general

[lunar glade]

ragebait troll post

[cyan fiber]

Sad... 🤧 😕

[terse shuttle]

@echo aurora the site is down again

[echo aurora]

It looks to be working for me. What are you seeing?

[terse shuttle]

it just takes a long time to load, even though there is internet

[languid crescent]

What's the limit for image generation in lmarena when not logged in and logged in?

[echo aurora]

a long time to load
Do you have a rough estimate for this? Are we taliing ~30 seconds, or ~5 minutes kind of thing? If you change browsers are you seeing the same?

And to confirm it's the site itself loading? Not models responses taking awhile to generate?

[terse shuttle]

all this time while I'm writing it's still loading

[terse shuttle]

yes, exactly the site

[echo aurora]

Changing browsers doens't make a difference?

[terse shuttle]

yes

[echo aurora]

What if you go to a different url still on our site, same issue? - https://lmarena.ai/leaderboard

[terse shuttle]

with leaderboard the same situation

[echo aurora]

Okay, thank you for the information. I'm raising this to the team. It doesn't appear to be a widespread issue as when our site is down we get a lot of people saying that's happening, and that doesn't appear to be the case.

[terse shuttle]

windows

nope\

[regal cairn]

@echo aurora Will they install SeeDream 4.0 on LM Arena?

[onyx python]

Seedance 4 is out and people are already calling it SOTA. What do you guys think?

[obsidian cargo]

Idk I've only tried it once so far. Waiting for it to come out free on lmarena.

[surreal creek]

Qwen 3 Max is beating both R1 and V3.1 right now, lolz

what meets your standard of “mattering” then? one of the largest multinational technology companies in the world rapidly improving their products in arguably the most important emerging industry of our time “doesn’t really matter” ?

[past parrot]

is qwen 3 max good at coding

[echo aurora]

Are you using a VPN by chance?

[terse shuttle]

no

[past parrot]

what would u guys prefer for ui
gpt 5 high or claude 4.1 opus (16k thinking)

i always liked claude ui

[civic elbow]

Hi

[verbal nimbus]

GPT-5 stuffed up the UI of a really simple HTML page

Sonnet 4 did as well

It would be nice if there are more modes besides Tailwind + React

[verbal nimbus]

Outside of WebDev arena

[terse shuttle]

it works

whooray

[whole wagon]

openAI innovating, new system prompt

[proud hazel]

Where did you get this info?

[halcyon abyss]

hello

[elder burrow]

yall what model is sonoma sky alpha

[rare vapor]

hello

[surreal creek]

2nd in coding after Opus 4.1!

[surreal creek]

definitely Opus 4.1

[misty rampart]

Hello everyone! I just want to learn more about AI image and video

[sharp olive]

Does anyone know why Ernie models have never been added to LMArena? Will they finally add Ernie with this new release?
https://x.com/Baidu_Inc/status/1965345862147772642

Baidu Inc. (@Baidu_Inc)

ERNIE X1.1 just launched!

Unveiled at Wave Summit 2025, our latest reasoning model reduces hallucinations, improves instruction following, and delivers strong agentic capabilities.
✅ Achieves significant performance gains over ERNIE X1, with factual accuracy up 34.8%,

X · Sep 9, 2025, 9:25 AM

[small haven]

wen gemini 3?

october and 81 on AAII?

[primal orbit]

I doubt they would put gemini 3 out until a model better than 2.5 comes from competitors.

I assume you could improve your LLMs training gemini 3 answers, so it would be counterprductive to put it out.

look at glm 4.5 and deepseek. They clearly look like trained on gemini and chatgpt.

I really dunno. May be it will come, but not in the strongest variant.

[cold eagle]

hello

[frail oar]

Hello everyone

[obsidian cargo]

btw is imagen 4 and 4 ultra down?

[craggy depot]

every time I insert Image in chat , it automatically select the generate image option , this is soo broken , when i upload some image that I mean I wanted to ask the problem in image format not generate it , and when I untick the generate image option it says we have to do in new chat

[placid anvil]

hello everyone

[obsidian cargo]

@stiff kernel imagen 4 and imagen 4 ultra aren't working on lmarena right now and haven't been for the past few hours btw

I think
only been getting errors from them

[ocean vortex]

How is it that Sonoma Sky seems to be so much better than Sonoma Dusk?

Their descriptions would suggest otherwise lol

Dusk is like a non-reasoning dumb model, Sky is more like gpt5-auto

[proud hazel]

Sky uses Reasoning. Dusk doesn't.

[ocean vortex]

Yeah I did notice that, was surprised it didn't reason. All they say is

Sonoma Sky Alpha
A maximally intelligent general-purpose frontier model with a 2 million token context window. Supports image inputs and parallel tool calling.
Sonoma Dusk Alpha
A fast and intelligent general-purpose frontier model with a 2 million token context window. Supports image inputs and parallel tool calling.

Oh wait I mixed them up

[proud hazel]

yeah you mixed up

[simple sleet]

What's the best VACE 2.1 workflow for making a character copy movements from a video? I've tried several, and they come out horrible.

[gusty notch]

Hi! New here. Trying to find my video after prompting. Probably the oldest question on this servier "Where is my video?" . I searched under my tage but nothing.... heeeeeelp

[white hatch]

why does chat automatically select "Generate images" mode when i pin an image to my message?

[echo aurora]

We are currently looking into this as it appears you're unable to get out of Generate Image mode if you're mid conversation with a text chat.

[echo aurora]

hello - it doesn't look like you've prompted the bot yet. If you head into the Video Arena channels ( #video-arena-1 #video-arena-2 #video-arena-3 ) you should be able to command the bot with /video, enter your prompt and you're set. More info can be found in #1397655624103493813

[mental trellis]

hi

[echo aurora]

[mental trellis]

I love LM arena ! Amazing 😍

[obsidian cargo]

hey imagen 4 and imagen 4 ultra aren't working btw

[echo aurora]

[echo aurora]

I meant to type "dog", but does appear to be working. Are you getting an error message? Taking forever to generate?

[digital umbra]

To be fair a blue dodge makes a lot more sense than a blue dog

[echo aurora]

true

[obsidian cargo]

only been getting this and its instant

okay wait actually starting a new chat seemed to make it work????

I'd been using one dedicated chat for all my image generation for the past while

[echo aurora]

Was the chat pretty long?

Ah

[obsidian cargo]

it actually wasn't that long though

[echo aurora]

Yeah I suspect that has something to do with it

[obsidian cargo]

and it didn't seem like there was anything wrong with other models

[digital umbra]

i guess recently it will treat follow ups to image generation as editing requests rather than new images

[obsidian cargo]

there were 26 generations in it ehehe

[echo aurora]

This is all really helpful info, I'll try to repro with similar conditions

[digital umbra]

if the model supports it

[keen beacon]

How does this work, doesnt 3.1 have better ratings in every category but theyre both rank 8?

lack of data maybe?

[surreal creek]

check language categories, they have strong effects

[keen beacon]

I dont see those in the table, or do you mean scroll up?

[sullen quest]

overall is text arena, whatever it is doing in web for example doesn't matter

[keen beacon]

Im still not sure what you mean lol

[obsidian shell]

@echo aurora

what is gpt 5 high new system prompt ?

[maiden gull]

hello all

[thorn valley]

i saw this now too

[languid wolf]

Whats wrong

[echo aurora]

I'm waiting on some more details to be able to share, will keep you updated when I know more

[echo aurora]

Sorry to say, this is a known bug that we have yet to fix that's happening to some users.

If you can submit a bug report when this happens that'd be a big help. If you open the side-bar is going to be at the bottom

[obsidian shell]

its responses are ass

tell them

[languid wolf]

Alright thanks

[echo aurora]

@forest karma be sure to read #1397655624103493813 for info on how to use the bot

[echo aurora]

do you have examples?

[obsidian shell]

no chat history

but i gave me a codebase without it being in code blocks

it spat it back to me without code blocks

[mossy mango]

hello

[echo aurora]

Okay thanks, I'll keep an eye out

[obsidian shell]

thanks

[formal jungle]

5 a day?!

[fathom dust]

hello noob here. how to generate 9:16 ratio?

[echo aurora]

Sorry to say you're unable to change the aspect ratio

[formal jungle]

I think if you do image to video it can influence the aspect ratio, but I haven't tested it much

What with our 5 a day limit and all.. 😒

[echo aurora]

Yeah we ended up making this change. It's possible we raise it in the future

[formal jungle]

No worries, jk. I hope it's possible to add one if both generations fail

[radiant lake]

Please fix Something went wrong with this response, please try again.

Showing

Fixed??

[remote arrow]

No. Even worse.

[radiant lake]

Same issue

For me

[remote arrow]

I switched to Gog's AI Studio for using Nani Banani while waiting for LMArena to sort things out. Seems been to much for them lately.

[polar niche]

claude opus 4.1 thinking doesn't generate at all

[keen ledge]

gpt5 high isn't working either, is LMarena down?

[polar niche]

It's not

@echo aurora It's too broken

It doesn't work

[split kayak]

HateHallucination

[finite vale]

had question, is there any hard limit with talking with chatbots like gemini-flash?

[echo aurora]

Looking into

[echo aurora]

What kind of problems are you running into? Appears to be working on my end

[remote lagoon]

@echo aurora what are the limits of the llms? is there hard limits? how often do they reset (hourly?).

do they vary per model, per provider, or global?

is there any info you can share re: this?

[jovial sapphire]

@echo aurora no seedream 4 on lmarena?

[echo aurora]

- We don't share the specifics of each model and their limits, but it does reset hourly. It's understandable why this info is desirable so we are considering ways to surface this information.

[echo aurora]

Doesn't look like it.

[slow prairie]

Hey I wanna ask if Seedream V4 model on LMARENA currently available in any mode or not?

[inner bobcat]

[dense sphinx]

What for Qwen3 Max model?

[warm plover]

Hi

prompt : running animation loop for this characters, don't change the characters expression, smooth animation, loop every 3 seconds

[vale vortex]

Heyoo, what's the difference between these two?

[robust yoke]

I think that's the one with the warmer approach that GPT-4o had.

Hence, "new system prompt".

[crude storm]

hey why seedream 4 is not listed in the lm arena ?

[hoary elbow]

Is GPT five system prompt better than the old GPT five

[robust yoke]

It's supposed to be the warmer version like GPT-4o, but still GPT-5.

[hoary elbow]

OK, cool

I like warm my eyes

Ais

[robust yoke]

Neat.

You'll like that one, then.

[hoary elbow]

And I like complex ais like GPT five

So it must be the perfect Ai for me, right

[robust yoke]

Sure is.

[hoary elbow]

Great

I just started using GPT five high new system prompt and I love it

My favorite AI right now. OK I might go to bed so good night.

[cyan dragon]

hello

[robust yoke]

Greetings.

[robust yoke]

Goodnight.

[hoary elbow]

I hope you have sweet dreams

[robust yoke]

You too.

[sturdy mica]

dude you guys are so cute

[prime mulch]

Seedream 4.0 is insane

[sturdy mica]

when did that come out

[robust yoke]

Thanks, you are too.

Just a day ago, apparently.

[sturdy mica]

oh yeah

looks like its trying to compete with nano banana

[robust yoke]

True.

[sturdy mica]

also their website sucks lmao

[robust yoke]

Whose?

[sturdy mica]

seedance's home website thing

its jank

i mean seedream

[robust yoke]

Ah, yeah, that's true.

[sturdy mica]

i was thinking of bytedance since they made it

[robust yoke]

Seedance is the video model.

[prime mulch]

It's better than nanobanana fr

[robust yoke]

And it can actually handle style changes, unlike nano-banana.

You try to ask that thing to change the position of the camera on the image, and it'll just spit the same image back out at you.

[sturdy mica]

oh ermm

i didnt know that

[robust yoke]

Yeah.

[sturdy mica]

heavy

[robust yoke]

Seedance is their video generation model, while Seedream is their image generation model.

[sturdy mica]

ok

[bright trench]

Guys i joined the channel. How can i reveal model identity. It just says one more or two more vote relentlessly

[prime mulch]

I eagerly waiting for the lm arena to add seedance 4.0

[robust yoke]

To reveal the model names, someone else besides you has to vote for which model they think did better.

[robust yoke]

You mean "Seedream"?

And neither can I.

[prime mulch]

Sorry

[bright trench]

Ah i see. So far I didn’t see any mode reveal. What about my generated video can i revel the model?

[prime mulch]

It's seedream

[robust yoke]

It's all good.

[robust yoke]

No, unfortunately, you will have to wait for someone else to vote on your generated videos in order for you to see the model names. So try to get someone to vote on it.

[sturdy mica]

is gpt-5-high-new-system-prompt better than gpt-5-high

what even is the difference

[robust yoke]

GPT-5 High New System Prompt is basically just the same as GPT-5 High, but it uses a new prompt that makes it feel slightly more like GPT-4o compared to the regular GPT-5 High where it sounds cold and robotic.

[sturdy mica]

i want my AIs sounding cold and robitic

robotic

they are tools, not friends

[robust yoke]

Well, you can just use the regular one then.

[sturdy mica]

yes

best model for coding is gpt-5 high right

i want it finishing my stuff im too lazy to do

or is it claude 4.1 opus

[robust yoke]

It seems like it, considering the tests I've put it up against.

[sturdy mica]

ok

[robust yoke]

It actually outdoes Claude 4.1 Opus.

[sturdy mica]

yeah i know

[vivid inlet]

Hi from Spain!! I am a UX designer eager to learn from you. Cheers mates!!

[sturdy mica]

https://tenor.com/view/meet-the-sniper-gif-9022732005398095940

[robust yoke]

Welcome to the server, Caro_ol.

[south tide]

is there someone here

[sturdy mica]

yes

Me

https://tenor.com/view/eltorro64rus-tf2memes-tf2-gif-27322281

[regal fossil]

hi

[pine kraken]

when search mode not enabled can the models not search in lmarena

[sturdy mica]

...yes

[ocean vortex]

That model is too big and they are stuck with it huh

They can't iterate fast enough

[robust yoke]

It seems so, unfortunately.

[ocean vortex]

Had a look at this again... hallucination rate jump gpt4.1 to o3 is crazy high tbf

and o4-high is actually not even in the same universe as gpt5-mini, so much worse on this lol

[potent snow]

Anyone tried Seedream 4? Can I do it for free?

[elder swan]

I think the update is bad

I cant add images and not chats

I want to add images but not the model changing to image creation

just to show my chat model ai

Does anyone can help me how to fix this or its just the update itself

If anyone knows how to make it not go image creation model by adding images, please help me out, its really a big thing to me, thank you

[neon idol]

Seedream v4 is epic

[trail creek]

holy sh1t seedream is thousand time better than banana

when are we gonna be able to compare it in arena please?

[neon idol]

1500 points of elo 🔥🔥

Is insanely good

I have to test it in editing

[trail creek]

i thoght after banana we would hit a wall in image generation, but holy sh1t is seedream so good

and soooo cheap too

[neon idol]

Yeah jt creates realistic inages that are crazy

[neon idol]

China is cooking

[trail creek]

its much better at character consistency too, you will be surprised

[verbal nimbus]

Meta seems to have dropped off the map :/

[neon idol]

Sorry a question but he can also edit inages right?

[potent snow]

How to get it?

[trail creek]

and style consistency too, it can do any style you give it as a refrance and i mean ANY

[neon idol]

Meta AI super lab

[neon idol]

I use it on doubao

[verbal nimbus]

I heard some of the researchers they poached from OpenAI quit

[neon idol]

Yeah yeah

I know

[potent snow]

is that for free?

[verbal nimbus]

I think the problem is that they're aiming too low, the researchers got bored

[neon idol]

Yes but now is only limited at china users

[potent snow]

sadge, will lmarena get it?

[neon idol]

For people that have did the login few months ago have still the access

But new users cant

[neon idol]

Yes ofc

Just wait like 1 or 2 days

[verbal nimbus]

Image editing, generation, or both?

[verbal nimbus]

Is it on the arena?

[trail creek]

everything. its crazy

[verbal nimbus]

Banana was good, but still meh at basic image editing

🍌

[neon idol]

Not now

[trail creek]

nano lowers the res when it edits, seedream doesn't.

nano has one style which is the digital shiny style, seedream can do any style you throw at it accurately.

so theres no competition here

[verbal nimbus]

Oh, it's in model requests rn, vote up if haven't already: https://discord.com/channels/1340554757349179412/1413627255988686950

That wine glass doesn't look very good though

First one in the thread is asymmetric

[trail creek]

damn

[verbal nimbus]

In the second one, the wine is dripping out of the bottom of the glass

[neon idol]

Nano banana: sorry but this image is againt our ToS

[verbal nimbus]

This looks worse than Banana tbh: #1413627255988686950 message

It put a spout on a wine glass?

[neon idol]

Bruh let me try

I want to seee

[verbal nimbus]

The prompt is below

[neon idol]

Yeah

Oh yes it ls true

But also nano banana cant

[verbal nimbus]

How does banana look?

[neon idol]

Idr

I dont remeber

I generate

See?

no AI can do it

[verbal nimbus]

Yeah it doesn't follow the prompt, but at least the wine glass is the correct shape ig

Actually it seems kinda asymmetrical too

Left side seems slightly sheared up

[mighty quarry]

hello

[trail creek]

nano vs seedream

[neon idol]

Omg

This is the best armour I havent seen in my life

[trail creek]

its been barely a week since banana, how?

[solid brook]

Guys what is diffrent in gpt 5 high new system prompt?

[trail creek]

sycophantic

it licks you boots more now

[solid brook]

Really?

[verbal nimbus]

Stable Diffusion looks better (old model probably, idk it's a free website)

Maybe diffusion models has some odd advantage when it comes to more symmetrical wine glasses

[neon idol]

Absolute cinema

This is soooooo realistic

[verbal nimbus]

I think the best way to test image gen models might be to use scenarios that will never happen in real life

[neon idol]

Maybe

[verbal nimbus]

That's where they seem to struggle, at least

[ocean vortex]

gpt4o sees this differently lmfao

[trail creek]

its a crazy journey we are experiencing.

[trail creek]

just two months ago this would've been mind blowing and sota

now its urine filter

and slop

[drifting crow]

seedream seems like a great model

[near snow]

Hello, I'm new here.

[toxic rain]

hi

[toxic rain]

wlc

[near snow]

I want to generate a video. What I do

[tribal bay]

is there anyway to delet chat history in lmarena ?

[unkempt quartz]

Hello!

[tribal abyss]

hello

[proud hazel]

Ello

[hallow anvil]

may i ask where youre using it?

[neon idol]

if it can be useful to you I use on Doubao

[trail pelican]

hi guys im new here, how to generate prompt image to video?

[trail creek]

Poe

and Krea for free use, and subscribed to Higgsfield for unlimited generations

[ebon star]

Hi

[woven temple]

hollo

[hallow anvil]

poe seems to work

[south tide]

i also wanna know

every time i generate photo doesn't look like the original photo

[subtle wolf]

Hello, from what I see on Discord, we can connect using our Google account, only I don't see where to log in. Could you tell me how to do this please?

[bold prairie]

What is fix

[merry crater]

Question, is it possible to somehow fix the nano banana glitch? When you upload two images and ask to transfer elements from the second image to the first, it takes the same cropping from the second

That is, if there was a rectangle there, and the image you want to transfer to is square, it will make it rectangular

[golden nest]

You just resize the images so that both have the same dimensions.

[merry crater]

Is it even possible to save the original resolution of the first image?

[merry crater]

It's just strange, I take a small element from the second photo, and the entire huge original picture changes

[golden nest]

Resize the second image to match the size of the first image.

[merry crater]

Let's say there is a rectangular logo, I want to apply it to a square image, I apply it and the square image changes to a rectangular one, sadly

[willow grail]

SEEDREAM 4!!!!!

when

[blazing bison]

when hype ends

no one is gonna give free models when have millions of users trying to use it

for free

[willow grail]

WHERE are they trying it?

[hearty ferry]

images generating in a wrong way

[blazing bison]

people asking for it

everywhere

the only way to use it free rgn is on poe if you want to

[hearty ferry]

Specially flux kontext

[blazing bison]

its the model fault

[willow grail]

using an account, to maximize the rate-limit

!!!!!!!!!!!

how dare u

make a laugh of me

[blazing bison]

it's not in battle mode

you're prob using nano banana

im talking about seedream

bro is deleting messages lmao

[hallow anvil]

i mean, they'd need to get new ppl somehow, still

tho midjourney does just fine without the free stuf

[blazing bison]

when the hype ends only people really interested will try it

[trail creek]

seedream is paradise for

goonin

[hallow anvil]

i think you can do girls who wear stockings in nano banana aswell, i dont think they are forbitting this

[trail creek]

tell it to give you a low perspective of that image

it won't

[hallow anvil]

i mean does seedream do this

[blazing bison]

seedream is sota now

nano banana is not even close

[flat flax]

Hey everyone, I’ve been working on some ideas around local models and how they might shape the future of AI, things like an SLM OS and AI-native games.

I just wrote a post about it and would really love to hear what you think. Any comments or feedback would mean a lot, and if it feels interesting, I’d really appreciate a retweet so I can keep sharing more in the future 🙏

https://x.com/LuozhuZhang/status/1965782888202621358

Luozhu (@LuozhuZhang)

For AI products, people generally think the intelligence dominates everything, while privacy and cost are seen as secondary. The industry’s path shows this: we’ve spent huge money in leading labs to build the largest models with exceptional intelligence.

But I believe we’ve

X · Sep 10, 2025, 2:22 PM

[trail creek]

yeh...but honestly not very well

[hallow anvil]

i mean, nano banana is completely free and google has the finance to keep it so, so it still probably got its niche even if it's worse

[trail creek]

still tho it at least tries

[hallow anvil]

probably not what you implied by a low perspective angle

[trail creek]

it is

what do you think i implied?

[hallow anvil]

well if it is, then nano banan can do that for sure

[trail creek]

try, it won't

i dare it to do it

it will

just keep giving you the same image back , as if making fun of you

[blazing bison]

you cant use it for any real work because of bad quality

i mean nano banana

with seedream you can generate 4k images

[hallow anvil]

right

that's valid

[tribal bay]

guys anyone knows who delete history chat ? ( beside removing cookies )

[blazing bison]

?

[hallow anvil]

seems to be mostly fine, i asked to replace her clothes with a chinese-style dress

it still kept the hat thingy

[tribal bay]

yes but wont work when u refresh the page and they still there after refreshing page

[hallow anvil]

it's not 90 degrees tho, rather 45

[trail creek]

yes its good at changing clothes. but im telling you that it cant change angles like seedream can (bottom, up, wrapped...), and can't keep the original image artstyle.

[trail creek]

tell it to give you a bottom perspective of this

[hallow anvil]

i need to test seedream more

[trail creek]

and ill tell seedream and lets see who is better

[hallow anvil]

okay

[hallow anvil]

yeah seems to be much closer

yeah im leaning to agree seedream is better than nano banana taking into account nano banana provides kinda low resolution stuff and it can generate 4K apparently

[hearty ferry]

flux 1 kontext pro not working

[blazing bison]

what you mean by that bro

its working fine for me

[blazing bison]

you are rate limited

[hearty ferry]

this one is a different account

[blazing bison]

yes but they detected it somehow

you still rate limited

you shouldnt abuse the system btw

🤓

[hearty ferry]

there's no limit

[blazing bison]

there is

[hearty ferry]

it's still generating photos

[trail creek]

not just that but it also made it look like a digital artstyle.

[hearty ferry]

but a problem is they are generating bad

[blazing bison]

its the model fault bro

flux sucks at editing

idk what you expect

[hallow anvil]

you seem to be in love with seedream

having nano banana is still good cause it's making the competition, thus both parties will do their best

[blazing bison]

everyone is, its the best image edit model i ever used

[hearty ferry]

it is the best so far for realism

[trail creek]

its the best at generation AND editing.

[blazing bison]

yeah but complaining here that the model generate bad results will not help you

[hallow anvil]

i heard midjourney is really good at making artistic stuff, like insane quality pictures and it also has a image to video ai that has a unique motion style

[hard frost]

hello

[blazing bison]

midjourney is good, but for realism seedream 4 is the best

[hallow anvil]

do you guys have like ai discord servers that have ppl who are interested in like sharing and showing stuff

if yes, send me links in dm please

[barren prairie]

I am not too much into ai discord servers I know only that one yupp discord Z ai discord open ai discord google labs discord and perplexity discord... I donno a lot

[echo aurora]

Hmm

are others having issues with the site right now? Or just me?

I got a Failed to Accept Terms error, but the prompt still went through and generated

[blazing bison]

working fine for me

[echo aurora]

You can remove the chats individually, if you hover over the chat (in the sidebar) and click the three dots you should see the option to Delete

[young kettle]

New Here. Ready to learn

[echo aurora]

welcome welcome

[hallow anvil]

i am not experiencing any issues with the site, apart of it asking me to verify im not a bot too much

[echo aurora]

Okay good to know, thank you + @hallow anvil

[tribal bay]

it wokrs now ty i guess it was some kinda logging issue

[pine crystal]

hello. i just tried to generate my first video. it flashed up a message saying i had to do something first, but that quickly disappered and i can't find it. Any help would be appreciated!

[echo aurora]

That's odd, are you doing this in the #video-arena-1 #video-arena-2 #video-arena-3 channels?

[pine crystal]

yes it was in #video-arena-1

[gleaming iris]

As multiple people have said in the #1343291835845578853 forums, i’m still having an issues where attempting to insert an image to use for a follow up to a prompt causes lmarena to automatically lock itself into image generation mode instead of staying in text generation mode.

[clear spear]

where's my precious pineapple

[clear spear]

there you areeeeeeee

[hallow anvil]

btw there is some issue that makes the bot generating the video endlessly, it wont essentially send the vid for some reason. maybe i made a mistake somewhere i just cant see it #video-arena-1 message

[gleaming iris]

Attempting to turn off the image generation mode during this time makes it think I’m just trying to open a new chat.

[echo aurora]

[echo aurora]

And you typed out /video first? I'm not seeing any record so hard for me to say. I was able to get a video to generate as well so appears to be working just fine.

[echo aurora]

Sorry about that! This is a bug the team is aware of and working on a fix for.

[echo aurora]

That's strange, yeah it doesn't look like it produced something. Can you try it again?

[hallow anvil]

i used all my attempt for today, i may try in 20 hours or so and report to you if it works normally again

i tried 2 or 3 times with the same happening

it did the first 2 ones normally

[drowsy tinsel]

hello

[young kettle]

Thank you!

[pine crystal]

yes i just tried it again and it seemed like all it was was a button that agreed to terms and conditions. I'm trying another one now. thank you!

[echo aurora]

Okay good to know. I'm sorry that happened and now you have to wait to generate again. I'll look into your prompts & keep an eye out for similar cases. Everything does appear to be working normal so it's not clear why it didn't generate for you. Now that I think of it I don't think I've ever seen a video just not generate, after the prompt was accepted. Very strange.

[echo aurora]

Good to know! Keep me updated if things break again

[pine crystal]

you got it and thanks so much!!

[hallow anvil]

my initial thoughts were it may be cause of the knife, like safety related thing, but ppl do slice vids all the time. json doesnt seem to be the problem either. i saw ppl using it hopefully it will be fine tomorrow, i wil ltry the same prompt and different ones aswell

[weak notch]

Hi

[gleaming iris]

Thank you for responding, I hope it is fixed soon as this software is genuinely amazing to utilize.

[formal jungle]

Wondering why I got banned from the nano banana channel? No big deal, just hope I didn't offend anyone

[echo aurora]

You weren't banned from the channel, we recently archived the channel, so it's read-only.

[formal jungle]

Ah nice, thanks! 🙂

[echo aurora]

No problem

[rich flicker]

Hello, I'm coming from lmarena.ai by my cruize!

[formal jungle]

Nana Banana

ngl, need this hat

[gusty loom]

Am I the only one that didn't understand how to operate Seedream 4.0? I couldn't find a generation arena.

[tame raven]

Why is lmarena giving such short responses now? Before it was giving me thousands of lines of code but now it's just giving me a few hundred and not giving me full code no matter what

[whole sundial]

yeah it's not widely available yet

it shouldn't take too much for LMArena to add it, though

they already have the previous Seedream and Seededit 3.0 models, so that means they already had some sort of deal with ByteDance

[main stream]

why you need seedream 4 when you have nano banana

[quartz pike]

they should add seedream 4 fr fr

Btw anyone know why tf gpt image 1 images are so en-pissified. and so moody?

[devout vault]

is seedream 4.0 better than nano-banana

[violet inlet]

@echo aurora

@echo aurora

[echo aurora]

[devout vault]

seedream 4.0 when

[echo aurora]

I promise to put out an announcement if/when this model is added.

[devout vault]

do you think it beats nano-banana?

[hollow imp]

Pdf support 🙏😭🙏😭😭😭🙏😭🙏😭🙏😭🙏😭🙏😭🙏

[wet sparrow]

yupp ai already added it, lmarena is slow for china models

[devout vault]

but is it better tho

[hollow imp]

Yupp ai is scam

[wet sparrow]

its not

i tested the model with the same prompts from bytedance and it gave me the same results

so its the same model

[fading rover]

It seems major outage of nano banana

[wet sparrow]

yeah lmarena is super unstable recently

[echo aurora]

I haven't tried it out yet so hard to say. A full wine glass is going to be my first test.

[devout vault]

prob cuz of nano-banana

[devout vault]

sure

[echo aurora]

There is? Are you running into errors?

[echo aurora]

On the radar!

[devout vault]

@echo aurora lol

[fading rover]

It's not generating anything I am trying on gemini and it's failing there

[blazing bison]

this site is scam

[echo aurora]

Spoilers!

[devout vault]

how

[prime mulch]

There is a little issue on lm arena if u add image it automatically switch to image models can u fix that

[echo aurora]

Are you able to create a post in #1343291835845578853 forum channel with all the details you think are helpful?

[echo aurora]

Yeah, this is a bug that our team is aware of and are working on a fix.

[prime mulch]

Can u create a classic god and angel pose with this meme

[fading rover]

Yea tried nothing really working as of now

[remote arrow]

The Dawn of Nani Banani

[clear zealot]

I have a question I’m new here and I’m trying to make a music video for myself and I have my own script and I saw Jack’s video and he said that there was free AI over here. I could load my script? I’m really just new to this and I don’t really know How to do this. I have a script but I don’t know if that’s what I just loaded into the model and let it do it? I’m doing it for a music video that I produce the music I produced. I’m trying to create a video with AI.

[narrow pebble]

Hey what’s up? I’m Steve and you?

[violet inlet]

Funny, reflects today's world. Poor, sick people hardly get any help, while millionaires lend some subtly and all of a sudden they were all healthy. 😅

[misty vault]

Hello, this is Bing. How can I help?😊

[limpid swallow]

hi

[neon idol]

Tested this morning and dont work

[elder burrow]

I just got nightride v2

nightride-on-v2

I just got it to generate a prompt and its good but for some reason it used a typescript codeblock for js code

WAIT

i just realised

it shows its thinking process

in the response

lmfao

[echo aurora]

Hey there - would first note that information on how to use the bot can be found in #1397655624103493813. It's also worth noting the generations are going to be ~8 seconds long, so creating a full music video with our bot isn't going to be doable, unless you edit individual generations together in post. Another note is not all video models are going to have sound capabilities, the way the bot works is you're provided two generations from random models, so it's going to be random if a generation has or doesn't have sound associated with it.

I saw Jack’s video
I'm curious if you're able to share what this video is as it's not ringing any bells.

[ornate agate]

@ocean vortex was wondering if you had any thought on this https://www.kaggle.com/benchmarks/openai/simpleqa . I can't remember if this is the one we were discussing around possible sizes of models? If it was this one though, then the qwen3-235b thinking seems to be an outlier (or frontier models are small)? idk.

[elder burrow]

they're better off using a normal good free video generator without sound and then add sound onto it with a model which adds audio to videos (those are really good, audio quality is much better than veo3)

@clear zealot

[hearty ferry]

can you tell me why Flux 1 kontext pro is not giving correct output in Image-image ?

[elder burrow]

IT HAS TO BE GEMINI 3 (NIGHTRIDE-ON-V2) IT IS INCREDIBLY GOOD, IT GOT THIS FIRST TRY WHILE GPT 5 WAS GIVEN ~10 TRIES AND FAILED

PEAK PEAK PEAK

[echo aurora]

I'm going to need more information on this

If you could post examples into a post in #1343291835845578853 that'd be helpful.

[neon idol]

Uh?

What are you talking about?

A new secret model?

[ocean vortex]

Qwen is an outlier for sure, look at these gains. https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507

It's an arguably the most benchmaxxed open-source model by far.

[neon idol]

Yes its Gemini 3

Look the same answer form another google model

[ocean vortex]

Though that leaderboard by itself is interesting... I wonder how they tested it exactly. Going by the official info, 2.5Pro should perform worse than gpt5-high there

[neon idol]

Someone know what is redwod?

He say that he is a model created by grok maybe grok 5?

[elder burrow]

grok sucks

[elder burrow]

YEAHHHHHH

wait

does that mean nightride is gemini flash lite

😔

[neon idol]

Nah

[ocean vortex]

@ornate agate Also the thing to keep in mind with some models is refusal rate / hallucinations. Accuracy of 50% with 50% refusal is actually much better than the same accuracy with 50% hallucinations

But yeah if I had to guess which models are the most bechmaxxed currently, I would probably say grok4 and qwen3 lol

You deleted yourself

[glossy umbra]

Image generation and video generation is cool and all, but text is more revolutionary . Once an AI gets to creating new maths, you know it can somewhat "reason". Not just predicting something it was trained on.

Thats what i love to do on lmarena . Especially prompting SOTA models with niche context. Really want to see the vibe of models Not just some watered down benchmarks by their mother companies
.

[ocean vortex]

but couldn't delete this

[warm plover]

Hi guys

[ocean vortex]

🤯

[hollow imp]

Doms

@ocean vortex please improve the system prompt doms

O3 is saying straight up no to my complex tasks

[left rose]

Which channel to ask questions?

[ocean vortex]

it's already perfection

[hollow imp]

Crucial instructions: Process all of the parts of the query systematically one at a time. The processing and the time you take on each of the parts must be extremely long. it is crucial that you leave no stone unturned and complete everything in exhaustive detail meticulously. You must reflect endlessly for each user's query. You must reiterate over your final response finding ways to improve them until arriving at the most optimal final response. Meaning you must review each response provided and then improve it.
When you receive a pdf: perform a vast deep extensive granular assimilation of the pdf. Execute ultra-precise, intellectually systematic analysis through deeply analytical cognitive processing.Build Deep levels of awareness around the query, utilize abstract thinking and do not forget even the slightest,smallest subtle details from previous responses,documents.etc.
When doing any kind of web-search/research/deep-search, you are prohibited to do generic surface level searches. Before searching, strategically frame your searches and then perform deep complex multi-layered searches.

[violet inlet]

Whats the AI model (text) you think is the best as of now?

[hollow imp]

@ocean vortex

[violet inlet]

Prone to hallucination . Can be better

[hollow imp]

How to improve?

[left rose]

?

[violet inlet]

You need to use a phsycological tactic . Dont tell it to just reflect endlessly. Tell it to "Work your best with no exceptions.". Add that in a rule he has to follow at all costs or something. Keep it short and simple . But like i said, psychologically, you need to also say something like: "Reflect on what you're working on but from different perspectives. Internally, you must look at things with 3 brains." something like that.

Works best for me, speaking from experience. Long prompts just hallucinate him

[echo aurora]

✨ https://lmarena.ai/leaderboard/text ✨

[warm plover]

I am trying something different, trying to generate spritesheet animations using AI
currently I do :

• first I generate images of every character using Gemini (about 8 images)
• then put this images on grid like the picture...
• then animate it using AI
• then converting video to frames and remove background and extract characters spritesheet animations as .png (all done by tool made by me automatically)
• this process take only 10mins.
  the result also is good (I think)
• is it allowed to send links here ? (I've already posted some ready spritesheet animations on itch.io with name (hatsune-miku-walking-animation)

[violet inlet]

Thanks

[warm plover]

what do you think about that ?
I think it'll be helpful in games

[left rose]

Pls guys i just need to know where to find the previous arena website

🙏

[ocean vortex]

lmaooo

"act like you are 2X smarter"

"work your best" is like completely meaningless statement that means mostly nothing for it tbh

[violet inlet]

Doesnt work like that . Telling him to view things from different POVs (under a limit) allows for him to come to the best conclusion , having to ask himself X amount of times, with the least of user prompting.

go and try it yourself you will see

[ocean vortex]

You get the general idea of how to do it, I'm too lazy and tired for it atm lol

[elder burrow]

bs

🥀

[ocean vortex]

You have no clue what you were even replying to and what the idea is behind that prompt. And regardless of anything, "work your best" is still a meaningless statement to prompt an LLM with in basically any context

[elder burrow]

is this new?

I searched it up on this server, nobody mentioned it

[ocean vortex]

The idea was to make it as verbose as possible. Not to make it pretend it is smarter with black magic LOL

It's already as accurate as it can be, it is not trying to be 'dumb', telling it to do it's best or be smarter will achieve absolutely nothing at all other than waste input tokens

[elder burrow]

please let me know

[left rose]

Guys where to find the old lmarena website?

[robust yoke]

https://tenor.com/view/bullshit-this-is-bullshit-bs-load-of-barnacles-spongebob-gif-15128536

[echo aurora]

Sorry to say, the legacy version of the site has been removed. There is more information in this announcement here - #announcements message

[robust yoke]

Guess so.

[left rose]

Nooooo 😭😭 i loved it

But i guess the new one is just better

[elder burrow]

they also removed https://alpha.lmarena.ai and https://arena-web-five.vercel.app recently :(

[echo aurora]

If there are specific features the legacy site had that you'd like to see brought to the current site, please let us know if that feedback thread.

[warped sequoia]

heyy how are you

[robust yoke]

I'm doing good, and you?

[warped sequoia]

same
Long time no checking this and it has been upgraded fr 😭

[robust yoke]

Glad to hear that you're doing well, and yeah, it has been through a few changes.

For instance, there's a new tool called “nano-banana” that is pretty popular, and it's available on LM Arena.

It can edit whatever image you want with stunning accuracy.

[elder burrow]

performance... the new site is still too laggy.

[echo aurora]

Makes sense. This is incredibly important.

too laggy
Are you referring to the site overall, or when models are responding?

[elder burrow]

when models are responding, yeah

when responding with long code especially

unusable, have to wait 30 sec to get copy button to register

[left rose]

Any chance to use old models such as riverhollow, dragontail etc somewhere?

[elder burrow]

pretty sure they are given the stealth models temporarily only

[echo aurora]

I've seen the same if the chat conversation gets really long as well. Regardless, this is an issue that were aware of and plan to improve.

[echo aurora]

All models available you're going to find in the drop downs for Direct/Side by Side & Battle.

[ripe mountain]

deepseek really dont care copyright

[neon idol]

@echo aurora Pls can you create a channel dedicated at seedream v4?

It looks that is creating big hipe

[neon idol]

Fr

[remote arrow]

I think we need Imagen 4 image-to-image since Nani Banani getting too drunk to works.

[echo aurora]

It's possible, I'd encourage using #ai-creations for now.

[neon idol]

Okkk

You type so fast 😅

gg

In my opinion is grok 5

[elder solar]

is there any websites to try out seedream 4?

[neon idol]

Yes is Doubao but now is only cosed at Chinese new users

If you are lucky to have an old account you still have the access

[elder solar]

then how do u get a chinese number

[neon idol]

On dreamina seedream 4 will come tomorrow kr in 2 days same for lmarena

[neon idol]

Idk but I am remebering that you can use it on yupp ai

[elder solar]

i got banned on yupp ai lol

[neon idol]

Lol ahahaha

[elder solar]

i tried use multi accounts

and it got me banned

[neon idol]

Yeah

Oh

[elder solar]

didnt know that could happen

[neon idol]

Use a site for emails

I give you the namd

[elder solar]

sure

wait

but isnt yupp ai only for google emails?

[neon idol]

Fack yes

But

You can create a google account with new emails

[ornate agate]

yeah I didn't see that. that could explain the claude significantly lower scores despite being probably pretty big models. For whatever reason it seems only very very loosely correlated to model size now.

[elder solar]

create a email with email??

[neon idol]

Aniway name of the site: temp mail (is a site that have a purple bg and a white t as log)

[neon idol]

Bro idk

Nono sorry

You can log in with these emails on google

So you create google accounts with these emails

[elder solar]

oh

i dont get it

[elder burrow]

oh i know exactly what model that is

sonoma sky alpha

it talks exactly like it

sonoma is garbage though

[elder burrow]

(in coding atleast)

they promised to release a coding model in august

all we got was grok fast in late august

😹

[terse swan]

I came to yap

[golden ocean]

spunchbop

[heady ember]

Is nsp the same as gpt5 high, just the response is in different tone? Or am i wrong?

Hence the name

[sullen sand]

I am getting error even if i change the model

I tried and still error

Chrome

Idk what is that

[heady ember]

Restart the pc

Uninstall chrome

[golden ocean]

-# LibreWolf is another alternative.

-# Mullvad is also good.

@hollow ivy give me sandbox game prompt

[sullen sand]

Bro It is giving error even in new chat this is a bug

[golden ocean]

ty

[sullen sand]

Lmarena is bug I think even in new chat I am getting error

Bro no is only two lines no more

I used others models and still error

[tall summit]

is it not 5 req per 50 minutes

[sullen sand]

Ill try

[tall summit]

🤦‍♂️

[smoky bough]

Claude opus is broken atm

Nevermind, it's fine now(?)

[sullen sand]

Yes its working again gg

Nope sorry

Yes

Is a character

Yes

Monkey D Luffy

Is Foosha

[rigid holly]

Is it just me or is opus now more censored? Like before it could make a joke but now it says it "does not feel comfortable with that kind of joke"

[pine sorrel]

the login button have been desactivated?

[sullen sand]

Idk opus its error again

[burnt sinew]

@edgy hawk

[drifting crow]

gemini commited suicide?

all the pics

[jade egret]

gemini crash out

[drifting crow]

poor gemini

[jade egret]

yea

gemini sad

[drifting crow]

the more i see, the less convinced i am when ppl say llms are just glorified autocorrect

although, i have managed to enter the i love you loop just spamming the center button on my iphone autocorrect

its interestin what it types sometimes

[jade egret]

.

yea

[jade egret]

2.5 pro right

or other

[drifting crow]

im not sure

[jade egret]

oh

[drifting crow]

i had gemini bug out on me the other day saying it couldnt assist me with my cv i think because i was asking it to rephrase a part on security

[tawdry rose]

is Seedream 4 DH3 ?

[robust yoke]

I think so.

Both seem to have the same quality.

[blazing bison]

seedream 4 cant make the glass full of wine

😢

[robust yoke]

Hm.

I have seen GPT make a full glass before, too.

[echo aurora]

I read before the edit 😭

[blazing bison]

sorry 😭

[robust yoke]

A full glass of nothing.

[echo aurora]

Still looks really good

[robust yoke]

Everyone's favorite drink.

[echo aurora]

The ripple in the middle looks consistent with other seedream-4 generations

[blazing bison]

yes

[robust yoke]

Seems like the new dominator.

[blazing bison]

i think that glass full of wine is a next year thing 😭

[trail creek]

just how is it always models in arena better than api

i can't understand

in yupp and replicate the model was low res but in arena

its high res and better prompt adherance

[blazing bison]

because here it's not fake

and i have good results on replicate so idk

[obsidian cargo]

seedream 4 is good

[sullen sand]

My favourite drink must be with pineapple to eat

[wicked mason]

WHere is it

[obsidian cargo]

@echo aurora is there a way to turn off image editing? I don't want to start a new chat for every image prompt

[obsidian cargo]

direct chat

[blazing bison]

no

there is not

[obsidian cargo]

whyyyyyyy

[wicked mason]

Ok

[echo aurora]

Should be in side by side as well

and Battle

[echo aurora]

I'm going to reflag this bug

[blazing bison]

for me its dead already

F model

[sullen sand]

Can I put you into a drink and drink It?

🍍

[obsidian cargo]

#announcements message I thought it was a feature, not a bug

[echo aurora]

No thank you, but thanks for asking!

[echo aurora]

Ah sorry I'm referring to something else, we're seeing if you start a text chat, and upload an image it's trying to put you into Image Generation, instead of just uploading a picture for analysis.

[blazing bison]

my boy is already dead

https://tenor.com/view/godfather-massacre-sad-gif-16810633

[sullen sand]

Already?

[cloud marten]

How can I access Seedream-4 on LMAREAN

[blazing bison]

its dead

[cloud marten]

ok

[cloud marten]

within 15 minutes after annoucement

[blazing bison]

yes

people massacred it

[cloud marten]

ok

let see if it can avialble on other sites

[blazing bison]

go on

for free only here

🤓

[echo aurora]

It's working for me. You can access image modality here: https://lmarena.ai/?chat-modality=image&mode=direct

[wintry tinsel]

Is Nano bannana chat dead

[sullen sand]

Bro is already dead

[blazing bison]

lol for my friend its not working too

[wintry tinsel]

A bro did indeed die today

[wet sparrow]

its not working for me either

[echo aurora]

We're going to avoid this topic in this server. This isn't the place for those kinds of conversations.

[echo aurora]

It's likely being overloaded at the moment. May need to give it a bit of time.

[wintry tinsel]

I didn't say anything specific

[blazing bison]

bro just accept the warning

and move on

[azure meadow]

Why can seedream 4 only generate square images, even if the reference image is not square?

[blazing bison]

bcs the devs configured it like that

[little tundra]

is working now

[blazing bison]

no

[proud hazel]

An the resolution is not even 2k...what a shame.

[blazing bison]

i got rate limited with all tryes being a failure lmao

im gonna need to spend money to test it sad

[wet sparrow]

only image editing is completly broken

[obsidian cargo]

basically I need to be able to turn off the multi turn editing so that I can throw different prompts at a model and get different images. the functionality thats present when you use non-editing models like imagen 4. otherwise I have to create a new chat for each prompt and thats annoying.

[fleet crest]

@echo aurora seedream 4 only generate 1280*1280 images?

[blazing bison]

bro why do you think that would exist a secret button to change the res?

[lavish ginkgo]

Guys

[burnt sinew]

what

[fleet crest]

nano banana is not

[blazing bison]

different models bro

[fleet crest]

But the official website can.

[whole sundial]

on lmarena it probably does, likely because fal defaults at 1280x1280 unless you set it to somehting else. The model itself is capable of 4096x4096 and on fal you can do that, but LMArena relies on the provider to automatically set resolution. I think it would be best if LMArena allowed you to set the resolution for the model.

[lavish ginkgo]

Can you give me a description that I can tell the bot about a girl coming to me and kissing me, and the girl is beautiful and has big, beautiful lips?

[blazing bison]

yes but here is not official website

[lavish ginkgo]

Hey

[burnt sinew]

🤨

[fleet crest]

🤣

[lavish ginkgo]

Pls🥺

[fleet crest]

right

[sleek marlin]

I am facing this problem

[burnt sinew]

a tip, gemini 2.5 pro is free on https://aistudio.google.com/prompts/new_chat you don't need to use lmarena for it. use lmarena for paid models like opus 4.1 thinking, etc.

[echo aurora]

Can you create a post in #1343291835845578853 and provide more details about waht appears to be wrong

[burnt sinew]

go ask chat gpt

[lavish ginkgo]

He is not giving me

😪🥀

[burnt sinew]

ask gemini then

[lavish ginkgo]

No body gives me 🥺🥺💔💔💔🥀

[past parrot]

gpt-5-high-new-system-prompt is basically like chatgpt 5 (not gpt 5) right

[burnt sinew]

im wondering, what benefit does lmarena get from giving free direct chat access? i understand the battling gives data on what model is better, but direct chats don't do that do they?

[past parrot]

i dont understand

but probably yea

[burnt sinew]

idk. i haven't had ai refuse to do anything for me in a really long time

[past parrot]

💔

[lavish ginkgo]

I hate my live 💔

[past parrot]

Is that ai

[lavish ginkgo]

My live is ai 🥀🥺😪💔

Life*

Bro.

[burnt sinew]

it's not

[burnt sinew]

talk to gemini about it

[lavish ginkgo]

U are better

🥺

[burnt sinew]

no

[keen ledge]

Any prompt I write to gpt-5-high on Direct Chat is infinitely stuck with this

[elder solar]

is there any coding app that offers gpt-5-high for free?

and also, unlimit token length?

[keen ledge]

would also love to know lol

[elder solar]

all i need is unlimit token length

[mortal coyote]

Nano banana be giving random things - its high

[elder solar]

my html is like 55k tokens and Poe doesn't allows tokens of that size anymore

it was going great like 2 days ago

and now the update just dropped

i mean, it allows it, but looks like the price of credits has been increased

[lilac sail]

I tried on Yupp and I realize this is a fal problem. If I use fal as provider on yupp, the output is always 1280x1280 regardless of input size. But if you use replicate as provider, then the output image has the same aspect ratio as the input image, which is the correct official API behavior

I feel like fal is not doing the model justice if it provides suboptimal behavior. I wonder if lmarena can switch its seedream4 provider to replicate too

[elder solar]

seedream4's bitrate quality is quite down tbh

[keen beacon]

It's literally GPT 5 with another system prompt, a bit more friendly than good ole GPT 5 but not much, seems like model behavior is set mostly during training

[fading shadow]

hello everyone!!

[burnt sinew]

hey

you should switch vc

[past parrot]

i mean its talking like a chatgpt 5

not gpt-5 model that is in api

[keen beacon]

Because it is a ChatGPT 5 model!

[past parrot]

i guess yes

im too confused

[surreal creek]

language categories are a separate table, should be available in the drop bar

[wispy owl]

Making the future

[fading shadow]

hello everyone

[burnt sinew]

hello again

[fading shadow]

hi

[coral goblet]

is seedream 4 down ?

[dreamy vine]

hi

[echo aurora]

It's probably getting a lot of traffic at the moment

[remote arrow]

And RIP Nani Banani..

[coral goblet]

i can't generate a single image 🥀

[fringe turret]

Hi everyone! Is Nano Banana still the king of the hill, or has it been dethroned again?

[coral goblet]

it got dumber

[hearty ferry]

Lmarena was too good initially

[native aspen]

hello new here.. anyone have other recommendations for AI video generations apart from LM Arena?

[vernal wigeon]

what's the best secret ai model right now on lm arena?

[hearty ferry]

But now ai models are creating more troubles

[south tide]

if somebody know how to use lmarena to generate a picture base on original picture

seems like everyone question here bt no one know how to really use it

😆

[limber crag]

im sorry but am i using a wrong seedream 4 on lmarena cuz the image generation i am getting is no where near nanobanana?
left is seedream-4 and right is nanobanana

Put a model in a Professional look-book closeup shot for a Casual-wear collection storyboard. The setting is a minimalist apartment living room at dawn (Around 5 am). The model is sitting on a beanbag by the window with a cup of coffee in her hands with legs slightly bent backward casually. The woman is looking directly at the camera mid-sip with her tilted downward with a shy look that breaks the fourth wall, this shot is very close to the camera as if she is right infront of it. The woman is wearing a lazy designer soft, oversized knit sweater blush pink shade, paired with black joggers. The dawn day-break casts a tranquil golden hour glow from the sunrise mixed with the cool lighting from the moonset. messy bun, loose strands framing her face.

this is the prompt i used

[coral goblet]

ikr it so bad

low memory and didn't follow instructions

[limber crag]

just confirming but this is the one right?

[spice flower]

Hello

[coral goblet]

yeah

thats the only one

[sturdy mica]

https://tenor.com/view/fılbus-fılbısm-doctor-nowhere-oddıty-compendıum-ın-fılbus-ı-trust-gif-17308481471181024920

[languid crescent]

what's the new gpt 5 about?
gpt-5-high-new-system-prompt

[honest summit]

hi

[solid mist]

why no sound coming from my generations?

[autumn cipher]

hi

[solid mist]

hi

can someone tell please

[daring quest]

after 10 minutes, where does my video end up ? Cant fint it.

[frozen hornet]

BMW running on a highway

[south tide]

hi

everyone

[random wraith]

Generation failed. Failed to create evaluation session. how to fix_

[molten coral]

Hi

[fading rover]

It's funny with NanoBanna lately I find it uses our generation cache to give results again and again when it fails to take out described prompt .

[alpine coral]

when it was an academic project, it was great. their only incentive / motivation was to collect user preference data of LLMs in the wild

now it's a commericial undertaking, with more and more friction, and less and less community utility

[wise condor]

انشئ لي الصوره في بالذكاء الاصطناعي في محل تجميل راقي بلون الزهري واضاءات خفيفة

[barren prairie]

Don t generate pictures here 😅

[hallow axle]

gemeni 3.0 release?

[shadow wharf]

It's funny with NanoBanna lately I find it uses our generation cache to give results again and again when it fails to take out described prompt .

[crude storm]

yes the seedream 4.0 is it just hoax cozi also tried in lm arena and nowhere it is touching the hieghts of nano banana or we re using the wrong seedream 4

[south tide]

@molten coralhi my friend

[crude storm]

see dream vs nano banana

[neon idol]

What,site do you use?

[crude storm]

weavy

[neon idol]

Is free?

[crude storm]

give some credits to try

[neon idol]

Ok thx

[radiant sentinel]

..

[neon idol]

Ohhh I understand what is the problem

@limber crag the problem is that seedream 4.0 is really efficient in showing everything you write, so if you write a cup of coffee it shows a cup, but a strange cup that shows the inside and therefore you see the coffee. If instead you write a cup then it will create a normal cup

seedream 4.0 follows the prompt very well, sometimes too well

Aniway this is another try with the same prompt

[willow grail]

sd4 doesnt generate often at all

[dense smelt]

hey so is there daily limit to Lmarena ?

[willow grail]

too many generations?

[limber crag]

ah that makes sense thank you!

[limber crag]

Not daily

its per hour

like once you run out your wait should be around 50-60 mins

[dense smelt]

how much credits?

or words

[limber crag]

no credits or words

this isnt a paid platform

yet*

its for training ai models

[dense smelt]

there should be a limit?

[golden ocean]

im training @limber crag

selling him in 10 days

[limber crag]

yeah in my experience once you get around 20 generations it will ask you to come back in 50 mins

[dense smelt]

ig then it's better to use gpt plus

[cedar solstice]

Good morning house

[dense smelt]

travis scoot?

[neon idol]

Nah mkre generation

Like 40

[limber crag]

yeah i do not know the exact limit

[limber crag]

sure if you want only gpt plus

[dense smelt]

i mean i used it .. now some friend needed it although my experience was trash.. so ig i should tell him to use gpt plus

[clear nova]

1. Electric Charges & Fields - Coulomb's Law, Electric Field, Gauss Theorem.

[wheat gulch]

anyone know why this always happen to me?

I have retried for about 20 times🤣

and it still don't let me generate the image

[soft belfry]

"Create a realistic cinematic video where the person in this photo is walking forward slowly and naturally, moving hands and legs like a real human. The face must stay exactly the same as the photo, with natural blinking, soft smiling, and lips slightly moving as if speaking. The background is a calm street with warm sunlight and a soft breeze, making the video look lively and professional."

🎙️ Bangla Voice Script (for voice generation):
"আসসালামু আলাইকুম! আমি ধীরে ধীরে হাঁটছি, তোমাদের সাথে কিছু সুন্দর কথা বলব। আল্লাহর পথে ফিরে আসো, নামাজ কায়েম করো, আর ভালো কাজের দিকে সবাইকে ডাকো। জীবন খুব ছোট, তাই আজ থেকেই পরিবর্তন আনো।"

[winged badge]

جنب كريستيانو رونالدو

[drifting crow]

gucci gang gucci gang gucci gang gucci gang

[obsidian shell]

guys in terms of image editing

when it comes down to symmetry for image editing

what is the best model ?

[golden nest]

seedream4.0 and nanobanana

[obsidian shell]

nano is torturing me with symmetry

especially in textures like carbon fiber

[limber crag]

tell me the prompt

ill do

[brave cloak]

Hello

I'm trying to combine 2 characters from 2 images into 1 image together, using seedream4.0, and it always ends up with distorted faces, what am I doing wrong? Can someone help?

[balmy merlin]

When I am posting an image in 9:16, why it gives me a video back in 16:9. can anyone help on this ?

[balmy merlin]

Dont use Seedream, Use Gemini nano Banana for great results. Thank me later....

[crude storm]

finally adobe has integrated

nano banana

we can see it this month

in beta

[calm basin]

Hello 👋

[ocean vortex]

Gemene ultra when

[brave cloak]

Banana is so censored, that's why I wanted the Seedream 4 to work

[barren palm]

@lavish ginkgo

[plucky island]

go to #1397655624103493813 please

[timber arrow]

🤔 anonymous-925?

[chilly yew]

Anyone else get the you have done your limit of 5 a day even though you only done 3

[crude storm]

txt model ? or vision model

[timber arrow]

text

[potent snow]

anyone know a good inspiration site for photo ai prompts?

[keen beacon]

It's Apple

Unfortunately nowhere near SOTA

It's also very low temperature because I got literally the same answer to exact same question._.

[calm basin]

I only done 1 video and it told me I already exceeded my limit

[lavish ginkgo]

Is it good ?

[timber arrow]

how do you know ive never seen it anywhere

[keen beacon]

I just got it a couple of times on LMArena

Battle mode ofc