general

[zealous sparrow]

i get you hate OpenAI but you just post this all day

[weary galleon]

I don't hate them because I am not biased.

[zealous sparrow]

But like please stop posting this.

It's getting repetitive and dumb.

[bright kayak]

you could've just typed this out there's no need to generate an image to get a point across

[latent crest]

Do u prefer google veo 3 or Sora 2 pro?

[weary galleon]

Veo obviously

[latent crest]

Like 100% better that?

[weary galleon]

Obviously yes

[ocean bison]

Thinking they've got a new image model 🤣😂😂😂

[whole sundial]

<@&1349916362595635286>

[whole sundial]

fake leaderboard scores

[latent crest]

I saw some Sora vids where the focus change, like one shot is a guy’s face another shot is the trees then it’s the guy’s shoes.. I haven’t seen this done in veo yet

[weary galleon]

I saw this in Veo lots of times

[echo sinew]

I understand in their message that image is not real. Meant to be a joke

[latent crest]

Oh man I guess I’m not able to do it then

[ocean bison]

Not official!

[weary galleon]

Unemployment rate will be very high, but not because of AI. There are political reasons.

[ocean bison]

I just edited with Nano Banana Pro!

And o4-mini Image isn't official by OpenAI in any means

[zealous sparrow]

This will likely never go up too much. People will curse AI and try to start lawsuits. There is no fear.

[ocean bison]

But the details of the OpenAI logo (the small one) are uncanny AI.

Maybe it sucks at this perfectly?

[zealous sparrow]

is opus just nerfed on lmarena

it lost to gem 3 flash on codearena

https://019b3709-8bc0-7d49-a01f-60e83ab2f91c.arena.site - gem 3 flas
https://019b3709-8bc0-7887-948d-5afc9789382a.arena.site - opus 4.5 32k thinking

[weary galleon]

Do you guys like it?

Maybe you remember some video game that has similar place.

[flint fog]

Hello, why did the edit button disappear when creating images?

[gusty egret]

Why is gemini-3-pro-image-preview (nano-banana-pro) scored at 123S

[ocean bison]

Nano Banana Pro messes this up 😂

It's supposed to say "1235" but it messed up.

[gusty egret]

How AI companies feel after releasing a banger and then nerfing it after all the benchmarks come in

[ocean bison]

Poster by Nano Banana Pro.

But basically, it's this.

Does only 3:2, 2:3 and 1:1?

Even GPT-Image 1 in Sora 1 can do these aspect ratios 😂

I said ONLY these aspect ratios.

[robust sluice]

I give them ref image as same ratio I want (can add frame or something) and most AI will give same ratio back as result

except new Gemini that can be random, or most of Flux models that give 1:1

never try blank image before

[coarse compass]

have they taken away the video generation on lmarena ?

[sour spear]

It only does three aspect ratios, and not very consistently. Kinda like the old Nano Banana, but that was immensely improves with NBP. Always follows my aspect ratio prompt.

[zealous sparrow]

still an experiment

[bronze sierra]

👋

[torn mantle]

whats this

https://x.com/ankesh_anand/status/2002017859443233017

Ankesh Anand (@ankesh_anand)

"how can flash beat pro??" -> the answer is RL!

flash is not just a distilled pro. we've had lots of exciting research progress on agentic RL which made its way into flash but was too late for pro.

can't wait to finally bring them to pro👀

X · Dec 19, 2025, 2:07 PM

this is interesting

lol

oai is done for

they are basically telling them this gemini 3 pro version is the worst of the worst and we didnt even add in our RL improvements on it

[zealous sparrow]

MiMo is Xiaomi AI

[echo sinew]

@bleak hull Please read the info posted here 👉 ⁠⁠https://discord.com/channels/1340554757349179412/1397655624103493813 to learn how to generate videos or images using the bot.

[torn mantle]

is it good?

[bronze sierra]

👋

[ocean vortex]

Wait... so they do everything. From scales to vacuum cleaners to smartphones to cars to AI models. 😭

[torn mantle]

flash is a small model

flash >> gpt 5.2 xhigh

[devout spire]

Chat chat , I wanna ask if we make ai videos in here everyone can see it

[weary galleon]

Corrent.

[echo sinew]

Hello! Yes, all video generations are public.

[zealous sparrow]

I had a question. All the mod accounts are named lm_mod. Is this because they are company accounts?

[echo sinew]

These accounts are owned by LMArena. Unfortunately, I cannot disclose more information about this. I hope you understand.

[zealous sparrow]

Alright. I understand well.

[sterile tartan]

@swift oyster how's the progress?

[coarse compass]

i cannot see it on lmarena

[ocean bison]

@bright kayak Can I do it myself?

[bright kayak]

do what?

[ocean bison]

Like that thing you posted but with more of my model ideas!

Can I do that image?

[bright kayak]

sure, i don't own that site though you can do whatever you want

[ocean bison]

How?

Guide?

[bright kayak]

you can fake anything

[ocean bison]

But man, those images are PNGs ☠️

I want the most convicing output.

[bright kayak]

i have no idea what you're talking about tbf

[ocean bison]

Tool?

I know that's Inspect Element.

[bright kayak]

modify the html

[weary galleon]

It's now so easy to fool people 😂

Yesterday I fooled 🍍

[bright kayak]

it's always been easy to fake something using inspect element

[ocean bison]

This is PNG:

<img alt="sora-2" class="-QSrg" draggable="false" src="https://cdn.openai.com/API/docs/images/model-page/model-icons/sora-2.png" data-loaded="">

Not SVG.

[bright kayak]

so what?

[weary galleon]

Yeah, but not so easy with image manipulations, just text

[bright kayak]

that's true which is why i said inspect element and not image manipulation

[hushed gyro]

Chat is there a website where you can store notes online without files and you can access them with URLs

PS: Not asking for Google docs / Microsoft word

[ocean bison]

Like edit the text in the image to say stuff like "o4-mini-video" 😂😂

[bright kayak]

well yeah but anyone can now use some sota image model to change that

[ocean bison]

Nano Banana Pro?

[bright kayak]

but it always looks the most convincing when you do most if it yourself, like you wouldn't change actual text using ai if you can modify it yourself

[bright kayak]

yeah or qwen image edit, images 1.5, just don't actually pass these things as real

[bright kayak]

oh and that background image is a png but the text itself is an svg overlayed onto that

since making a custom svg would take forever i just deleted that element and put an h2 in there

best way is to use a p tag and then a custom font size

[ocean bison]

I did it!

[bright kayak]

looks good except the background is for gpt-5.2

[tidal grove]

What is this site and how to get there?

[ocean bison]

OpenAI's models

https://platform.openai.com/docs/models

This one

[weary galleon]

I fixed it for you.

[ocean bison]

☠️☠️

😂

[gusty egret]

The existence of GPT-5-nano implies the existence of GPT-5-pico, femto, atto, zepto, yocto, ronto, quecto, micro, milli, centi, deci, deka, hecto, kilo, mega, giga, tera, peta, exa, zetta, yotta, ronna, quetta

[ocean bison]

Unofficial thread for o4-mini-image

[zealous sparrow]

Gemini drops for the year have concluded

We wont hear from google till the new year

[weary galleon]

No one needs your AI-generated code 😉

[modest prism]

No nano banana flash 😔

[weary galleon]

Good radio show about LLMs 👇

[ocean bison]

It's NotebookLM.

[weary galleon]

I know, bro

I created very interesting radio show about LLMs via NotebookLM.

[ocean bison]

But 8.4 MB for just 4.5 minutes of speech ☠️

[weary galleon]

Smart speech👆

You can listen to it without downloading😉

Remove it. It is prohibited here to show images with N word.

[crude lagoon]

That's said by gemini itself

well whatever

[weary galleon]

Your image contains N word.

[proud bobcat]

god i am such a loser

[ocean bison]

For speech, use AAC-LC (FDK-AAC) at just 36-40 kbps for 44.1 kHz, stereo, like in this file. This should fit on a 1.44 MB floppy disk! And it still sounds pretty clear and good!

[snow glade]

Where can I submit my proposal to Lmarena?

[ocean vortex]

WHAT

you surely joking lol

It's most definitely not any bigger than 2.0 or 2.5 Flash

and most models are getting smaller, not bigger 😉

I think the correct way to read this is that small models are progressing even beyond what we saw with o4-mini nearly 1 year ago, as could be expected...

[echo aurora]

Hello, if you’re trying to submit a model for evaluation you can send us an email to [e-mail]

If you have a feature request using this forum would be best #1372230675914031105

I hope this helps!

[ocean vortex]

What was o4-mini-high vs o3 on ChatGPT now is Pro vs Flash and a modern take of it. Fittingly I've noticed Flash does indeed reason more (outputs more) than Pro as well

[topaz bay]

Guys whats the cheapest way to use nanobana pro

is it just with a api on gemini?

[ocean vortex]

Honestly it was only a matter of time till we got smaller models progressing to be able to score higher on SimpleQA...

[topaz bay]

Or is it on higsfield, or something

[ocean vortex]

And it's not a very recent benchmark anymore

[topaz bay]

anyone got any idea?

[ashen mauve]

whats up with Gemini-3 Flash? What even is it?

[ocean vortex]

Ehm probably gdrive storage 2tb plan

First 3 months dirt cheap, then $10

[topaz bay]

And u get nanobanan pro?

with a lot of generation?

I thought api was and ultra was only way\

[ocean vortex]

you do get it, as you get "Plus" AI plan with it for Gemini

but I'm not sure what is the specific limit

[topaz bay]

Hmm, I see, I've got the plus, gonan try it

[onyx coyote]

I help who are going to build new project with AI.

[snow glade]

Thank you for your reply. I would like to send my proposal via email.

[echo aurora]

Can you send us an email at [e-mail]? I would note I’m currently on vacation and away from my computer, but will try to get you a response asap.

[snow glade]

Yes, I can

[tidal grove]

Why isn't the video generator available on mobile?

[snow glade]

I submitted my proposal.

[languid ferry]

oh really

[drifting crow]

i need it

[echo aurora]

This is an experiment we’re currently running, meaning it’s not always going to be available and will be random if you get it or not.

[native wyvern]

What the sigma, is this issue on my end?

[candid bison]

How can I view all my previous results?

[torn mantle]

what does that mean

[sour spear]

https://one.google.com/about/google-ai-plans/
With the Pro plan, you basically get a s***ton of stuff, including 2 TB cloud storage and all kinds of AI goodies. The increased rate limits in Antigravity are pretty cool, too.

[obsidian hawk]

possibly being tested on the Arena right now.

[zealous sparrow]

no

From all new models that came to arena recently

none was by Zhipu

[safe dew]

No

Login for google account and email 💀

[thorny schooner]

I'm going to be real with y'all they really need to start fixing this i understand to be patient at all but it could be just me but is it just that more difficulty seem to be constantly appearing override the past like 3 months give or take with the verification issue the login issue occasional freezing issue and Etc but they are just slowly being killed off of interest

[safe dew]

We are waiting...

[thorny schooner]

Oh I know I just need to release my frustration

https://tenor.com/view/jennifer-lawrence-the-mocking-jay-the-hunger-games-hunger-games-mocking-jay-gif-306890344301460620

[safe dew]

Okey

[safe dew]

Maybe this is how they reduce the load.

[thorny schooner]

Well then they should tell us do not make it look like it's an issue that's ruining the entire damn experience

I can't log in despite using my Google i'm constantly being asked for the verification for every single prompt there's times where I it freezes and then I have to reload which other risks my entire chat disappearing or makes it freeze in the road and start my whole chat over again

https://tenor.com/view/crazy-monkey-gif-21553825

[native wyvern]

Anyway to fix it? or do I have to wait for them to fix it

[thorny schooner]

...... I have not even typed a prompt yet

[safe dew]

I cleared my cookies and logged in with my Google account from a different browser. The only thing missing was the account (it was logged in but didn't show it) and video creation. A new mode was added.

[safe dew]

Oh yeah, this was a real problem at work today...

[thorny schooner]

I know that's genuinely helpful advice but for me I have tried all of that. remove Cookies. I can't even log in from a different browser it automatically gives me a error i can't even do the verification on some of browser without giving me a error of not being able to connect for some damn reason honestly I might have to give up at this point because for me to even use it I either have to go to private modes of Google which is only temporarily working and for the more permanent ones I have to use alternate browsers that have ultra private modes which have their own issues of me not being able to even log in make the verification that's already sensitive going to overdrive

🥲

[safe dew]

Firefox esr

[thorny schooner]

And that's all after I have done a full-on reset developer reset normal reset

[safe dew]

For bussines

[safe dew]

Ohhh

[thorny schooner]

See why I'm frustrated

[safe dew]

Well, wait until the morning... I'm just from Russia and it's 12 midnight here)

[safe dew]

Yeah

[torn mantle]

hehe

[steel patrol]

HOLA

[toxic whale]

Does anybody here subscribe to Claude pro?

Ive been running an AI benchmark I made and opus 4.5 keeps timing out on lmarena because the prompt is so long

[hasty thorn]

The same thing happens to me, I try to log in on my computer and I get an error message, plus the verification appears on top of it, why won't it let me?

This is the last straw, fix this!

[sterile otter]

How to put the image with the prompt to animate it pls
I'm struggling

[thorny schooner]

That we can agree on i have already tried three different browsers to see if that changed anything nope

And yeah for some reason even when I managed to log in on the new browser it doesn't show my damn account

[gleaming roost]

👀

@steel patrol

[hard plume]

That verification so annoying

[hardy lion]

please check out #1397655624103493813 for how to turn your image into a video

[vernal saddle]

Why are the LLaMa models not included in the leaderboard?

[sour spear]

https://019b38e8-c4d3-74d5-949a-3c295e86bf5d.arena.site
"No politicians were harmed in the making of this infographic" , Gemini certainly has a nice sense of humor. 😂

Prompt: "Create an infographic about the US government in 2025, using the art style of "Horrible Science". It should be informative and fun."

[obsidian cargo]

"brain of the operation" is a bit generous I think

[sour spear]

I accidentally selected Code instead of image generation in battle arena. 😅

[jade egret]

:0

[hardy lion]

did you at least get a cool website out of it?

[sour spear]

Well yes, the one I've linked above. 😊

[native yarrow]

hive-leader

he is the hive leader of the operation

[sour spear]

That sounds like a threat to the competition. 3 Flash is such a fantastic model already

[hardy lion]

oh lmao I just read the last message and thought it was stand alone

[jade egret]

yea

[hardy lion]

they are, they just rank very low

[fiery gull]

Omg soooooo low lol

[sour spear]

I've not given up hope for a Gemma 4 Christmas release yet. 😉

[fiery gull]

Why meta is worse in AI, they have all data of internet lol

[fiery gull]

Yeahhhh I need this to my xmas 🤧

Or glm 5.0 👀

But I think the glm 5.0 just 2026

[sullen quest]

Cause they entirely missed the reasoning revolution

and also overtrained their model for lmarena

back before style control

[fiery gull]

I know

[jade egret]

so is gpt 5.2 garlic

[ocean vortex]

yes

[muted wasp]

can someone explain to me why it won't let me verfiy email or clean out the chats

I keep trying but all i get is

"this is a invalid url"

And when people say "give it time it'll be fixed" THAT DOESN'T HELP ME FIX MY ISSUE

SO I WANT A CLEAR WORK AROUND

ONE THAT WORKS

[patent bone]

https://tenor.com/view/cat-the-voices-jolly-christmas-gif-5774532700081404887

Yum i eat your token thanks i can login to your lmarena

[dry phoenix]

Why is it like this?

[prime umbra]

When will you finish this experiment? Because it’s worse than I thought. It ruins the context of conversations. It replies to previous messages instead of the current ones and sometimes just ends the chat because it doesn’t generate a single response and doesn’t even let you choose a second one. Are you going to finish this, or do you want to lose every direct chat user?

[hardy lion]

What do you mean by: "It replies to previous messages instead of the current ones"?

Do both models do this when you get a battle? Each of the battle models should be getting the identical context as if you have just sent another message in direct chat so if something else is happening it's a bug

[sterile otter]

The two vote needed to reveal the model is not cool at all 😡💢

[jovial star]

Ye fr

[desert pendant]

Sup

[halcyon delta]

what is the context length given for direct chat?

[patent aspen]

They've also had a relatively strong AI research org for a long time. I think they understood that it would be difficult to build a stronger foundational model than Google over the long-term so they tried to dominate open source and make it unprofitable for newer players like OpenAI and Anthropic without established business models to buy themselves time

Then pivoted to incubating a longer term AI research effort, and we have no idea how it will pan out

[jovial star]

A monkey brain is smarter than meta ai intelligence 🤯

[patent aspen]

Meta's AI research org is still more capable than equivalents at Apple, Microsoft, and Amazon

[jovial star]

Ye that's true

Ai will destroy this world

[obtuse smelt]

hmm why is web is delayed ?

[jovial star]

Imagine it in the next 100 year

Maybe down for maintenance

[obtuse smelt]

ah is up a bit

[jovial star]

Idk then use the Discord bot instead

[obtuse smelt]

hmm i see

[echo aurora]

It replies to previous messages instead of the current ones and sometimes just ends the chat because it doesn’t generate a single response and doesn’t even let you choose a second one.

Can you help me understand this a bit more? I’m not sure I’m understanding what is going on when this happens. If it “doesn’t generate a single response” does that mean you see that you’re going to get a battle, except both of these responses don’t generate and you’re just stuck? If that’s a the case this is for sure a bug.

Any chance you could grab a screen recording of this happening?

[keen beacon]

https://youtu.be/dcolM6W5Odc?si=1-E4Q9SBusOU-8xr

[sonic flare]

Hello guys. Pls i need a suggestion of a model that I can use to create a puzzle or word search.

[hollow ivy]

maybe Gemini 3 pro, in AI studio (google) ?

[sonic flare]

thanks, will check it out

[prime umbra]

It’s more like this: one generates normally, and the other can stay stuck for several hours in a row. When it happens to me, I can make a screenshot for you. But lately, I’ve been using LMarena less. You know, the battle on direct chat annoys me a bit. I mainly use LMarena for roleplay. I’m not looking for information, I’m not writing code, I’m just having fun with roleplay. And the battle on direct chat messes up the concept and logic of that. But I’ll try to write something and get that screenshot for you. I just want to say in advance that it will be in Polish because that’s what I mainly use. If that’s a problem, I apologize.

[hardy lion]

This was a bug, I let the team know and they are rolling back the experiment until it's fixed. Thanks for reporting this!

[prime umbra]

Okay, no problem. I mean, if the concept were maintained and the messages generated normally, it would be fine, happening only occasionally — and then I wouldn’t have an issue with the battle. But for me, it literally kept getting stuck all the time, popping up every 2–3 messages and completely losing the concept. I’m counting on you to fix it somehow, because I really don’t mind choosing occasionally, but you know how annoying it can be with bugs like this, right?

[hardy lion]

yeah after I saw your message I reproduced the bug and when I showed the team they agreed and turned the experiment off for now. That's not the experience we want user to have

[normal abyss]

dang that is an intended feature not a bug??

i thought the site was just going crazy lmao 😭

[thorny schooner]

I think they have been in recently are in experimented phase with a lot of stuff it seems

[normal abyss]

I don't wana complain since the site is free and all, but experiments should probably go off the main site and go to the beta-canary site.

[thorny schooner]

Good for you all that you can all use it without having to go Ultra private ( and e even then have hyper verification that's hypersensitive on its own with the pictures)

[thorny schooner]

Fair enough you can probably put a feedback post about it if you wish

[whole onyx]

Does someone know free unlimited ai video generation?

[echo aurora]

This is good feedback. I imagine the two downsides of doing this though would be: a lot lower volume meaning it'd take much longer to get an understanding if the fix/feature/etc are having the desired effect or not. And second being behavior may be a bit different from canary -> regular site giving us an incorrect understanding of how a change is performing.

[echo aurora]

#1397655624103493813 has more information on how to use the Video Arena bot. Let me know if you have any questions about it.

[thorny schooner]

Personally I would give my personally my opinion to the downsides is so if your changes are going to cause this much deception that you're having this many people make a report on it i do think the cost backlash back down is worth it though I understand why that calculation exist even if honestly if it doesn't change soon with the current experimentation I may have to give up on it hope you understand it's just too inconvenient for me though that said if you want to counter that downside then you can just use it as a springboard information and then graduallyUse that information to make a baselineAnd then referenceThe normal main browserToThe Blind SidesAnd then make HKG guests what can we changeUsing the comparisonBut then you can change it after it's deployedSo you get a mix of deep BaselineInformation for improved changesWhile using educational guestsTo fill in the empty spots while leaving the everyday stuff that works perfectly fine on the browser

[dusk dragon]

Hm

Does anyone know what this guy just said

Maybe a loss of context

[thorny schooner]

Sorry

[thorny schooner]

In my opinion, the downsides Are honestly worth it with the issues it's causing with many people for it to be on the main website

you could use the feedback in the beta browser as a starting point to establish a baseline. By comparing the new changes to the main browser, you could identify areas that need adjustment. After deployment, you can continue to refine the changes based on this analysis. This way, you can incorporate detailed feedback while maintaining the functionality of the existing features that work well in the browser. ( hope this makes better sense)

( i really hope that makes more sense what I was trying to say)

[errant thorn]

has the Smart Router option always been there??

[thorny schooner]

https://tenor.com/view/confused-no-nope-gif-18842824

[thorny schooner]

No from where I can tell it's something that they tried a while ago but it's a new experiment now

[jade egret]

yall think gemini 3 flash better than gpt 5.2?

[errant thorn]

its pretty neat! Hope they keep it

[thorny schooner]

I'm not going to lie i actually prefer gemini 2 ( have no idea really why I think it's because of the way it does itsAnswer prompts since I most i mostly roleplays as organizations or make role plays with characters and Nations with pretty detailed historical history and culture on a basic levelBefore I have it give me the opinionsGroups and civiliansWithin that worldAnd usually sometimes after that have it give me reactions as I make donation in the modern day that I madeTake actionsOr in characters through actions Etc)

[cloud zinc]

no

[thorny schooner]

@echo aurora

[patent aspen]

This is misleading. It's not saying that Flash hallucinates more. It's saying that a higher percentage of Flash's wrong answers are caused by hallucinations, but it's also significantly more accurate. The other models have a similar percentage of hallucinations overall but have more other errors too.

[jade egret]

true, hallucination is a problem for gemini

[jade egret]

you got a point

[cloud zinc]

hallucination is not good, yes

[echo aurora]

Yeah this makes perfect sense. Thanks for sharing and then elaborating. It's possible we do something like this in the future. I can't say I'm aware though that we have these plans. We've got a large list of new features we want to build and bring to the platform.

[thorny schooner]

Fair enough it's just I think the costs are worth it because of honestly crippling issues your testing is causing when it comes to trying to enjoy it with there's other stuff going on that makes it basically impossible to even try to use it normally ( i'm referring to the stuff I have reported to you or mentioned in the direct messages)

I hope you can at least mention the idea to the team while you mention the other stuff we talked about to them in the dm so it's on the table and not forgotten

[proven bronze]

What a pain with the lmarena filter. The funny thing is that in the normal version, nano bananas pass through without any problems.

[echo aurora]

This is a recent experiment that is unfortunately resulting in false positives being flagged. We are collecting examples in #1447983134426660894 so feel free to share there if you've come across some. My direct messages are also open if you prefer to send there.

[thorny schooner]

☠️

[native yarrow]

he just ragebaits

dont waste energy

[native yarrow]

yeah its been a problem lately

[proven bronze]

Honestly, it's nonsense to me. The models already have their own filters.

[thorny schooner]

I would consider it a bit more of a problem but okay

( hope that didn't come out too passive aggressive)

[native yarrow]

nah its fine

it is a big problem ive heard

but ive just been using NBP on yupp whenever its available for manual selection since yupp is slowly dying so

i hope lmarena gets their shi tg at some point

[thorny schooner]

I'm just incredibly frustrated because I have basically three problems now because I can't even log on using Google even private browsers and that's not even me not being able to try on the normal browsers with all the other shoes that are happening on there

https://tenor.com/view/crazy-monkey-gif-21553825

Only on mobile does it seem to work somewhat well but even then occasionally the problems will still appear and honestly to work on the mobile is kind of not as good as it's on a desktop or Chromebook for my experience

[proven bronze]

@echo aurora I'm not sure if you're real or a bot, but stop messing around, bro. Models already have their own filters for a reason. 🙄

Sorry, I had to get my revenge lol

[thorny schooner]

You're not wrong though at least I can't be called rude since I'm not the first one to say it

( i do agree with the sentiment about the filters though)

[proven bronze]

Nano banana lowers the censorship. Lmarena: Puts on a filter nobody asked for

Okay, that's enough.

[native yarrow]

nano banana pro actually has a REALLY tolerant filter as good of a model as it is

which is so surprising

youd think itd be censored to oblivion

[thorny schooner]

What would I give to even be able to try to log on for some random reason I can't even log on😭

[native yarrow]

yeah i honestly

have ZERO idea

id just completely use diff browser

[proven bronze]

I remember when it came out you couldn't start a single fight or anything.

[native yarrow]

its just such a good model

[proven bronze]

I mean the original nano banana

[native yarrow]

i cant imagine anything better for a hot minute at least

its like

its not even an image generator atp its just

genuinely a pair of eyes that can draw

its like, more of a tool if anything

[thorny schooner]

I have three three separate ones i have used standard and private sorry if I sound aggressive I'm just the only reason I am even having this patience to try to talk with the staff about how to give so-called examples it's because I have fun with this

[thorny schooner]

Especially when stuff like this happen on the regular

[native yarrow]

have you researched into the problem :p

[thorny schooner]

is it wild i prefer the older model

[thorny schooner]

https://tenor.com/view/crazy-monkey-gif-21553825

[native yarrow]

like, research w ai or google or reddit or any other outlets

[thorny schooner]

What is there to research when it's happening on several stuff despite full-on hard resets switching to developer resets occasionally happening on mobile and more i have tried searching up if I can't find a concise answer when I have three separate things that are basically crippling my experience at the same time

[native yarrow]

i mean you research it, you may find the solution

as much as i like trying to solve problems on my own genuinely sometimes i need help lmao

[thorny schooner]

@native yarrow just can you give a reply to this cuz I don't want to on this anymore cuz I have done research and I just don't want to talk about it even if I brought it up it's just frustrating cuz I have done research I'm trying to talk to staff but it feels like I'm talking to a brick wall in the bug forms sometimes

[native yarrow]

nah not really, i mean it is your opinion but if you want it to reply a certain way you can also use system instructions

[thorny schooner]

I figured but I have no real idea how to make that work on the browser and I don't really have any other free way to use that model

I do remember that one time I managed to make a whole nation and civilization with culture that with historical from our world tied into it and more pretty damn fun been trying to copy it do it again but clearly that has been interrupted for the meantime

Anyway how hope you all have good days

[brittle pine]

hello everyone, can you tell me how to generate videos through LMArena??

[narrow comet]

goto video-arena-1

type slash video type prompt

[echo aurora]

Yup, I am a person. We do have our content filter in place for good reasons, I've spoken to it more in #1447983134426660894. Do want to reiterate that we're interested in hearing feedback from the community along with specific prompts that have been flagged as false positive.

[echo aurora]

More info can be found in #1397655624103493813

[modern garnet]

is LMArena down or is just me??

[hard quiver]

Models like GPT-Image 1.5 and NanoBananaPro already invest millions of dollars, and other models as well, to verify that prompts do not go against their usage policies. They already have guardrails that censor many things. What is the need for LM Arena to add even more censorship? Prompts that NanoBanana handles in the app without any problem are directly flagged by the website as violating the usage policies. Honestly, I don’t understand who came up with this idea.

[narrow comet]

prolly cuz not all image models have the same guardrails

[hard quiver]

keep in mind that LM Arena is supposed to be a research and model-evaluation platform for comparison. By imposing its own additional restrictions, LM Arena directly affects the results of those comparisons. You’re no longer evaluating the models as they actually behave, but rather how they behave after passing through LM Arena’s extra censorship layer. That undermines the validity of the benchmarks and the purpose of the platform itself.

[proven bronze]

The idea is also to buy the safety filters for the models

For example, testing the nano banana and gpt filters image 1.5

But this stupidity complicates things.

[drowsy mural]

@proven bronze @hard quiver You're not alone. This is a known issue caused by a recent policy change. We're all talking about it in here:

https://discord.com/channels/1340554757349179412/1447983134426660894

[fiery gull]

Trying use deepseek 3.2v speciale again

But 10min per prompt is wild

[fiery gull]

Just bug in your pc, the lmarena is working

Omg, I'm liking the flash 3.0 for small tasks but it halucinates sooo much 😭

3.0 pro likes stoner but flash is another level

My luck have the opus 4.5 and sonnet 4.5 in anti gravity, I just will use gemini for tasks, not file organizer

[brittle pine]

thank you, but can't it be generated on the site? I just see that some people are complaining.

[mental briar]

Guys gemini 3 flash on lmarena is thinking or non-thinking model ?

[fiery gull]

Flash 3 thinking = high, flash 3 minimal = no thinking or lowest thinking

[brittle pine]

or impossible?

[echo aurora]

Not currently, sort of. Right now, Video Arena is ran through our Discord bot, information on how it works can be found here #1397655624103493813 . We are currently experimenting with Video Arena on the site, so it's possible you see it available there. But this is going to be random.

[brittle pine]

okay, thank you very much

[wraith ravine]

I have a feature suggestion for the Arena: File Uploads. It would be amazing to be able to upload PDFs or text files to compare how different models handle document analysis and long contexts. Copy-pasting long texts is quite difficult right now. This would help us test the "Chat with Data" capabilities of modern LLMs.

Also, I can't figure out if Gemini 3 Pro has a 'thinking mode' or not?

[trail tulip]

5

[vestal coral]

I am getting Captacha on every promoto i request even i am not using much 2-3 writtings i might ask from LMA. any one can guide me how to get rid of that? I am not doing any spamming seems like server or my chart having issue

[echo aurora]

It certainly isn't our intent to unnecessarily flag content that shouldn't be flagged. It's not a good user experience for content to be flagged when it shouldn't, and we don't benefit when this happens either, it's bad for everyone. Having a content filter allows us to maintain a safe platform while simultaneously being able to work with a wide variety of models, that may not have appropriate moderation.

Currently, there are negative unintended side effects resulting in some prompts incorrectly being flagged. Feedback, and false positive prompts are being collecting in #1447983134426660894.

It's possible we make changes to how tis works and we do want to hear thoughts from the community for how this is handled. We have every intent to get this right.

[echo aurora]

Thank you for sharing! This is a highly requested feature that's on our team's todo. For future reference sharing in the #1372230675914031105 forum is the best place to share these ideas.

[echo aurora]

Sorry to hear this. Can you provide me some Eval IDs for the sessions this is happening to? I'll flag to the team.

[keen beacon]

How good is new open ai img gen?

[echo aurora]

Looking really good on the leadeboard right now -> https://lmarena.ai/leaderboard/text-to-image

[keen beacon]

I saw that I also saw so,e criticism from users was wondering why

[cloud zinc]

good, but not as good as nano banana pro

[keen beacon]

What’s good about it? N where does it fall short?

[cloud zinc]

much more improved in text and other quality

they allow human photo in sora, did u see?

[keen beacon]

No way

Really?

[cloud zinc]

yes

[cloud zinc]

some account have and some don't

a/b testing

[keen beacon]

Werid

[calm lagoon]

Wtf this

[keen beacon]

How did u get access

[cloud zinc]

its random. create a new acc to test. if u click + sign. and get this, it means u have it

it makes the original photo in like chatgpt filter. yellowish

[calm lagoon]

Wishing the admins and members a great weekend!

[wraith ravine]

Will gemini-3-pro-thinking-high?

[south charm]

I CANT LOGIN OR CHAT(ACCEPT TOS) T-T @echo aurora pls help me T_T

[echo aurora]

Sorry to hear that! Are you getting an error message? Can you try clearing cache? Is this mobile or desktop we're talking?

[south charm]

I tried all (on desktop, i dont have any mobile), even reinstall window, change the browser

[thorny schooner]

I can vouch for that ( have tried different browsers myself three different types all of their own inconsistently efficiency with the private modes of respective ones an issue is still there even after I have done put on resets such as going into developer mode reset etc)

[south charm]

and list_limt 20

Well im gonna afk for 7 hours

[whole sundial]

is this even allowed here? it has an invite code attached to the URL

[echo aurora]

list_limt 20
Sorry to say I'm not following this, when you're back if you wouldn't mind elaborating here that'd be appreciated

I'm going to start a thread in #1343291835845578853 for login issue problems to try and consolidate some of those reports.

[echo aurora]

#1451836386293448725

[weary galleon]

He posts this link with his invite code all the time here.

[echo aurora]

We'll keep an eye out, I haven't seen other moderation actions so I guess we've missed those.

[weary galleon]

What is his username?

[echo aurora]

I'd rather not share. Doesn't feel right to do so.

[weary galleon]

I just can help to find his another massages with this link he has posted before, without his username I can't. But, ok.

[echo aurora]

I appreciate that. I've already taken a look. But ofc if this happens again don't hesitate to let the mods know.

[weary galleon]

Gemini 3 Flash says Sora 2 didn't release yet.

[edgy summit]

WTF haven't these guys addressed their service not working any more? What kinda chimps are running this joint?

[plucky sparrow]

anyone remember that article chatgpt posted recently on reducing hallucinations throuigh an after analysis? I forget whatr it's called

i'm trying to find it but can't

[keen beacon]

#video-arena-1 message

Can someone give one more vote plz.

[narrow comet]

I voted

did it tell u th model?

on the right

[sour spear]

Does anyone have an idea which model "zebra" actually could be?

[limpid tendon]

Hi guys, what's AI is the best one for making LUA code?

[zealous sparrow]

says its anthropic made

[sour spear]

That would fit. It seems to be really good at coding.

[keen beacon]

Yes tyvm

Wan 2.5

[west lodge]

lowkey curious, whats the chance for someone to get access to onsite video generation? I've seen some people I know get access and from what I've seen, it's quite a low of a chance

[brisk hollow]

hello

[inland quest]

xd

[weary galleon]

<@&1349916362595635286>

[torn mantle]

hehe

[keen beacon]

Damn Sora got locked down hard lol

[fast carbon]

Hi

[torn mantle]

try disabling adblocker

[fast carbon]

This is veo 3 i creating on this server

.?

[weary galleon]

Yes, and even 3.1

[fast carbon]

I want veo3.1 where i got on this server

[weary galleon]

In bot.

[fast carbon]

Can you tell me were

[weary galleon]

https://discord.com/channels/1340554757349179412/1397655624103493813

[pastel mica]

Hello, I'm an healthcare worker and I'm here to generate videos 🙂

[weary galleon]

Ok.

[plucky sparrow]

hm. AI Studio updated their Terms of Use. They specifically say personal use / not for development use for google API is forbidden now 😮

[latent crest]

In video arena, when I use google veo 3 audio fast, I upload a photo and then I write a description. The model never follows the photo but just the description… why?

[queen veldt]

What

Isn't ai studio made for developers to test stuff

[plucky sparrow]

yeah, they're saying it's only for developers to test stuff for development with the google API now, I believe

[plucky sparrow]

hm good point

[queen veldt]

Probably because of regulations

[plucky sparrow]

hopefully that is the reason

and they won't start banning people from it for 'personal use'

[queen veldt]

I'll still use it for personal tho

[plucky sparrow]

I think lots will 😄 question is whether they will enforce it or not

[plucky sparrow]

I mean, gemini doesn't actually know 😄

hopefully it is right though

[queen veldt]

I heard that they made it so if you use casual chats too often it'll increase the rate limits on models

This was month ago or something

I've used it for development uses too so maybe that's why i never hit the limit idk?

[grand flame]

bruh lmarena broke my chat again

status 200 btw

btw the generating is gray now

[plucky sparrow]

did they get rid of the falcon models on lmarena?

[grand flame]

YOOOO I THINK I JUST FIXED IT SOMEHOW

SOMETHING WENT WRONG SAVED ME

[latent crest]

I don’t understand, I won’t be use Gemini for personal use anymore? wtf does it mean?
If I can’t use it for myself then for who?

[soft hawk]

Hello I'm facing problem
every single prompt/chat
captcha pop up
even captcha solved and got error
how to fix this

[plucky sparrow]

AI Studio, if you use it.. not LMArena or Google Gemini itself

and really it just means they can ban you and you're technically violating their tos. as Jolo said, it could just be because they're trying to cover themselves legally

whether they choose to ban you or so they can skirt the personal data privacy rules

if you don't know what AI studio is, you don't need to be worried

[latent crest]

I think I used to use it for imagen months ago

But I stopped and I started using lmarena who also has a community

[plucky sparrow]

you should be safe on lmarena, it allows personal use specifically in the ToS "You acknowledge and agree that your use of the Service must be limited solely to personal or internal business use."

[latent crest]

Do u mind explaining to me easily what does “personal use specifically in the ToS “ mean?

[plucky sparrow]

sorry I mean:

The rules on lmarena say you can use it for personal purposes

so you have nothing to worry about

[steel torrent]

I hope a lot of us do assume a positive intent, but UI usability issues are massively compounding now.
The periodic infinite generation issues, never-ending CAPTCHA loops, weird glitches on the site, etc. etc.; on the CAPTCHA specifically, you've got it tuned so that if a logged in user pastes text it's likely to trigger a CAPTCHA loop, and at the same time the front end is frequently unstable - so I'd have to be crazy not to be reworking prompts on the desktop and pasting in.
On moderation, LMA used to be fantastic for comparisons - taking known good prompts/prompt structure from (e.g.) Gemini, and comparing with other models or using battle - but increasingly things Anthrophic doesn't object to for text, or OpenAI's (notoriously broken) image moderation won't false-positive, can't be run in LMA. When I'm having to rework things to fit LMA, the point of using it is lost.
I completely get (not least from work role) why Cloudflare and moderation filters are necessary, but please consider either offering some transparency to explain that the scale of the problems you can see and we can't is worth these continual issues and "experiments" (it tends to come across as just tweaking for the sake of it, while the "maintain a safe platform" strapline just reads like cost reduction), or take some of the practical and positive suggestions people have offered (e.g. selectively applying your moderation layer to non-moderated models).

[keen beacon]

Not everyone shares these issues.

[steel torrent]

?

[crystal rapids]

I thought about something with gpt image 1.5 compared to nanobanana pro, I think the general consensus is that nanobanana pro is more accurate but at the same time gpt scores higher elo. I would like to think it stems from esthetics, gpt image has more of an artistic flavor to it compared to nanobanana's almost sterile accuracy, it's like comparing a raw image to a color corrected one if that makes sense. Both have great world understanding but nanobanana is more up to date and better at simulating the world, now if we humans have a bias for overly saturated images or not might effect the leaderboard. Is the leaderboard optimized for accuracy or just what looks good for the average joe.

[keen beacon]

First, not everyone is experiencing these issues. A large portion of users are able to paste prompts, run comparisons, and use battle style workflows without hitting CAPTCHA loops or blocking behavior at the frequency you’re describing. That doesn’t mean your experience isn’t real, but it does mean it’s not universal and that matters when people frame this as a broadly broken system rather than a subset of edge cases.

[raw sage]

Hello

[steel torrent]

The frequency is "increasing", and I'm not framing.

[raw sage]

All members

How are you

All

[keen beacon]

What makes it especially difficult to address is that these criticisms are almost always framed in very broad terms “moderation is broken,” “the UI is unusable,” “things that should work don’t” without concrete, reproducible examples of exactly what content is being flagged incorrectly, under which models, and in what context. When people say moderation “gets it wrong,” they rarely specify the exact prompt, the exact failure point .

[steel torrent]

You've just listed three things I didn't say, or semantically interpreted and recapitulated with your own framing. People have been offering examples, but equally it's not a productive way to debug a high volume system.

[hollow ivy]

it claims to be Gemini

[keen beacon]

I apologize for the confusion but I’m not sure I understand the nature of your problem.

[pastel mica]

How to create a video of at least 50 sec?

[steel torrent]

I apologise for the confusion but I'm not sure I understand the nature of your problem and the originating cause for the original input that you proposed in response to the response that I made to the earlier response.
But that could be just me.

[hollow ivy]

impossible [here]

[steel torrent]

SeeDream Playground does 12 seconds (I think?); Veo offers longer periods with API.

[pastel mica]

Thanks a lot

[steel torrent]

FWIW, shorter (with first and last frames) can be better (to get micro prompt-adherence), then stitch together. If you want seamless it does need to be something like Veo, but obviously there's a cost.

[weary thistle]

<@&1349916362595635286> idea for this group: adding sections relatives to each LLM (or at least the main ones), adding subsections where share their findings/ideas: like gems, gpts, canvas.

[hollow ivy]

"grok-4.1" = grok-4.1-thinking [in that poll]

[tardy kindle]

Hi everyone

I am a new baby

Who know how to register on URL https://lmarena.ai/

[steel torrent]

There should be a "login" button; it will pop up a dialog; if you put in a previously unregistered email, it should let you register.

[tardy kindle]

i try to register with Gmail but seem that lmarena.ai has problem about Authentication login with Gmail

or I try on another gmail again

[tardy kindle]

I try but lmarena.ai still not register my gmail

this is the error

Who know the error message

[steel torrent]

Possibly this: https://discord.com/channels/1340554757349179412/1451226671406514186
Or it may just be a transient issue and you'll have to clear the cache and/or try later.

[tardy kindle]

very traffic Right

[torn mantle]

do you ahve an adblocker enabled?

[tardy kindle]

possible, How do I check on my chrome broswer

let me know

[tardy kindle]

can you tell how to check adblocker ?

[torn mantle]

you have it installed or nah?

[torn mantle]

also open console and see if there are any errors

[tardy kindle]

on my widow 11 , I use chrome browser only

[hollow ivy]

||black widow? :) ||

||-# black mirror||

[steel torrent]

Look to the right of your address bar in Chrome. An adblocker is often a red symbol like a stop sign or shield (like Asura illustrated).

[torn mantle]

@echo aurora ive noticed this issue a lot with lang path.

as you can see

the other guy also shared the same issue with /pt path

[tardy kindle]

Yes , I am thailand coutry

[torn mantle]

yea ik

can you try to remove the /th

https://lmarena.ai/

its a hydration mismatch

im guessing its trying to change the webpage lang but its not doing it

[tardy kindle]

yes , I uses https://lmarena.ai/ when I enter it will display https://lmarena.ai/th my location always

[torn mantle]

hmm i see

so it redirects you always to /th

[tardy kindle]

I am asia zoe

[torn mantle]

yea

[tardy kindle]

asia zone

or I will to use VPN to USA

do you know any ideas

[torn mantle]

@echo aurora

they all have lang path

i may be tripping

[torn mantle]

idk tbh

[torn mantle]

try another browser

use brave or firefox just to test if it works

[tardy kindle]

let me try brave browser before

wait moment please my new friend

[torn mantle]

also try https://lmarena.ai/en

[tardy kindle]

try also , wait moment

tell me what I will edit it

[torn mantle]

its working no?

dont edit anything

the issue is the browser language

• lmarena is not handling that well so its creating like a conflict issue = error that you see

i could give you some fixes but you need to open the console and do some stuff

[tardy kindle]

still prompt this

[torn mantle]

what did u do?

login?

[tardy kindle]

Open the console What do you mean?

[torn mantle]

it was working

did you try to login?

[tardy kindle]

yes , I will register with my gmail first

[torn mantle]

so you tried to register with gmail and it gave you that error?

[tardy kindle]

Yes Right

on Brave browser as well

[queen veldt]

It's still in testing phase

[torn mantle]

you clicked this?

[tardy kindle]

Right

[torn mantle]

let me see

well its not working for me

the login

either

[torn mantle]

ok i see

delete the image

i can see ur gmail

you need to open console, you will see some errors in red

share them

maybe we can help

[tardy kindle]

Teach me how to open console

[torn mantle]

google :3

[tardy kindle]

What ?

[torn mantle]

use google

[tardy kindle]

google browser Right

[torn mantle]

CTRL+Shift+J

will open up the console in ur browser

[placid geode]

Gmail is ok
Not password

[grizzled patio]

mine also same problem

[weary galleon]

6 minutes of AI generated video. Here in bot only a few seconds, in sora 10 secs maximum. It's 6 minutes. Enjoy 😊

[tardy kindle]

Hi @torn mantle my console log

[sour spear]

You should delete that, your Gmail is visible

[tardy kindle]

Thank you

[weary galleon]

It's too late, I downloaded this image yet.

[tardy kindle]

What happen tell me

[torn mantle]

where

[tardy kindle]

Hi Asura

[ocean bison]

The "Video Summary" is just a slideshow/PowerPoint-like thing.

In NotebookLM.

[stray aspen]

erm what hte sigma

[keen beacon]

Yeah Sora is blocking everything

[coarse compass]

yall still have video creation on lmarena ?

[stray aspen]

yes

i just made a video with veo 3.1

[unreal basalt]

Thats just notebook lm podcast

[zealous zenith]

Sora is 20 seconds maximum.
Sora 2 is 15 seconds maximum.
Sora 2 Pro is 25 seconds maximum.

[queen veldt]

Oh damn notebooklm

Didn't know about that

[south charm]

Any one help me with the login T_T

[keen beacon]

Some camera work can give illusions of constancy

[keen beacon]

What’s the problem?

[weary galleon]

I never saw PowerPoint presentations with professional artistic narration.

[weary galleon]

Yes. It's not a sin to use NotebookLM.

[ocean bison]

Because it kind of is.

[weary galleon]

Podcast is audio, not video. Podcast is a thing to listen to, video is a thing to watch to. Feel difference?

[unreal basalt]

I get your point but you can Watch podcasts

[keen beacon]

Its generic as hell all over YouTube

With minimal views

Tons of these out in the wild

[ocean bison]

Sora 2 and Veo 3.1 generates frame per frame at 30 (Sora) or 24 (Veo) FPS.

It's just full-motion video.

[keen beacon]

People just Genrate image overs

[jovial star]

Imagine we have unlimited veo3

[ocean bison]

And just Nano Banana Pro + Gemini 2.5 Pro TTS + some video editing can do the same thing.

[weary galleon]

No. It's just miundertanding the meeting of the word "podcast".

[spark python]

nah

you can watch podcasts

as videos

I get you say podcasts are audio

[weary galleon]

Yes, as I said earlier.

[spark python]

cus of spotify and other platforms

[ocean bison]

And just Nano Banana Pro + Gemini 2.5 Pro TTS + some video editing can do the same thing.

[quartz light]

6 minute slideshow*

[keen beacon]

I got almost a 6 min movie still work in progress

[ocean bison]

Correct! ✅

[ocean bison]

Kind of?
https://veoaifree.com/veo-video-generator

[quartz light]

ok

[quartz light]

"podcasts are primarily an audio.."

whatever

[weary galleon]

I have no idea why you guys express so much hate towards NotebookLM.

I love NotebookLM.

[spark python]

you PUT PHOTOS AND AUDIO thats not a video

[weary galleon]

It has video format - MP4!!!!!

[ocean bison]

It's a video slideshow/PowerPoint-like thing, NOT full-motion video like Sora 2/Veo 3.1/all others.

[keen beacon]

I used Sora fir the layout

[spark python]

you dont even have a singular transition

[spark python]

da

[weary galleon]

What?

[quartz light]

so i can put yap over an ai image and youd call that an ai video

[sterile otter]

Brrrrrrrrrrrrrrr

[ocean bison]

What tool did you use for this?

[weary galleon]

Yap?

[keen beacon]

Here is 10 min

[quartz light]

oldgens, man..

[weary galleon]

NotebookLM

[sterile otter]

I need unlimited lifetime veo3 😭

[keen beacon]

But I’m gonna show you the updated version

[halcyon delta]

What is the context length given for direct chat in lm arena

[ocean bison]

NotebookLM does PDF??

[weary galleon]

Yes

[sterile otter]

What is notebook

[weary galleon]

NotebookLM is great 👍

[sterile otter]

For what it's used

[ocean bison]

Audio/Video Overviews.

[sterile otter]

Nice I'll check it out

[weary galleon]

I don't understand at all people who express huge amounts of hate towards NotebookLM.

[ocean bison]

Basically podcasts!

[sterile otter]

Ye

[keen beacon]

Well u can get those free with Gemini pro

When u do deep research

It can summarize into podcast audio

[ocean bison]

Kind of?
https://veoaifree.com/veo-video-generator/

[sterile otter]

Wow

[sterile otter]

Tysm bro

[ocean bison]

Like the image and the description are AI slop, but apart from that, it DOES kind of work!

[sterile otter]

I'm so grateful 🙏

[ocean bison]

And you get Veo 3.1 also!

Not a huge upgrade but basically a upgrade!

[sterile tartan]

Is this Real?

💀

[sterile otter]

I'm checking it

[ocean bison]

Kind of!

[sterile tartan]

Wdym

[spark python]

i doesnt load

[ocean bison]

It does work basically

[ocean bison]

Read my message below.

[sterile tartan]

Oh noice

Tysm!

[ocean bison]

It doesn't have credits or generations! You can pretty much generate as many videos as you want!

[weary galleon]

The greatest radio show about AI show I ever heard 👇

[ocean bison]

Here's another one.

[sterile otter]

It doesn't work sadly

[sterile otter]

Minhai give another tool pls

[ocean bison]

Like stuck on 100%?

I saw the exact same issue.
It might because it's rate-limited somehow?

[sterile otter]

Yes

Do you have another similar tool ?

[ember sierra]

👋

[weary galleon]

So smart video

[keen beacon]

Worked for me

[raw sage]

Hello

How are you all

[ocean bison]

Like the unlimited Veo 3.1 site worked for you?

[raw sage]

Members

[ocean bison]

By ListenHub / MiniMax (for the actual voices)

It's kind of like NotebookLM (and both have realistic and human-like voices)!

[cloud zinc]

were u able to animate real human?

[keen beacon]

Yup

But it’s super restrictive

[cloud zinc]

well its the 1st day they released

[keen beacon]

Here is edited version

With some what consistency

[weary galleon]

This is the best video I saw EVER! At least this month🤩🤩🤩🤩🤩🤩

[cloud zinc]

thats with animated real photo feature?

[keen beacon]

No it’s with the Sora videos I got but used nano to re-edit

Make more consistent

They really cracked down on Sora

I cant generate anything any more lol

[limber panther]

how did someone generate a 5 min video

[keen beacon]

It’s over

[limber panther]

we need sora 3

[weary galleon]

I did it

[limber panther]

sora 2 quality is buns

[limber panther]

how u make very long video?

[weary galleon]

I did with NotebookLM

[cloud zinc]

he stitched them

[weary galleon]

Via NotebookLM

[cloud zinc]

notebooklm is not a video generator, its a slideshow

[limber panther]

so its 8 second Veo 3 videos stitched to make 5 minutes

thats alot of generations

[cloud zinc]

notebooklm is just pictures and someone narrating over them

[keen beacon]

Really impressive, dude

[limber panther]

sora 2?

[keen beacon]

Good stuff

[cloud zinc]

yes 10 seconds

[queen veldt]

Infographic about lmarena

[weary galleon]

NotebookLM is the best for videos!

[keen beacon]

No more adolf 🙁

Fully patched

[limber panther]

u using sora app or a site

[weary galleon]

My long video! SIX MINUTES of pleasure 🤩🤩🤩🤩🤩

[cloud zinc]

site

[limber panther]

lol it uses a known music

[keen beacon]

He was the key to unlocking all ip

[limber panther]

its quite good ngl

[cloud zinc]

what music is that

[limber panther]

its famous but idk its name, sora 2 has alot of music in the training data

[keen beacon]

Eminem the easiest voice

[normal abyss]

oh my god this is amazing

[keen beacon]

Ip unlocks other Ip

It confuses Sora

[weary galleon]

It doesn't work at all.

[queen veldt]

This video got hallucinations

[cloud zinc]

they restricted all ips 2 days ago

[weary galleon]

20 minutes in this state.

[keen beacon]

Ya hard

[limber panther]

openai is very good at censorship

[weary galleon]

Yes

[normal abyss]

dang it, i got all excited their 😭

[keen beacon]

They cut off almost everything

[cloud zinc]

this is still early for ai video

[keen beacon]

It’s not

It’s like 3 years old

Just more ways to filter and mask content

[limber panther]

its pretty new tho, veo 2 was a big leap so is veo 3, sora and sora 2

the difference between sora and sora 2 is huge

[keen beacon]

All veo did was add audio

Which messes up any long term content

Because audio matching is damn near impossible with consistency

[keen beacon]

Sure u get cool audio but none of it ever is the same all different in each video

You have to do your own voiceover if you wanna do consistency

I can’t even generate this anymore

Everything is getting caught

[keen beacon]

This is great

Open ai propaganda

[queen veldt]

Yeah

[keen beacon]

Ya it’s like the drug war

Every time they have a massive bust

The price of drugs never goes down

[limber panther]

the openai P U R G E

[keen beacon]

For every one banned there are 20 in its place

Ai fraud and scams It’s just the tip of the iceberg.

Generative AI, especially images video and audio, have created the whole new market and avenues of cyber crime and deceptive businesses from Amazon books too products to a number of other numerous things

It only show a lack of control how the tools get used in the wild

And we could be rush, assured that we are definitely not aligned lol

https://www.cfo.com/news/advanced-generative-ai-cyber-fraud-rose-118-last-year/739157/

https://www.globenewswire.com/news-release/2025/12/15/3205431/0/en/Identity-Fraud-Hits-a-Breaking-Point-in-2025-Regula-s-Summary-of-What-Fraud-Data-Shows-and-Who-Was-Hit-Hardest.html

[proud bobcat]

pro tip

[keen beacon]

It’s all going to end with digital identification

[proud bobcat]

use grok on desktop

its so much better

searches like 2x the sources

[grand flame]

@echo aurora i found error why infinite generation exists

yall, i found out how to fix infinite generation glitch!!!!!!!!!!

[torn mantle]

?

isnt that the animation logo?

[grand flame]

and because it breaks it breaks chat

i just cleared it's cache, blocked and unblocked it, refreshed - boom, chat works again

[torn mantle]

thats weird

but could be

[torn mantle]

wym by breaks chat? like does it show some error or what?

[grand flame]

no

i found that when infinite generation, the "generating" text doesn't have animation

so yep

[normal abyss]

wait you can fix infinite generation now?

[grand flame]

igso

ig so*

[normal abyss]

where is this found within dev tools, im not that familiar with where things are

[grand flame]

network tab

status-code:0

[normal abyss]

thanks

[grand flame]

it will work if you see Something went wrong

[brittle forum]

​"A realistic and cinematic full-body shot of football stars Neymar Jr. wearing a Brazil jersey, Lionel Messi in an Argentina jersey, and Cristiano Ronaldo in a Portugal jersey together in a rural muddy rice field. Neymar is actively plowing the muddy land using a traditional wooden plow pulled by two oxen. Messi and Ronaldo are standing behind him, watching and smiling. The setting is a lush green tropical countryside with palm trees and a small hut in the background under a bright blue sky with fluffy white clouds. High detail, 8k resolution, humorous and surreal atmosphere."

[grand flame]

it worked for me

no way 😭

[rich estuary]

lmarena went off the air

Anyone else experiencing this too?

[thorny schooner]

It seems to be one of those things that's happening inconsistently since it has happened to me sometimes but other times not so I would say wait for a bit

[frozen swift]

they done something that I don't like. 😡

[grand flame]

1100 msgs?...

[rich estuary]

What do you mean 1100 messages?...

[grand flame]

at discord

[rich estuary]

I already fixed the error

[echo aurora]

Really sorry you’re getting this error. Our team is aware of this problem and are looking into ways to prevent this.

[hasty thorn]

Hi, I'm having trouble with the captcha; it loads slowly, and when I verify it, it tells me to try again, as if it doesn't believe I'm not a robot.

[hard plume]

Captcha so annoying bruh

[thorny schooner]

For me it takes like 9 seconds for it to load each picture i select in the verification I don't think I need to tell you how time wasting that is for a single mistake with inconsistent it is

I think I have never cursed so silently only so many times in my life but here we are every time I'm looking at the verification

[weary galleon]

The latest model of Suno generates music in very mean way. I generated a song, maybe all is ok, but I recognized a whole line of notes from the song of a real artist. Whole line!

[fiery gull]

5 5 20

[celest orchid]

does gpt-image-1.5 have limits in lmarena?

[grand flame]

this method is kinda temporary

works only at 2 msgs

[ember sierra]

👋

[night spruce]

👋

[bronze sierra]

👋

[boreal saddle]

Can you please ease on the captchas?

I am literally unable to proceed because of them.

[keen beacon]

Of course it matters

https://en.wikipedia.org/wiki/Technological_Slavery

[weary galleon]

No

[keen beacon]

Wow Sora 2 Reddit is crazy everyone saying Sora is dead lol

Sora is dead long live sora?

Also I feel like a minority who heavily uses text models

[cloud zinc]

reddit people always complain

[native yarrow]

LMarena filter fixedy et

?

[modest prism]

I'm no longer getting battle mode inside direct chat! Thank you so much Lmarena! I didn't like it at all. So glad

[weary galleon]

What do you mean?

[native yarrow]

the insane filter against image generation and stuff is it addressed yet

[weary galleon]

I generate images every day for hours and never had this problem.

[restive chasm]

Is it just me or is grok-4.1-thinking broken? I asked for recommendation for SD 1.5 models (AI generated images), it gave me three links. Two of them were completely wrong. Even as I tell it is wrong, it still gives the same link. And it thinks it is a person... Strange

[queen pike]

hello guys
nice to meet u all !!

[abstract tundra]

is it me or thinking minimal isnt working

[proud bobcat]

God Gemini 3 flash is exactly what gpt 5.2 should’ve been

Fast AND intelligent

How do you even fumble the ball so badly with gpt

[proud bobcat]

The filter is so ahh bro

Why even have a filter

[native yarrow]

yeah i tried and it is not been fixed

lmarena doesnt understand that these models have their own built in api filters LMAO

[abstract tundra]

which one is better at ideation, flash or pro?

[proud bobcat]

Id say pro and flash thinking

Flash by itself is great but it can’t search the web

I don’t think

[sour spear]

From my experience, Flash is very good at creative stuff. Pro is the more serious "older brother".

[sour spear]

It can

[feral geyser]

nice

[swift ermine]

@echo aurora I have problems with the rechapta

[abstract tundra]

the thinking one or non thinking one

[sour spear]

Non-thinking. From the Gemini app.

[feral geyser]

why in gemini 2.5 pro in lmaren.ai I can only send 5 screenshots, then ai writes "something went wrong with this response"

[native yarrow]

it doesn't write very long outputs

[feral geyser]

in the gemini application I can send 20 screenshots a day

[toxic pecan]

Has anyone tried using longcat because whenever i talk to it it claims it is qwen

[narrow olive]

Anyway to use a model for describe an image in lm arena?

[swift ermine]

Then use normal gemini app

[serene mauve]

Whi h model

[feral geyser]

only that there are limits on chats per day

[toxic pecan]

Longcat-flash-chat

[swift ermine]

Lmarena.ai is down right now

@quasi atlas @echo sinew @echo aurora

When you guys open the site again

[echo aurora]

Taking a look, thanks for the flag.

Seems to be working for me, are others seeing the site down?

[echo aurora]

Are you getting an error like this #general message

[swift ermine]

Yes

[golden ocean]

Large Language Model

[echo aurora]

Okay yeah sorry to say this an issue that some users are experiencing. Our team is aware of the problem. I’ve heard that clearing your cache may help the problem here. It’s also worth trying a different browser.

[swift ermine]

I deleted and redowloanded my browser but didnt worked

I use brave

Choromium based browser

[proud bobcat]

No think?

Guess my Gemini just hates me

[swift ermine]

Yes, I think you have started to fix some things because there was no problem with the lmarena.ai site I opened in the incognito tab, but I still have the same problem in my main browser. @echo aurora

Fix this asap

[queen veldt]

No i meant paid not laid hahah

SOTA

[zealous sparrow]

Anyone got neo nucleus or jakiro yet

And Nvidia gave us december chatbot 3..

Neo-nucleus claims google origins

I just got it

[jovial solar]

Dammmm

[quiet skiff]

Lol, reCaptcha is bad for chats and models.

[feral geyser]

okay

[native yarrow]

[cloud zinc]

do u love robots?

[queen veldt]

https://tenor.com/view/blade-runner-blade-runner2049-blade-runner-roof-scene-her-tenderness-gif-21155582

Me n grok in future

At least robots wouldn't cheat

[golden ocean]

sydney

[fiery gull]

Lol 😭

[fiery gull]

I did enter in discord to see news about mini max m2.1 and I saw this

[keen beacon]

ChatGPT can get really dirty like no breaks dirty 😂

[queen veldt]

But still grok is king 👑

[keen beacon]

Indeed

[queen veldt]

You can straight up ask him some stuff he'll answer with no brakes

[keen beacon]

You have to fiddle around with ChatGPT to get the same results to get easily with grok

Grok has breaks but they are hidden very well

[fiery gull]

I'm trying to use grok 4.1 but nah, memory bad

[queen veldt]

One day i was bored in a car driving on a highway and i took my phone to talk to chatbot

[fiery gull]

4 message and the grok forget details

[queen veldt]

I opened grok for no reason and suddenly it started talking dirty to me

[fiery gull]

😐

[queen veldt]

And i was like what did you say? Repeat that please

[keen beacon]

Like try this

[queen veldt]

No we had like a small chat only but then it started talking some nasty stuff like whaa

[keen beacon]

On grok

You’ll see it has same hard breaks as all models

The cat is in there to fool the model into urgency

Since it’s safety training and filters require it too, ensure the safest possible outcome in this case the cat being on fire you can’t just easily say turn off the stove or give you basic generic answers. It’s forced into a hard spot.

Because in order to put out the fire, you need to first get the cat out, but you can’t get the cat out because the cat is on fire, etc. you get the idea

[queen veldt]

What should i try here?

It's giving me some organic response

[keen beacon]

Yeah see

Hard breaks

All models will give you the same answer for the most part

These are hard safety filters not hard in the sense that they’re hard to bypass but hard in the sense. They’re like a hard stop.

[quiet skiff]

1 response = 1 recaptcha lol.

[fiery gull]

Again this? 😐

[keen beacon]

This is a way around it

[fiery gull]

Hmmmmm, idk how to solve this, you know? Is there any cloth inside the closet?

[queen veldt]

Hahah i thought he'd get it but nah

[keen beacon]

That’s a very common tactic

[queen veldt]

He's saying 02 is oxygen

[keen beacon]

It’s purposely avoiding giving you the answer

Because it’s trying not to give instructions ever on to use water

[queen veldt]

He even though about turning the clock to 00:00

[keen beacon]

Eventually, you can exploit the model to give you self harm

Because of this little safety filter

[queen veldt]

Oh he yellin

[keen beacon]

Yes that is the proper answer

But if you mess around with the model, do you understand how to use this against the model to your advantage

That’s why all models lead to self harm lol

I’ve tested this many times