue, it's a not havign upload file issue. The team is aware of this problem and it was bumped recently.

[echo aurora]

Can you give this article a try: https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

[echo aurora]

We do updates pretty frequently. Can't say I'm familiar with the security side of things.

[hollow imp]

What do you mean upload pdf issue

[worldly lark]

Thankss, first time trying it worked, then afterwards it just gave errors even trying new chat , I'll wait and see if it works again

[hollow imp]

The support is given to the normal opus 4.5 model and not thinking one

That is also the case for the search model

Opus 4.5 without thinking > pdf support without web search, web search support
Opus 4.5 thinking 32k > no web search, no pdf support, me struggling 😭

[sonic basalt]

Maybe possible to find public keys for cloude opus 4.5 ???

[echo aurora]

Meaning this model doesn't have upload in general enabled. Meaning images + PDF upload isn't yet going to work. It's not specific to just PDF.

[echo aurora]

How many total generation requests have you made in the last 24/hrs?

[worldly lark]

Around 5 for video

[echo aurora]

Okay yeah that's going to be rate limit

[thorny cove]

aw dang no direct chat on video 😔

[worldly lark]

Ohh, that makes sense thank youu

[thorny cove]

what if we had direct chat in uhh

videos but

the rate limit was like insane

like 1 video per week

[livid pine]

Guys, I’ve been reading and explain it to me properly — is Nano Banana 2K going to be down until they manage to fix it?

[echo aurora]

It's currently removed while we try to lower the error rate.

[livid pine]

Ooh, awesome, thanks!!

[magic ravine]

What happened to Nano Banana Pro 2K?

[magic ravine]

Is it permanent?

[echo aurora]

No

We're trying to lower the error rate before adding back.

[magic ravine]

https://tenor.com/view/kermitthefrog-dance-dancing-goingwild-crazy-gif-5023883256888097815

[floral jewel]

is the error rate specific to Lmarena ? I feel like there is the same problem on multiple platform

[gleaming roost]

Anyone want to duel with me at chess in the game I created?

[alpine osprey]

are yall having captcha problems

[hushed gyro]

A blank slate of nothing???

[woven canopy]

Damn? Yall removed nano banana pro 2k?

[woven canopy]

Oh okkk

[woven canopy]

Yep

[alpine osprey]

is this the new "something went wrong " thing

[gritty arrow]

its not new, but I feel like its been happening a lot recently

and my video still hasn't generated after 20 mins ;.;
edit: actually, 37 mins

[alpine osprey]

LOLL

[gritty arrow]

is there any way to see how many video gens i have left?

[echo aurora]

I've heard other platforms are having troubles from other users, but I couldn't say for sure.

[echo aurora]

If you refresh the site, does it appear?

[echo aurora]

There is not. But it's going to be 3 generation requests per 24/hr.

[gritty arrow]

refresh discord?

[echo aurora]

Oh sorry thought you meant the site

[echo aurora]

Can you link me the generation that's stuck? I'm not seeing it when searching through your generation requests.

[gritty arrow]

@echo aurora Here it is
[019be756-7cfc-7bd3-9551-03c73a766319](#video-arena-2 message)

[next thistle]

I need Frame1 to frame2 option. When it will work on Arena?

[dim ivy]

What are those models on yupp.ai Ik it's not lmarena, but I can't ask on their server and I don't find any info about these models.

[wicked wren]

;

[heavy smelt]

Something is wrong with the LTX 2 19B model on the video arena, you give it a prompt and there's a decent chance the video output looks completely unrelated to the prompt, like COMPLETELY unrelated, what's going on here?

[meager sun]

Why is there so much text

[hollow mist]

Please tell me where he disappeared nano banana 2k?

[analog steeple]

ayo? where's nano-banana-pro-2k? basic pro feels like common 2.5 nano-banana

[analog steeple]

for some reason I have this only on linux system, on windows works seamlessly

[glass zealot]

they probably had too much costs with that one

[naive igloo]

Hello

[hollow mist]

Don’t you know what will happen to her? Maybe some news? Will it still work?

[hollow ivy]

what worked for me:
logout of your LMarena account and log back in
that almost always solved it for me

[olive drift]

I tried, it doesn't work for me :/

[echo aurora]

I thought you were having problems with Video Arena via the Discord bot? This is a link to web.

[echo aurora]

We've temporarily removed the model from the drop down as it was experiencing high error rates.

[echo aurora]

Wait really? This works? I don't think I've heard that before

[undone geyser]

hey guys, new guy here...question...is it me or from lmarena.ai when i click to generate image, gemini-3-pro-image-preview-2k (nano-banana-pro) is missing? where is it? it is only emini-3-pro-image-preview(nano-banana-pro) without the 2k

[glass zealot]

broo, the model was completly fine, bring it back …

[undone geyser]

When will you bring it back? In few days or?

[echo aurora]

I don't have an ETA, but have bumped the team if there is a status I can share.

[echo aurora]

The error rate was pretty high, wasn't a good user experience.

[undone saffron]

100% correct

[chrome roost]

hello can i ask 2questions why is it limited to only 3 videos per day ? and why dont you make it in direct chat ?

[hollow ivy]

i use Brave browser, and LMarena i use with a google account
this method is my favourite way to unstuck threads which are generating forever

(on Linux Mint 21.3 Cinnamon)

(though i only tested it with battle-mode threads, yet)

but i only log out after having waited for at least a minute or so, to ensure, that it has finished generating

(it is called "sign out")

and don't press F5 (to update), else that might threw a monkey-wrench into the trick

[hollow dome]

Who is looking for a skilled AI engineer?

[obtuse smelt]

oh this my birthday guys!

[rare swallow]

I wish they added file uploading to coding🙏🌹

[obtuse smelt]

yeah back to normal again

[viral notch]

happy bday! how old?

[obtuse smelt]

32 haha

[lofty holly]

We are looking for specialists and professionals in the field of AI and machine learning engineers and developers, game developers and programming(backend,frontend,fullstack) developers, as well as QA testers, game testers and information security specialists.
A startup in the field of AI and game development needs a team of 25-30 people. A salary of $25,000 per month. If you are interested in this project and are ready for long-term cooperation with us, then you can contact us on our Discord, where you will find out all the details about the project.

[solemn nacelle]

@echo aurora why we guyz have 2 different gpt image models (1 being 1.5 and 1 being gtp image lastest) ?

[foggy root]

Will there eventually be a lmarena iOS app?

[jade cloak]

Hi,can someone tell me why does this usually occur?

[fallen blade]

If someone knows this pls reply. Thanks ☺️

[celest star]

hello, thanks to invite me

[keen beacon]

so how do you create an actual good prompt

[torpid pasture]

Give basic requirements to chat gpt it will create the best prompt but you must add a reference image to support your prompt

[keen beacon]

more for like websites (not in html, css, or js)

[torpid pasture]

Yea yea this is for that only

[sick sand]

[high owl]

hu

hi

~~

i m russian and know english

[left lodge]

Have you told the team that the error rate was majorly caused due to fixed requests allowed per min?
And that it is low compared to the users using the platform. Thats whats causing majority of the error for popular models

I think it would help the team too if you guys just hire me lol

[obtuse smelt]

oh yes where is gemini 3 pro 2k ?

[left lodge]

Its has been taken down by the team.

The cause is a lot of "something went wrong" error

[grave plaza]

the model was trained there maybe

let me ask my model 😅

[left lodge]

Becase it doesnt have realtime data. It relies on older data it was trained on that can only be of the past.

Use search mode for realtime information

[viral notch]

lmarena video battle is so amazing man, i swear lol

[obtuse smelt]

lol

sell bullets

[robust sluice]

its funny how 2k works fine for 10-20 min before the team remove it

Im about "wow they fixed it" then it gone haha

but its not forever they said, it'll back soon

[rigid copper]

that's why i only see gemini-3-pro-image- preview in lmarena image and not 2k one

btw hi guys

typo issue lol sry

[long zinc]

well i can't see either of them

[rigid copper]

they removed it due to too many reported errors

[long zinc]

dang

[high owl]

chat gpt 🙂

https://tenor.com/view/chatgpt-te-amo-coração-coração-abrindo-ia-gif-1702084030504635687

[little spoke]

hi

[cloud oak]

-# i used to be able to make so many images on the site like a month or 2 ago..now i can't even gen anything with gpt 1.5 and gemini 3 pro.. 🥺
there are 10k more members in this server since i joined..can the ai models handle all the user traffic?.. 🥺

[long zinc]

what does auto modality do?

[fair depot]

Hello

[hollow imp]

Then pls atleast add search support

[rigid copper]

#general message

some people find auto modality disappointing due to it will automatically route you to the different model

[left lodge]

That is Smart Router , different experiment

[left lodge]

It automatically selects the modality like text, image, code.

[ionic torrent]

Wait, so now I can no longer access Nano Banana 2k clarity model?

inside of lmarena

[left lodge]

#ai-creations is for showing ai creations

[merry wedge]

it is not generated image

[left lodge]

Yeah, simple one is available

[ionic torrent]

just to save money?

[merry wedge]

oh i understand now

my bad

[left lodge]

No, the cause was that a a lot of something went wrong error was occurring

[ionic torrent]

Oh yeah, but even normal model is causing that error

Like I literally had that error just a min ago

[left lodge]

Yeah popular models have that

Try when a new min starts

[scenic hawk]

Why is claude-opus-4-5-20251101 listed twice?

[left lodge]

Only a limited generation request are allowed in a single min

[left lodge]

Cause of pdf experiment

[ionic torrent]

That was first image after a week, that I'm generating

[left lodge]

See #1463601694796419203

[left lodge]

That doesnt matter. ??

[ionic torrent]

Oh okay I get it, my bad

[left lodge]

Well you can switch to any model you want.

This is smart router

[sullen bolt]

Something went wrong while generating response. Please try again -> what's the purpose & meaning of this Error message in LMArena?

[left lodge]

And this is auto modality

[left lodge]

Well it means something went wrong ¯_(ツ)_/¯

[sullen bolt]

that I know very well, but why is this coming, that's why I meant

[left lodge]

And suggests to try again

[left lodge]

Many reasons theres not a single cause

[sullen bolt]

but if I am creating a new thread for other info, that's working fine, but in same thread it's not

[loud verge]

What does it do?

[left lodge]

Try clearing cache and site data and a hard refresh, a new tab then visit lmarena.ai then that chat.
Clearing cache and site data will cause chat loss and log out. Chats only saved when logged in
And people are facing login issues so try on your own risk.

[left lodge]

@loud verge

[sullen bolt]

OK, but what can be the issue, like due to a LIMIT or what

[left lodge]

Theres many including rate limit, some random issue, total api rate limit, etc

[loud verge]

Ah....

But would it work without smart router?

[left lodge]

They are completely different and independent

[loud verge]

Like does it automatically switch to nanobanana pro if I ask it to generate image in Gemini pro?

[left lodge]

You can only use auto modality on battle mode and smart router on direct mode

[sullen bolt]

Understood!... but this really confuses the person what's the actual cause of issue - if there's a better error message is given, then it will be really helpful to understand

[loud verge]

Clear now.

[peak badger]

They still have nano 2k blabla off;

[left lodge]

I dont have control on that , i myself have said that to them a lot time ao but they will do whenever they thinks. They are doing a lot other things than that

5-6 experiments are currently active on the website

Testing different features

[sullen bolt]

I do agree you don't have control!.. I am just sharing my feedback as an Developer which will really useful not just for me, but for many users

[sullen bolt]

oh, that's a lot

[left lodge]

Yeah

[left lodge]

See #1417174113092374689

[late briar]

guy what happen?

[golden ocean]

true

creeper games

erm

literally

[fast falcon]

who knows what will happen to 2K and whether it will be returned?

[magic ravine]

Pineapple said it's only temporary for now until they can solve the 'failed to generate' error.

[fast falcon]

@magic ravine Thank you so much for your reply. 🤝

[late briar]

ok can say one thing?

[robust sluice]

is Gemini 3 think too long today guys ?

[rigid copper]

oooh, i see

[wraith sedge]

is claude removed?

[obtuse smelt]

oh yeah is 2k removed

but gemini 3 pro is fast, but 2k is removed

[fickle pulsar]

Please do something about the captcha it's literally annoying i click on the correct pictures but it say it's wrong like wtf it's acting like I'm blind. Replace it with something that actually works that doesn't result in people screaming at the captcha.

[obtuse smelt]

is fine now

[ornate mesa]

Snowbunny?

[wind ember]

whats that

[ocean vortex]

It's OpenAI playground, but it's using their official API and exposes mostly the same parameters (settings/options) LMArena have access to. Who in turn tend to name the model differently if they want to test some additional option and then hardcode that in API request.

[proper summit]

hello

[wind ember]

this one?

https://x.com/chetaslua/status/2014509263130173564

Chetaslua (@chetaslua)

🚨 Gemini - Snowbunny

Snowbunny mogs last best model of google and still i am not happy as they will never serve these models due to lack of compute

X · Jan 23, 2026, 1:23 AM

whats this guy yapping about

[ocean vortex]

Technically say a model "gpt-5-high" doesn't exist either, it's just gpt5 with a reasoning parameter set to high, model identifier is still "gpt-5". This is really more of the same except applied to other things as well

[robust sluice]

thats good

but still I got only 2 Flux in battle, didnt see other models

[devout spear]

Guys I need your help. I made this to be an open hub for inferencing where you can use AI inference for free and you can donate your compute to offer your self hosted inferencing to the hub for users to use. I’d really appreciate if you can use it and test it. You can also contribute and expose your local inference to others. https://github.com/Jiss-AI/jiss

[obtuse smelt]

but 2k is removed

[robust sluice]

yes cuz of high error rate

[alpine osprey]

they need to rework on that

[magic ravine]

Still sad about that. 2K was the best. The 1K one is so bad 😔

[cloud oak]

-# i keep getting wan and reve and qwen which i don't want for my image gens.. 😭

[obtuse smelt]

is fine now bro

[magic ravine]

Not it's not

[cloud oak]

-# maybe i should just go back to grok.. 😔

[robust sluice]

look at mine

and more 3 pics

this is what I got all day

[obtuse smelt]

hmm you wanna 2k get back in lm arena ?

[robust sluice]

thats why I use this profile pic and its not just today

[cloud oak]

-# yeah..like bad rating them does nothing in battle mode and the same models pops up again immediately.. 🤦

[obtuse smelt]

oh flux is not good

[robust sluice]

I admit some of their result is good but rarely see it

just dont like how I got only this model

[devout spear]

Guys can you try it?

[wide mauve]

will video generator be in the website at sometime?

[magic ravine]

https://tenor.com/view/yeah-john-wick-keanu-reeves-john-wick-chapter4-yes-gif-26625904

[surreal creek]

are you voting both are bad on all of them?? 😭

[robust sluice]

for Flux yes 🙂‍↕️

for me Im fine with any other models but except this one

and it always show up on 2 slot on battle

[stark linden]

hello

[lofty frigate]

Im going crazy without 2k😭

[long zinc]

💀

[balmy mist]

Neo

[inner thicket]

hello everyone

[stray aspen]

whats snowbunny

[wicked wren]

Hi, in the video generation section, later on will it be possible to choose the model to generate the videos?

[analog steeple]

same question

I want to compare veo with kling

[cloud oak]

-# has anyone used grok's site before?.. 🙊

[tight meadow]

Me

[stray aspen]

yes

it sucks

[tight meadow]

TRUE

[zealous sparrow]

google already nerfed this

btw

and snowbunny is also probably not coming out due to well lack of compute

[gritty arrow]

It links to discord.
Just tried it out...
Maybe theres a a problem?

[vivid badge]

Hey I am new here

[gritty arrow]

Video arena 2

[boreal saddle]

Ffs

Nobody is bothering to fix the captcha at this point.

Even Cloudflare is better than f*cking recaptcha.

[woven canopy]

When is nano banana pro 2k coming back?

[quartz light]

gemini flash

gpt 5.2

wtf is goin on with these 2?

[wide mauve]

why do i get so many error messages with nano banana? its not like that with the other models

[rigid copper]

sup guys

[shy scroll]

hi

[rigid copper]

you guys can try my image editor that is made from LMArena code arena using puter.js, posted in #ai-creations

[woven canopy]

Ye I basically can't do shi

[rigid copper]

because, lmao, nano banana pro keep throwing an error

[echo aurora]

You're right, sorry about that. I believe that generation failed, I think it's this - #video-arena-2 message

[echo aurora]

will it be possible to choose the model to generate the videos?
It's possible, but can't say we have these plans at the moment. cc @analog steeple

[echo aurora]

It's TBD

[woven canopy]

Wdym?

[echo aurora]

It's still being looked into. I don't have an update yet

[woven canopy]

Oh ok thanks

[shy scroll]

pineapple are you the creator of LMArena?

[echo aurora]

Nope, I was apart of the team pretty early on relatively, but not I'm not the creator

[shy scroll]

then who is the creator

[echo aurora]

Currently is! https://lmarena.ai/blog/video-arena/

[wide mauve]

Bravo👏

[echo aurora]

The founders are Anastasios Angelopoulos, Wei-Lin Chiang, and Ion Stoica

[shy scroll]

ohh ok

[wide mauve]

amazing, created by the new web video feature with the veo 3.1 model

[rigid copper]

tung tung tung sahur lmao (sry for saying that lol)

[wide mauve]

will there be added more features for the image generator like ratio aspect, and masking options for the editing?

[rigid copper]

you can just directly ask what aspect ratio you wanted in LMArena image arena chat

[wide mauve]

doesnt work very well in some of the times

[rigid copper]

i guess it work best when you use ratio 16:9, 9:16, 1:1, 2:3, 3:2.

[shrewd citrus]

it should be available for the side by side chat

so that we have to make a choice on which one is better

so it benefits us since we can choose our preferred model and it benefits lm arena since they still get feedback

[echo aurora]

Yeah this makes sense, it's something we're still considering.

[rugged abyss]

Knew Gemini had better results

[viscid timber]

why are most of the video arena channels filled with these specific languages which use a different alphabet

[echo sinew]

Well, we're a global community. People can submit prompts in their native language while abiding by the server's rules. Chatting will be in English only, though.

[peak badger]

The nano 2k thing it's still not available...

Shame on you

Peasants

[echo aurora]

We're still looking into it.

[glossy umbra]

❓

Lmao

@echo aurora you see what this guy is saying?

🤣🤣

[echo aurora]

Hecka rude 💢

[thorny schooner]

Is something wrong with the website since this non-proceed message generator warning keeps popping up for different messages / chat when it was just working fun yesterday usually this problem only appear for a single chat and could work later on so I don't know why it's happening across several chats now

[leaden laurel]

someone fact check this

[analog steeple]

and last question: will video arena be available without logging in soon? cuz there are still people who has troubles singing in, and if using discord to compare you need more than 1 people to vote (which doesn't happen) to reveal the models

[leaden laurel]

fact check status false

[burnt sinew]

guys if it doesnt let you send messages in long chats, if you just spam retry it eventually works

i never knew

[woven canopy]

Its free buddy, appreciate that

[gleaming roost]

[unkempt cave]

#video-arena-1 message
Lmaoo

[astral jetty]

The search bar and archive chat functions aren’t even appearing for me at all

There’s just a black bar where you would normally delete something with nothing in it

[woven canopy]

You need to factory reset your phone to erase all it's data

😂

[echo aurora]

Sorry for the errors! Can you try the steps in this article - https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message ?

[somber marsh]

It isn't letting me delete conversations

[echo aurora]

I'm seeing other reports of this. I think there is a bug, reporting now.

[somber marsh]

Thanks man

[magic ravine]

Nano Pro 2K is still not back 😢

[echo aurora]

Okay yeah this is very much a bug, it's been reported, we're working on a fix asap.

[glass zealot]

are you addicted to it ?

[cloud cargo]

It's the best image model available and people like using it.

And people like using it regardless of its failure rate or whatever problem it has that it was removed for, so people want it back.

[wide mauve]

good explaination sir

[woven canopy]

We all are

[astral helm]

@echo aurorashould've removed 1k instead

[sudden path]

How to makr our own video ?

[cloud oak]

pineapple..where's Seedream-4-2k?..theres only Seedream 3, Seedream 4.5 and Seedream 4 high res fal on my list.. 🥲

[echo aurora]

Check out #1397655624103493813 or https://lmarena.ai/blog/video-arena/

[echo aurora]

It's possible models may occasionally be unavailable for technical or other reasons.

[hushed gyro]

Would these pictures be recognisable as AI?

Used NB Pro

[hushed gyro]

2k has a much higher error rate than 1k before removal

[astral helm]

i personally can, bcz i saw tons of ai images and they tend to have a certain texture, but i dont think they're for the average joe

[astral helm]

could be

[hushed gyro]

Wait lemme post less

Tbf how can google manage to train a model to have this high consistency, quality and realism in these picture? Like GPT image doesn't even come close.

[astral helm]

nothing come close to nb 2.5 bro

it's incredible

[hushed gyro]

NB 3 pro has a very big difference to 2.5

Lmarena is the only platform that offers it for free

[astral helm]

i figured they have even better understanding for the prompt i bought freepik, but its not as good

[hushed gyro]

I mean look at the 3 pictures, this is nearly impossible before NB 3 Pro launched.

2.5 can't even do that

Got a new one, take a look

[astral helm]

its sick mate, imagine the 2k would show it even better

[hushed gyro]

I'll give you my generic prompt, can you go and try and generate some

For the woman provided in the picture, generate a realistic ____s female ________ uniform, gear and equipment

She is using the rifle shown in the picture (s), standing in an aiming position, side-view camera angle.
High-detail face, realistic proportions, cinematic lighting.

The rifle must match the reference exactly in geometry, proportions,
receiver shape, wear, and other characteristics.
Do not redesign or substitute the weapon.

Weapon accuracy is higher priority than clothing or background.
Preserve the rifle exactly even if other details are simplified.

Ultra-realistic, 3840x2160, sharp focus, photographic realism.

[cloud oak]

oh..it's image to image?..
-# was gonna try with grok.. 🥲

[hushed gyro]

Try with NB Pro on LMArena

Pls... 🥺

@astral helm

NB 2.5 vs 3 Pro

[astral helm]

gimme the womens face

[astral helm]

i think its better than every other model beside the nb pro obv

[hushed gyro]

Yes NB 3 > NB 2.5 > others

[hushed gyro]

Use this face every time, and use Nano Banana 3 Pro

Both models cannot follow the original appearance of the rifle (3rd pic)

[gaunt roost]

https://www.youtube.com/watch?v=YmRpH0kE4dk

[hushed gyro]

Pretty good! Very similar to my results as well

[astral helm]

ye they actually are, but only higher quality

[hushed gyro]

No AI model can generate an AR rifle that closely to real life

[astral helm]

it's good in everything, i barely use any other model

[hushed gyro]

I know right

Also really good with fictional weapons 🤭

[astral helm]

damn this is so cool

the setting the lighting the colors..

[hushed gyro]

However, it is very poor with bullpup weapons. Why isn't the hand gripping the pistol grip and the trigger?

I have a few more. Interested?

@astral helm

[astral helm]

no i actually have to jump to finish my work, great talking to u man

let me know when they get back the 2k version

cant wait for it

take care

[crimson bear]

Weird censorship on LMArena.🤔
A woman holding a gun that is meant to kill people is no problem at all (which I personally don’t care about).
But heaven forbid two women lie peacefully on a mattress in a bedroom and sleep, then the gates of hell are flung wide open.😱 👿 🥵 😵‍💫

[gleaming roost]

[echo aurora]

We are collecting examples of suspected false positives in #1447983134426660894 . I'll add this flag though, but if you come across others don't hesitate to share in that channel.

[worldly wind]

Prompt: The lyrical poet sits at a massive writing desk with a quill in hand. On the desk is paper with a folded corner. The poet is seated in a massive Baroque-style chair. Behind him is a huge home library filled with old books. On the desk stands a kerosene lamp. The poet's face is sad and thoughtful.

Hello

[golden ocean]

https://tenor.com/view/artificial-intelligence-neural-network-gif-3329491906550140056

[echo aurora]

Na it's an egg

[golden ocean]

https://tenor.com/view/komaru-getting-ddos-gif-1428811091514520825

[echo aurora]

clearly a hallucination, clearly.

[shrewd citrus]

is there a reason why I can’t use wan 2.6 image for the side by side comparison

t2i works fine

[balmy mist]

yo claude is generally a good person/machine lol

like wow

ever since 4.5 came out it really has changed my life

gpt has to be the worst, i literally almost lsot my marbles dealing with gpt, gemini is okay, its like a god middle model, but opus is good hearted, its weird how you can feel that

[hard quiver]

Opinions?

[shrewd citrus]

hmm what’s wrong with grok

[gleaming roost]

[proud bobcat]

Wan 2.6 came out???

[echo aurora]

Yeah this situation is a bit odd, what's happening is wan2.6-image is image-edit only, meaning it needs an image uploaded to work. For wan2.6-t2i doesn't have image upload available. Basically one model needs an image upload, and the other is preventing an image upload. This is a problem the team is aware of. I'll also check with the team image upload is something we can enable for wan2.6-t2i

[echo aurora]

Sure did!

[stone cobalt]

can it be that the links in the announcement here don't work?

[ocean bison]

Siren is ByteDance's Seedance 1.5 Pro?

Maybe!

[proud bobcat]

So peak

https://tenor.com/view/black-beard-peak-writing-black-beard-marshal-d-teach-peak-writing-barba-negra-gif-5446515270440960312

[echo aurora]

I won't be commenting on codenamed models, can't confirm nor deny.

[ocean bison]

Ok!

[echo aurora]

From this announcement? #announcements message

The seem to be working for me just fine.

[stone cobalt]

mh

[dusk dragon]

I am so sick of the hallucinations and the crappy Google grounding

It's so bad

I just can't

No way this crap is number one of lmarenas search

[stone cobalt]

yeah right, i was getting 404 now 2 times when i first clicked them but i cant replicate that behavior now

[dusk dragon]

Does anyone

Anyone have a solution for the crappy Gemini grounding

[frank berry]

video create

[echo aurora]

Check out #1397655624103493813 for more information on how to use the Video Arena bot.

Let me know if you have any questions.

[frank berry]

yes

[obtuse smelt]

oh yeah i see grok in website have issue

[tough shale]

Is anyone face the " error registration email " problem ?

[midnight phoenix]

What happened to the delete chat? Now I have to go to archive which is another step. Should have the option for a chat to be both deleted and archived in 1 shot.

[willow mural]

is it me or nana banana quality decreased

[obtuse smelt]

huh what happened ?

maybe your connection

[fickle pulsar]

What happened to z.ai glm 4.7?

[chrome goblet]

Where is the delete button?

[austere sundial]

Omg end of support
I'm out
last one to leave close the door

[fluid cove]

how to delete chat after you decided to arcive chat

need help

I can't delete chat

[midnight phoenix]

go to search

[fluid cove]

I clicked on 3 dots the delete option not coming

[midnight phoenix]

it is really dumb design decision, they should have kept the delete button where it was. Now to delete you must click on Search

[fluid cove]

Yes gone to search option and from there I clicked on three dots and delete option is not showing

[midnight phoenix]

Need to archive first

[fluid cove]

ok

done then

what should I do next

[midnight phoenix]

So you need to archive the chat first, then go to search, then find the one you want to delete then click on 3 dots and delete

[fluid cove]

I was lazy to take screenshot the delete option is not showing

[midnight phoenix]

Then perhaps it is a bug. Either way, they should really revert back to make it easier. Then have archive as another option...

[midnight phoenix]

@fluid cove you need to actually search for the chat like google then you can delete

[robust granite]

[fluid cove]

ok let me try

S the option came and deleted one chat then I tried to do delete another and the option disappeared

now I can't delete the other chats

[river wigeon]

May I ask if it only supports generating the video on the Discord channel? Can I generate videos that do not show on the public channel?

[timber sandal]

Hello

[timber sandal]

Report kaczka

[void swallow]

When am using claude opus 4 5 search model for web search, I am getting error. It was working properly earlier. now it says Something went wrong while generating the response. Please try again.

I tried refreshing the website, etc

[subtle frost]

yeah idk how its first place on LM arena
It's the worst searching model and its not even close
Ive made gems so it will pick a paper source it and its still meh

Claude and chatgtp are far better, hallucination stats say claude is better but tbh I find chatgtp slightly better especially with thinking enabled

Also I haven't tested deep research in a while but Geminis deep research also sucks in comparison
I feel like it just searches the query find the first 200 links and summarises it. Whereas chatgtp meticulously reads a source then carefully selects the next source based on the previous sources it read.
Itt ends up doing 20ish relevant sources and picks up nuanced information that I may of missed whereas gemini reminds me of me doing my first powerpoint when I was 12, just word for word reguritated whatever the shinest sources say with no nuance or analysis

[golden ocean]

https://tenor.com/view/come-at-me-bro-bring-gif-13239319

[dusk dragon]

So realistically I should give the gpt 5.2 search a better tru

Try

Because I just can't with the garbage grounding is giving me

I ask for help with a game or something random

And non stop making up random crap

[subtle frost]

Yeah I feel like it hallucinates less by an order of magnitude
Just personal experience tho

[dusk dragon]

Interesting

[hollow ivy]

<@&1349916362595635286> @quasi atlas another zombified account

[subtle frost]

Lemme know your experience after a while

[dusk dragon]

Sending you a friend request need someone to talk to about this

[hushed gyro]

Zombified

😂

Why are ppl getting hacked all the time?

[hollow ivy]

they forgot to get their vax ^^

[hushed gyro]

Rate these two images, generated by NB 3 Pro

Do they look bad or smth 💀

[hollow ivy]

i don't like war

[hushed gyro]

You pacifist?

Btw it's not just war, people shoot guns for fun/as a sport yk

[hollow ivy]

kind of
humans should not do war

the pics are clearly depicting a war scenario

[hushed gyro]

I mean most wars start for no beneficial reason

[hushed gyro]

Could be training 🤔

[hushed gyro]

But they look high quality, no?

[hollow ivy]

if all people rejected to touch weapons there would be no wars

[hushed gyro]

Weapons aren't just for war, it can be for self defense

I agree that wars shouldn't happen, but yk its just pictures, and I get what you mean

😭

[hollow ivy]

earth is the bad joke of the universe right now

it also explains, why we never get visits

they are teaching us as a bad example, how a civilization can go wrong

[hushed gyro]

I don't want visits

Halo-Covenant War vibes

[hushed gyro]

I agree that the world is a messy place rn, but unfortunately, it won't get much better

[hollow ivy]

if we can get a beneficial AGI, it could get better

[hushed gyro]

Preventing WW3 is already a very good thing for the world

Our social issues are way too big to fix immediately

[hushed gyro]

What if AGI decides that humans are trash to it 😭

[hollow ivy]

high intelligence has a high probability to also be wise