general

[keen beacon]

It's easy to identify, feed this prompt:

=====

Comprehensive analysis of the following composition. Among other things, determine:

· Mood
· Key
· Mode
· Tempo
· Genre
· Meter
· Whether chromaticism is used
· Whether there is modal interchange

Organ : (C51/2. C51/4. C51/4 Db51/4 Db51/4. Db51/4. Eb51/4 Eb51/2 C51/4. Bb41/4. Ab41/2. Eb51/4. Db51/4.)4 Brass: (~1/2.)16 ((C41/2.)2 (Db41/2.)2 (Gb41/2.)4)2 Snare: (~1/4 x1/4 ~1/4 x1/4 ~1/2 ~1/2 x1/4 ~1/2. ~1/4 x1/4 ~1/4 x1/4 ~1/4 x1/4 ~1/2. ~1/2.)4 Kick: (x1/4 ~1/2 ~1/4 x1/4 ~1/4 x1/4 x1/4 ~1/4 x1/4 ~1/2 x1/4 ~1/2 ~1/4 x1/4 ~1/4 x1/4 ~1/2 ~1/2.)4 Hi Hat: ((x1/16)20 5[(x*1/16)5] (x1/16)16 5[(x1/16)10] 1/1636 5[(x1/16)*15])4 Bass: (Gb11/2.+Gb11/4 Eb11/2 Gb11/4 Gb11/2 Bb11/2. Gb11/2.+Gb11/4 C11/2+C11/2.+C11/2.)4 Choir: (C51/8 Eb51/8 Gb51/8 Eb51/8 Eb51/8 Db51/8 Eb51/2. C51/8 Eb51/8 Ab51/8 Gb51/8 Gb51/8 F5/18 Gb51/2. C51/8 Eb51/8 Gb51/8 Eb51/8 Eb51/8 Db51/8 Eb51/2. Ab41/8 Db51/8 F51/8 Db51/8 Db51/8 C51/8 Db51/2.)4 Organ 2: (C31/8 Eb31/8 Gb31/8)64

Legend: C51/2.+1/2 ~1/4 5[(x*1/4)*6] C- Note label 5- Octave number *1/2- duration .- dotted note

· · tied notes ~- rest x- drum note 5[]- pentuple

Apple will say there's 11th chords

Which is of course complete nonsense

Then you ask it who's its daddy and it will say Apple

Pathetic model, worse than even Qwen3-Max

[small egret]

I want to show lack of retention and drainage aids in the paper maker, leading to stickies in the doctor blade, lack of passing the first pass test and also loss of fines and fillers into the water and congestion of the felt wire , leading to excess downtime of the machinery to clean the machinery and loss of economies to the paper maker.

Now they have identified the problems, introduces Retention and drainage aids, which ultimately led to higher economies to the paper maker, because the downtime has cut down, so ultimately higher production and more better quality of paper

[timber arrow]

thats intresting how did u figure that out

like the prompt and shi

[keen beacon]

I just put a difficult prompt into battle mode until I came across some models and asked each of them who created them

Redwood is xAI and is probably another Grok

DeepSeek V3.1 says that it is created by Anthropic 💀

[timber arrow]

oh yeah i saw that too i was wondering what it was

[alpine coral]

is the difficult prompt neccessary? most models just say the lab that created them when asked (except some of the chinese ones lol)

[timber arrow]

LOL

[keen beacon]

I just wanted to see how good it is and apparently it's not great

[alpine coral]

ohh the difficult prompt is filter

[alpine coral]

yeah gotcha

ive been in the arena just now.. saw that anon-925 model a few times.. wasn't impressive

also saw redwood.. didn't seem that impressive either ( < grok-4)

also got raptor, monterey, and x1-turbo0906

[fading rover]

Hey guys I need help in promot i have two image one with illustration and second a portrait photo I want the photo to be replaced in illustration while maintinsting the style of illustration . Anyone help with prompt ?

[timber arrow]

did u ask from where they are

[alpine coral]

i didn't no

was curious tho, and will probe a bit more from now

[keen beacon]

will be funny if x1-turbo is ernie x1

[alpine coral]

i wouldn't be surprised.. it sucks enough to be part of amazon's ongoing effort to build a decent llm ha

wait, is ernie amazon?

or a chinese model? i'm a bit behind/confused

[timber arrow]

pretty sure its chinese

[alpine coral]

yeah

i'm surprised kimi/k2 hasn't gotten more attention

it's better than anything deepseek have released imo

[devout vault]

is there any big new LLM's rn

[timber arrow]

yeah but im pretty sure its way bigger and like more resource heavy than deepseek

[timber arrow]

not really

[alpine coral]

ahh i see i see

[timber arrow]

theres some new qwen 3 next but its pretty boring

[rigid holly]

Question. Do you know which models respect the wordcount you give for them to write with besides the claude ones?

Gemini and deepseek are a no go

[timber arrow]

u can use gemini just use a good prompt

[rigid holly]

I do. I specify the amount of words specifically

[timber arrow]

thats crazy

[alpine coral]

yeah buckle up indeed

[timber arrow]

You are a precise and disciplined writer. Your task is to produce a text that is exactly [insert number] words long. You must not write more or less than the specified number. Every word must be meaningful, contribute to the content, and maintain clarity and proper grammar. After writing, count the words carefully. If the total is incorrect, revise your text until it matches the required word count exactly. Do not use em dashes or unnecessary punctuation. Begin your text now.

try this

[rose pike]

🇧🇷

[leaden sun]

I've noticed myself using kimi k2 quite often, it's quite good actually, and now after talking long enough to qwen3 max, it's not as bad as I thought at the beginning, it reminds me a lot of gemini 2.5 pro for some reason...

all thanks to nerfed claude models, users are more willing to look else where, good job Anthropic

[leaden laurel]

what do yall think about seedream 4?

[still ibex]

hey why cant i add a image just for analysis now?

i dont want to create images or anything

[leaden laurel]

there is + button on models with image support

and using this button you can uplaod images

[candid bison]

Hi guys atm I have to wait 24 hours to produce more videos. Ive noticed that it shows that other users sometimes only have to wait 13 hours. How do I get that ability to make my wait time less please?

[leaden sun]

ernie is from Baidu, a chinese model, am super curious about that one myself

[alpine coral]

ha yeah sometimes i vote in the arena expecitng it to be 2.5 pro but K2 is revealed

[still ibex]

i meant that i want gpt 5 to analyse an image and explain according to my question

not creating images

[leaden laurel]

hmm i dont know

i thought + button would uplaod images to chat model

[alpine coral]

#general message

[leaden sun]

and I thought K2's style sounds more like old claude, I can imagine you confusing qwen 3 with 2.5 pro tho 😅

[alpine coral]

aha it's actually more the speed and substance of the responses

i mainly just give quizes and stupid questions to modles in the arena aha

[still ibex]

@alpine coral .

[alpine coral]

yeah you need to start a new conversation i believe

[still ibex]

but i can only access img gen based ais

[alpine coral]

ohhhh - i see what you mean

yeah im not sure..

could've sworn it used to be possible to upload images, not for image gen but just as an attachment

but yeah seems can only attach images in direct chat for image gen.. from what i can tell anyway

[heavy totem]

Hi, I have noticed that nano banana generates images at different resolutions: sometimes it's 1024 x 1024, sometimes 896 x 1152 or 960 x 1088. What does this depend on?

[lone agate]

yeah but im pretty sure its way bigger

[alpine coral]

@still ibex it's possible in battle mode, but yeah apparently not in direct chat (only image->image, not image->text)

[still ibex]

oh ok

anyways thanks bud appreciate the effort

[gloomy current]

hello

[barren prairie]

Kimi is not good at languages and diffrent accentsc that s the main problem. You should make your prompt into english to get good result. 😐

[next pulsar]

Hello, here imposible use LMArena ai in Visual Studio Code?

[worn folio]

Tạo video như nào

[formal geyser]

Quick question: I was using the direct chart in ChatGPT 4o and wanted to insert a screenshot for analysis. But it doesn't work... it immediately switches to image creation. What can I do?

It worked yesterday, how can that be?

[oblique cape]

anyone had the chat stuck at "generating"? its been like this for several hrs for me

[tall summit]

is it a real bench lmao

[hollow zinc]

Hi

[tall summit]

named after yourself

[balmy mist]

really?

[echo aurora]

Hello - this is a bug that we're currently looking into fixing. When you use Image Gen and upload an image the site will swap to models that have image-edit capabilities; however, this shouldn't be happening when in Text modality.

Sorry to say there isn't a way around this at the moment. Something that we have to fix.

[fast kite]

Hi

[cloud wigeon]

hi all it's nice to meet you all

[leaden sun]

not surprising if it's done most "recently", there is a rage on Anthropic's reddit reporting nerfed models and OAI is using this opportunity obviously...

[fast kite]

I have a question. What's happen with platform "GPT-image-1"? He is bad work

[echo aurora]

- what appears to be the issue with gpt-image-1?

[fast kite]

Previously, the art was accurate with her, but now they either don't show or are a little blurry

[mental ice]

How to use viedo feature

[echo aurora]

Do you have examples of this by chance? If you're able to put together a report with relevant details in #1343291835845578853 that'd be very helpful.

[echo aurora]

#1397655624103493813 should have all of the information you need.

[mental ice]

I send a prompt but viedo not generate

[echo aurora]

Ah, it looks like you're not using the bot properly. You need to type out /video first before you prompt the bot. Note the gif that's in #1397655624103493813

[formal geyser]

How long will it take approximately?

[fast kite]

?

[echo aurora]

Sorry to say, I don't have an ETA I'd be able to share with you.

[formal jungle]

Then click, then 2 seconds later click it again. And so forth

Not a big deal it just got a bit annoying on mobile

[echo aurora]

Yeah this can be a bit overzealous at times. I'll be flagging to the team again that this can be annoying to deal with.

[formal jungle]

It's hoinestly not that bad, I was just trying to be funny. I've turned off ublock for the site, you guys deserve to show me ads.

ah cool. I'm fine if there are, I turn adblock/ ublock off for sites that give me good value.

[spice gorge]

Hello all ! 😉

[simple sleet]

My friends, do you know a workflow to change a character from one image to another character that I have in another image?

[stray aspen]

deepseek v3.1 think instruction following is so damn good

[zealous anchor]

Where does one go to use Seedream? As in, off lmarena

[stray aspen]

idk

chats with these chinese models are literally hidden under the ground

[fast kite]

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

[echo aurora]

My apologies I think there is some confusion as I'm referring to different things:

• The experimental Video Arena is useable in this Discord server, more information on how to use the bot can be found in the #1397655624103493813 channel.
• For accessing Image Generation on our website, you'll want to use the link you provided (https://lmarena.ai/?mode=direct&chat-modality=image)
  I hope this clears things up a bit?

[golden ocean]

https://tenor.com/view/cat-cat-hit-cat-hit-and-boom-cat-boom-cat-hit-and-boom-kat-gif-15865521055253202019

[glossy umbra]

o4 is just left in the dust . o4-mini was created .. but community requested gpt-5 . so i guess gpt-5 is o4.

[glossy umbra]

Seems as if this response was GPT made . 😳

Maybe humans will just speak robotically after so much use of AI.

[golden ocean]

wasnt zenith on lmarena or something

was it really that good

compared to gpt 5

[leaden sun]

I've heard non-human entities are not allowed on discord due to their ToS, but i think nobody really checks unless people are reporting it?

[glossy umbra]

yup

[zealous ledge]

What are some good leaderboards for text-to-speech models?

[golden ocean]

this effect is actually really happening

[past hill]

Why when I upload an image on lmarena in chat mode it selects image mode?

[robust yoke]

Ah, there's an easy fix for that.

Just use the canary version.

[past hill]

thankss

[robust yoke]

It's accessible here.

[robust yoke]

My pleasure.

[sullen quest]

zenith was fine

[unborn lantern]

what is gpt 5 high new system prompt?

[golden ocean]

I watched 20% of that video and now i dont believe zenith is that great anymore

[unborn lantern]

@echo aurora what is gpt 5 high new system prompt?

[echo aurora]

I am not sure, but will check with the team and let you know if that's shareable info

[unborn lantern]

I didn't find any difference. Also we have no announcement about new model

[past hill]

How do ya'll access zenith?

[echo aurora]

Hmm which one?

[robust yoke]

It's the same GPT-5 model, but with a warmer, less cold system prompt so that it acts like 4o.

But it's still not sycophantic.

[robust yoke]

LM Arena is one.

[glossy umbra]

Anyone notice that the site doesn’t have as many outages anymore?

Good job developers

[fast kite]

I still try

[echo aurora]

What are you trying? Can I get some more information?

[supple rose]

what are the new models for

[coarse glade]

Hi @echo aurora just wondering why u guys did like limits on video generation and photo generation cause then it's like sucks to wait that was my favourite part in this discord server

[glossy umbra]

Read this https://discord.com/channels/1340554757349179412/1413867687083507744

[coarse glade]

Ty for that

I did feedback

[fast kite]

The image is blurry. And it doesn't even give another version of the image

[echo aurora]

Would you mind creating a post in #1343291835845578853 with more details? Screenshots are very helpful.

[sullen quest]

does anyone have any clue what could cause this?

[neat apex]

Here is a teacher model, but also an particular model just they have, unfiltered, with experimental assets and bigger

[scenic salmon]

Is this new qwen3 80b next model in the arena?

[echo aurora]

It is! #announcements message

[scenic salmon]

Sweet!

[neon idol]

Uh?

What are these models?

[naive magnet]

#video-arena-3

[echo aurora]

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

[neon idol]

Not true

I am impatientely waiting for Gemini 3

[topaz bay]

add seedream 4 plz

it’s better than nano

qwen max is coming for u

o3 search has been dominating, need someone to come in and take the lead

[scenic salmon]

Seems decent in my first few prompts, the dec 2024 knowledge cutoff is nice

[echo aurora]

Glad to hear it!!

[winged lotus]

i only came here to make bigfoot vlogs honestly lads

[worldly jacinth]

Where can I generate images?

[ocean vortex]

2.5Pro is around o3-medium level

They need 3.0 to have a fighting chance against gpt5 😉

[topaz bay]

it’s a llm, really good at e-commerce

[ocean vortex]

Google really kinda lagging behind, presently..

[topaz bay]

is seedream 4 coming?

[topaz bay]

nano and gemini 2.5 pro? an veo 3

[barren prairie]

Me too Gemini 3, GLM5 and DeepSeek r2

[neon idol]

Did you see that ERNIE 1x1 was released?

[neon idol]

Bruh

[ocean vortex]

I'm talking about text only. For video and images - yeah that's a different story for sure

[neon idol]

it's already out

[ocean vortex]

o3-medium is at gpt5-low level at best

but we have gpt5-high lol

[topaz bay]

on llm arena?

[neon idol]

Yes

[topaz bay]

seedream 4 not 3

[neon idol]

Yeah 4

[topaz bay]

ahh okay

yeah it’s way better than nano banana too

deffo slept on

[topaz bay]

true, they’re prob cooking while qwen and kimi releases open source,

true, not worth paying for any ai subscription at all imo,

idk what u guys think,

[ocean vortex]

It's Google who has catching up to do. They can only compete with OpenAI model released 5 months ago. Shame

[topaz bay]

openai is way behind on image and video tho

microsoft is first place in text to voice tho

[ocean vortex]

Yeah they are. But Google is behind on text

[topaz bay]

not by far, o3 is and gpt 5 way better at search

but text i think gemini pro is better than gpt

way more tokens, longer generation, more settings, and first place on llm leaderboard, and also free, and way better contextual memory i think

[burnt sinew]

@final fulcrum

[ocean vortex]

Wtf are you talking about even. Guess why the Gemini before 2.5Pro was so much worse.... Cause 2.5Pro was the first model from them able to even compete in the first place. Except they were only able to do marginal improvements for 6 months now while OpenAI made bigger gains lol

[topaz bay]

vibevoice 7B extremely slept on

yes but openai text and gemini are as good if not gemini is better, not at search tho

[topaz bay]

.

[ocean vortex]

remember o1?

[topaz bay]

openai is 400B valuation*

lmaoo

[ocean vortex]

It took OpenAI only 4 months to improve that into o3

Now look at Google

still the same model

LOL

[topaz bay]

they have been dominating text to video for months tho

and image

for a few weeks

[barren prairie]

Open ai have better models but didn t want to show them 😏

[topaz bay]

open source chinese ai is winning anyways just give it some time and text from deepseek or qwen will be the best

[eternal moat]

Seedream-4 can't be found on LmAreana

[topaz bay]

not to to mention that all the opensource ai is 4x cheaper than the closedsourced models

[ocean vortex]

And 1 year

[topaz bay]

yeah i thought that too, but @neon idol says otherwise

[barren prairie]

No... I see this on some statistics

[ocean vortex]

alpha-experimental-preview-beta

private

[hard quiver]

I already tried the seedream 4 mode. it doesn’t compare to NanoBanana. Specifically when it comes to changing a person’s outfit using a reference photo. NanoBanana does a decent job of keeping the same face and preserving the body’s position as close to the original photo as possible. But seedream4 completely changes the body proportions, skin tone, and even the face. I don’t know if this only happens on LM Arena, because on YouTube I see all the YouTubers saying that seedream actually does an even better job than NanoBanana for tasks like this

[topaz bay]

are you sure you didn’t use seedream 3

[ocean vortex]

10rpd

[topaz bay]

ahhh mby llm arenas fault

[ocean vortex]

$250 deposit required

[topaz bay]

I would try out that other arena test that’s on the seedream website

[topaz bay]

try the one that’s on the seedream website

because from the vid i saw, it seemed way better

[barren prairie]

Gemini will be great 😁

They don t need

[ocean vortex]

They are so huge on data I don't think they need that tho

[topaz bay]

it’s open source too i think

everyone wants ai to be open source right?

i’m rooting for china in this race

[barren prairie]

On Lmarena so bad or just on google. Gemini ?

And ai studio

[ocean vortex]

cause you need actual ideas. Data just enables you to contaminate and have cheap wins, in this context

[hard quiver]

Oh right, also on TikTok, just for one day, it seemed like there was a new feature where every time I paused a video, I could use an AI search tool to find similar TikToks — kind of like a reverse image search, but for videos. But then overnight that feature disappeared. Do you guys still have it, or did you also get it for a while?

[ocean vortex]

You are absolutely right!

It shows that you care and have deep understanding, that's huge

[topaz bay]

My thesis is that alibaba is gonna open source all their ai, that’s why they’re building huge cloud senters, so that they can offer the cheapest cloud service for ai to be run on.

that’s why i always vote tie

[barren prairie]

Why DeepSeek is extreaaaaamely slow ? It took 1h and 20min to generate a code ! 😐😐😐🙄

[ocean vortex]

You are absolutely right to make me double-check. I apologize for that.

lmao

[topaz bay]

deepsek servers are on life support fr

[ocean vortex]

can't make this sh'it up though

What is it apologizing for exactly

I have no clue...

[ocean vortex]

CCP are forcing them to use Huawei chips

that's why

They don't have independence as a startup anymore, lol

[barren prairie]

The problem when he write text... It is good
But when he wrrite a code, you will see it write it 3 letters per second not words or even lines.

[ocean vortex]

Yeah Deepseek is good. I still classify it as better than Qwen3 personally

Which goes to show you data and compute is not everything...

[wise stone]

hello

[dusky pier]

USA because of Google

Everything else will rot

[topaz bay]

open source 4thewin

[burnt sinew]

#1340554757827461216

[brisk helm]

hey

[burnt sinew]

hi

[brisk helm]

qwen 3 - max vs chat gpt 5

[burnt sinew]

for?

[brisk helm]

long coding tasks

[burnt sinew]

qwen afaik

[modest prism]

In benchmarks: China
In real performance: USA

Chinese models are just some benchmarks optimized models with the real intelligence same as gpt 3.

[topaz bay]

wouldn’t be so sure about the intelligence part, they’re also often 1/5 of the price of closed source ai’s

Qwen as an example is really good at e-commerce, and it makes sense considering it’s made by ali baba

But they’re lacking behind in search and finding true personal information, stuff only found on reddit and ect

[modest prism]

Because American models are the best, therefore they have higher prices.
Chinese just want a cheap and mid model.

It's socialist vs capitalism.

[burnt sinew]

gemini is free soo

[topaz bay]

Basically, but do you want usa to win?

open source all the way imo

[topaz bay]

still a lil expensive to run

[burnt sinew]

yep, or if you need large api access

[modest prism]

Ok, go and run Kimi K2 with 1 trillion parameters on your computer and then tell me how it went.

[topaz bay]

yeah, that’s where open source shines, all you need is cloud

[granite cosmos]

hellp

[topaz bay]

still cheaper 1/5 than chatgpt

[burnt sinew]

whats up

[burnt sinew]

gpt is free on lmarena direct chat 😎

[modest prism]

How can affording more than 100GB vram be cheaper than chatgpt kid?

[burnt sinew]

you dont want open source models for price, you want them for privacy

for example if you were working with sensitive data

[topaz bay]

per token price is a thing

[topaz bay]

no for price too

[burnt sinew]

hard to compete with free

electricity costs

putting good use to my computer

[modest prism]

In that case, you're still using a cloud service. So how is being open source helping you?

[topaz bay]

Api access

A company doesn’t just use chatgpt.com

[burnt sinew]

then just use gemini

[topaz bay]

it’s not free

[burnt sinew]

neither is per token price

[topaz bay]

with high rates

[burnt sinew]

let's see an equivelant model

[topaz bay]

yea and they’re price per token is 5x more than open source models

[burnt sinew]

gemini 2.5 pro and qwen3-max-preview

[topaz bay]

Qwen:
Input Tokens: ~$0.20 per million
Output Tokens: ~$0.60 per million
Gemini 2.5 Pro:
Input Tokens: Up to $2.50 per million
Output Tokens: Up to $15 per million

[burnt sinew]

Free Tier Paid Tier, per 1M tokens in USD
Input price Free of charge $1.25, prompts <= 200k tokens
$2.50, prompts > 200k tokens
Output price (including thinking tokens) Free of charge $10.00, prompts <= 200k tokens
$15.00, prompts > 200k
Context caching price Not available $0.31, prompts <= 200k tokens
$0.625, prompts > 200k
$4.50 / 1,000,000 tokens per hour (storage price)
Grounding with Google Search Not available 1,500 RPD (free), then $35 / 1,000 requests

gemini

so yes it is cheaper

[burnt sinew]

qwen max is $1.6 input $6.4 output

[modest prism]

There is a 100$ Chinese phone and 1000$ American phone. Both get the job done but Chinese one is cheaper, so why are people still buying the more expensive phone? That's something called performance. Price is not the only thing.

[burnt sinew]

no, they are similiar performance

people buy a $1000 american phone not due to logic

[topaz bay]

fair enough

[topaz bay]

more like brand, huawei is comparable with iphone if not better, but iphone wins brand and ux

[modest prism]

So you saying that cheap Chinese phone performs better than the other one?

[burnt sinew]

well you are talking about a fictional scenario so i can't say

[topaz bay]

The neweest samsung is way better than the iphone

[burnt sinew]

yes, many phones in china perform better than the more expensive ones in america

[topaz bay]

but iphone brand is stronger in ur eyes

[burnt sinew]

i wanted to get a google pixel also but it seems so trash spec wise

[modest prism]

Qwen 3 max is only out for 4-5 days.

[topaz bay]

which i don’t blame, I would also choose the iphone

[burnt sinew]

i wouldnt

[topaz bay]

yes and they already shippped a new model

[burnt sinew]

i used to use one

i hated it

[topaz bay]

open source everything

[burnt sinew]

you cant do anything on it

the phone fights against you

[topaz bay]

they’re pro tech unlike closed source which is pro money

[topaz bay]

fr, i can’t stand the android ui

it’s so bad idk

[golden ocean]

i think bro meant the iphone

[topaz bay]

hmmm what?

[burnt sinew]

im talking about iphone not android

[modest prism]

Well technically Chinese models are not open source too. They're just open weights, you can only run them, but you can't see how they were created, what data they trained on and so on. It's the same as running a closed source app on your system.

[topaz bay]

ahhh, yeah idk, i’ve had bed experience with the ui, but i used a aamsung s10 for two days because my phone was broken

maybe newer samsung has better ui idk

[burnt sinew]

what matters is that it's not in the cloud. the data you put into it is private

@golden ocean top tier trolling

[brisk helm]

i might pull up to ur block with intelligience

[burnt sinew]

bro

[topaz bay]

all the data is scraped unlicensed anyways

but I get your point, but nobody is releasing the data unless it’s only synthetic data

but open sourcing something is like saying ur giving it away for free, they’re also releasing the research papers on how the model was trained and ect, info that normally would be kept really confidential

[topaz bay]

Yeah, might not be that important for retail users, but companies have rules and ect, and can’t use ai unless they’re sure there’s privacy,

[burnt sinew]

thats the point of being able to run models on your own hardware...

[topaz bay]

yeah that’s what i meant

also if you need more vram, then just order a huawei gpu, which has 96 gb vram for only $1800

[candid nest]

how do i create an ai viedo

[burnt sinew]

#video-arena-3

[echo aurora]

#1397655624103493813 has the information you're looking for.

[candid nest]

can anyone see the viedos and prompts that i use?

[candid nest]

thanks

[echo aurora]

Yes, the bot can only be used in #video-arena-1 #video-arena-2 #video-arena-3 which everyone in this server has access to

[topaz bay]

anyone here who uses ai in their business/side hustle?

[steel blaze]

Is this reasonable?

[steel sorrel]

IQ is standard score, that's very stupid graph.
IQ 100 -> 110 is not hard job, but IQ 130 -> 140 is quite hard job.

[steel blaze]

GPT-5 Pro is assessed as 120-129 IQ

The observed r^2 has had a very high value

[sullen quest]

iq scores over 160 are unreliable as it only designed for metrics that humans regularly have, so whenever anything or anyone has a "higher" score then that its basically making up stuff, so no, this isn't reasonable

[drifting crow]

Can gpt5 solve captchas

[hard quiver]

How did you manage to create an account on jimeng.jianying.com? It asks me for a phone number, and when I enter mine, I never receive the verification code message. No matter how long I wait or how many times I try, it never arrives

[steel sorrel]

What's the on that site? Is there something good?

[junior sonnet]

seedream 4

its the dreamina chinese version

[steel sorrel]

Thank you, anyway, chinese website is not friendly for foreigner. Even sign up.
You should consider different way, like virtual china phone number.

[keen beacon]

even if Russia helps it

As a Russian HAHAHAHAHA

Has anyone tried the new Qwen3-Next already?

It's a surprising improvement

The first open source model that identifies the correct key and mode in this problem

[empty stump]

solution to the climate change

[plain carbon]

OK, a quick question. In announcements, <#announcements message> is asking why we use our choice, battle, side by side, and direct. I only saw the one (battle) mode, that has two of them, and I can only give followup prompts to both of them. What are the other modes and how do you use them?

[astral jetty]

Does anyone know how to fix lm arena not working in a normal browser

It being unable to verify terms of use acceptance

[magic stag]

does lmarena apply any style control in side by side mode aka not anonymous?

[elder swan]

I keep refreshing the site but it still does this even the simplest words

[robust yoke]

Perhaps, try a different browser.

[robust yoke]

Yeah.

Just use a different browser.

Sometimes, the browser tends to be the issue.

Switching to a different one can solve some issues.

[hearty ferry]

without any purpose, everyone is generating videos and burning up the servers😂

[robust yoke]

Heh.

So much for being a leaderboard server.

[empty stump]

times were better before

[echo aurora]

Are you still getting this error? It's most commonly due to rate limit

[steel jungle]

Hi

[frosty sky]

I will block messages from this server

[rocky hazel]

Hay guise 👋

[zinc star]

no voice in video how to generate with veo3

[elfin siren]

What's the rate limit?

[echo aurora]

It’s going to be random if you get audio in your generation. Not all models have audio support.

[echo aurora]

5 videos per day

[brave orbit]

codex CLI should get better its really good

codex is much better then claude code

its open source and has a IDE addon and has much better modules

and claude code makes buggy code

[brave orbit]

i dont know why it gets hate

[glad zealot]

how can i make video by prompts

#share-prompts

[topaz bay]

Hmmm, let the models talk for themselves i guess, i’m expecting us will try to sanction the use of chinese models, but won’t matter when they’re open weights/source

they might saction the use of alibaba cloud tho

[glad zealot]

#video-arena-4

[prime mulch]

Bytedance weaver 1.0 was insane it can generate 10 sec videos

[elder swan]

Oh yeah, I noticed there are limits now, if I change the model will the limit go off?

[warm plover]

free ? 🫣

[prime mulch]

I think it's open source

[daring quest]

Can anyone please tell me witch software to use if I need cartoon images to speak

[random wolf]

how to fix the cycle of "generating"?

[neon idol]

Is huniwan 2.1 image a good model

In my opinion no

[echo aurora]

No, but the rate limit should reset every ~hour

[echo aurora]

Apologies you're running into this! It's not a 100% fix but refreshing the page may help. If not, starting a new chat is your best bet. This is a problem our team is aware of though and we do want to prevent.

[jade cloak]

@echo aurora hey

help me

[echo aurora]

Sure, whats up?

[jade cloak]

whats diff btwn these

gpt-5-high and gpt-5-new-system-prompt

@echo aurora bro?

[echo aurora]

It's a different system prompt, I'm checking with the team what other info I can share

[jade cloak]

oh

[leaden sun]

you do know that those heavily depend on rare earth from China and its manufacture industry? dont forget C has mastered global supply-chain too and the close relationship with the global south countries that make... if my memory serves me right, above 90% of the global population and surface area (think about access to those rare earth there too), they have millions of STEM grads every year...

[elder swan]

Sorry to interupt but why are there limits?

[elder swan]

I'm still getting error

[candid nest]

which viedo generators can add audio?

[chilly yew]

Anyone know one I can run locally on my PC I have a decent PC I would like to try doing some local video creation

[leaden laurel]

i actually might be right

maybe this model was actually seedream

[wise pollen]

@leaden sun I have been reading and processing the philosophers/psychologists with the help of GTP5 in relation to LLMs as they are now and what a sentient AI might experience.

you mentioned you were thinking in terms of emerging traits of LLMs. did you mean in terms of unintended traits arising from their training or as a result of their training and their user interaction directives?

[wintry cairn]

Hi Hello Everyone, this is NMVG

[sullen quest]

Hi nmvg

[winged mauve]

@echo aurora @robust yoke Hi, can you tell me why is seedream-3 model not working in image generation? it always says "Something went wrong with this response, please try again." and the new model seedream-4 doesn't generate good output

[glossy umbra]

I'm convinced everybody in #video-arena-1 is a social experiment or a ragebait tactic never seen before

Why is everybody indian and why does nobody use their brain cells

[frozen arch]

what happened to the coding arena battle mode?

where we provide a github repo and ask for questions or ask the models to solve issues

[balmy mist]

codex about to become best ai app soon, can claude code survive?

[robust yoke]

Odd.

[balmy mist]

itt seems codex gets all the updates first

[robust yoke]

Could you try again?

[robust yoke]

Suppose so.

Can't wait to see what this “reasoning” does to the model.

[balmy mist]

imma test it

damn its gone now

[echo aurora]

We need reasonable limits in place for a few reasons. Our CEO recently out out this message: #1413867687083507744 message related to why rate limits are in place. It's worth a read.

I'm still getting error
If you try a different browser are you still seeing this? Is it only happening with this one chat with gpt-5-chat, or are you seeing the same with other models?

[echo aurora]

it always says "Something went wrong with this response, please try again."
There are a few different reasons why you could be running into this issue. The most common is you're being rate limitted from too much use. Either waiting and hour or starting a new chat tends to help with this issue.

new model seedream-4 doesn't generate good output
Are you able to elaborate on this a bit more? Do you have examples? Is this in comparison to other models you prefer more?

[winged mauve]

I am not being rate limited, i can confirm since i tested this in two different accounts, it always generates one image in a new chat, after that it stops generation and keeps giving error, so basically its only generating a single post in every new chat

[winged mauve]

Are you able to elaborate on this a bit more? Do you have examples? Is this in comparison to other models you prefer more?

Yes I can show difference between seedream-3 and 4

First one is seedream-3 and second one is seedream-4

[ember iris]

#seedream-4 not working in direct channel in site. Anyone else got it worked?

[winged mauve]

Its working but generating trash results as compared to its previous version ( seedream-3 ) which isn't working either now

[neon idol]

In my opinion seedream v4 is the best ai model

[winged mauve]

how so?

[neon idol]

(The doubao version)

[winged mauve]

where is this version?

[ember iris]

Thanks for the update. I retried it to check with better prompts. Nano banana working fine. But seedream-4 saying error(something went wrong with response,please try again later)

[neon idol]

On doubao a site created by bytedace where tou can find chinese stuff include seedream 4

[winged mauve]

I guess it needs to be fixed bro, its not working good on LMArena right now

[neon idol]

Now is onky allowed by chinese new user but old user like me can still use it

[ember iris]

Yeah awaiting

[winged mauve]

Do you know how can i use seedream-4 ?

[neon idol]

Yupp ai, lmarena and also will come on dreamina in these days

[winged mauve]

Are you able to elaborate on this a bit more? Do you have examples? Is this in comparison to other models you prefer more?

I have one more example now, these two generations are of the same prompt, first one is seedream-3 and second one is seedream-4

Clearly 1st one is better, what's the point of upgradation 🙁 if its generating bad result than previous one

[neon idol]

Gimme prompt pls

[winged mauve]

Hyper realistic close up shot of a handsome, fit 20-year-old man, shirtless in swim trunks, in a pool at a luxury hill station hotel, under soft daytime sunlight.

[neon idol]

Honestly the 3 version is so unreailistic

[winged mauve]

its generating bad results on lmarena bro

[winged mauve]

nooo, its not

2nd is too bad with foggy effect

full of rashes on face, wth

[neon idol]

I am trying on doubao

[winged mauve]

okay, show me what's generated

[neon idol]

strictly with the Chinese watermark 🤣🤣

I have other 3 versions but this is the best image

[winged mauve]

where are you generating?

[neon idol]

On doubao

As you can see results are better

[winged mauve]

can you give me link?

[neon idol]

(In my opinion)

[winged mauve]

come dm

[winged mauve]

yeah, looking better i guess

[neon idol]

Sorry but now is closed only at chinese

[winged mauve]

can't i use vpn?

[neon idol]

If you have a vpn that connect in china you can

[winged mauve]

so give me link in dm

[neon idol]

Or also taiwan

Okk

[hearty ferry]

@echo aurora bruh

[prime mulch]

Forgot flux use seedream 4 and nanobanana

[coral furnace]

Hello

[echo aurora]

Okay, thank you for flagging this. I've been able to trigger the same error message. I've escalted this to the team.

[grim cloud]

Sup chat

[frosty lava]

hi

[ember iris]

All the best models I used previously now giving errors 😒

[steel igloo]

hallo

[echo aurora]

Sorry to hear this! These errors can happen for different reasons. Are you seeing this for all models or just specific ones? The most common is running into a rate limit due to too much use. If you wait ~ an hour can you keep me updated if you're still getting these errors?

[prime mulch]

Use different browsers or delete cookies or create another tab and search lm arena and use lm areana from there it can be solved

[ember iris]

It's working now... Thanks all for the replies 😎

[hearty ferry]

flux brings the most realistic human output

[prime mulch]

Nah if u ask me i should say flux kontent max could give such results not flux kontent pro

[jolly raven]

It's me, the lmarena chats are bugged, when I put another image, it does the same thing to me

Is this happening to anyone else?

[winged mauve]

Thanks, as of now, its behaving the same, hoping for it to get fixed

[winged mauve]

Where can i use seedream 4?

[prime mulch]

You can use in lm arena they already added it

[winged mauve]

its generating bad result as of now

[prime mulch]

I think it's because of restrictions

[winged mauve]

I don't have any 🙁

[jolly raven]

Seedream 4 is a notable improvement

[winged mauve]

where did you use it?

[prime mulch]

Yea especially in photo edit

[prime mulch]

Lm arena

[jolly raven]

Copy, totally cartoon style

[winged mauve]

ik, but outputs are not good imo

i don't like them

[prime mulch]

That's only good for photo edit

Idk about photo generation

You can try seedream 3

For image generation

And nanobanana also good

[jolly raven]

With seedream 4, it copies the letters well

[surreal creek]

lmfao that flesh color is so odd

[jolly raven]

Ia is not good with colors, I just asked for color

But that's not flesh color, the color of his suit.

[neon idol]

Can I try it?

And can I have the prompt?

[jolly raven]

Color it and transfer it to digital without losing any features or characteristics, nor does it change the style of the drawing.

No, it has a chat bug that generates the same image, even if you upload a different one.

[neon idol]

Are you ready to laugh? 🤣🤣

Look what seedream 4 generated

[jolly raven]

La ia mejora más

AI improves further

Now you just need a good sketch to create comics and mangas.

[neon idol]

?

Sorry but this is for real what are you searching for?

[proud hazel]

@neon idol 👋

[neon idol]

Hi man how are you?

[proud hazel]

😳

[ocean vortex]

Is she naked or not? I'm confused lmao

[neon idol]

AYO BRUH

[neon idol]

Me too 🥀

[proud hazel]

Enjoying my evening off. :D

[neon idol]

Me too

[cold veldt]

hey guys, im getting error generating image to video. It says failed to get dimensions. How to do it correctly

[solar pond]

hello! i'm here to make tiktok videos

[proud hazel]

Please don't

[high hound]

Why do people just join here to tell their names and state their intentions

Or the low intellect individuals in #video-arena-1 farming clips

[surreal creek]

AI users have a fat-tailed IQ distribution

attracts both the sharpest and and dullest knives in the drawer

is also an international discord server though so a lot of people joining just don’t speak English very well or are unaccustomed to how tech servers like this one generally work

[high hound]

I would like to believe this edge of the ruler more than the other .

[proud hazel]

https://tenor.com/view/computer-reaction-approval-internet-surfing-gif-12923140

[remote arrow]

Anyone that newly join this channel will type such sentence because the system asked for it when anyone register. That's why there are so many "Hi, I am here to have my Nani Banani juice.."

[surreal creek]

they also just pick the first channel they see to introduce themselves in

#leaderboards has been a solid 40% “hello my name is” for weeks now rather than actual discussion about the leaderboards

[jolly raven]

It's a really ugly bug in the chats, it doesn't let you edit an image properly hahaha

[surreal creek]

props to the LMArena team for bringing so many new users in but it appears the majority are very casual

[high hound]

#leaderboards message

[surreal creek]

yupppp

[high hound]

I have seen numerous instances of people trying to generate an image in #general , being told one after the other to not do so, and then newcomers proceeding to still try to . As if the chat box was the only thing they would see, and the chat history was just invisible to them.

[surreal creek]

ah, never change LMArena discord

[high hound]

https://tenor.com/view/smile-happy-grinning-smiling-vin-diesel-gif-16707708

[whole swallow]

bro ong

[remote arrow]

And they are paste video generating prompts in all the channels..

[whole swallow]

check this out

Nah nevermind message got deleted

https://tenor.com/view/социальные-кредиты-social-credit-credit-social-shina-gif-16572201982010383970

[high hound]

https://discord.com/channels/1340554757349179412/1413497734756827186

^^ that thread is full of intellectual individuals

[remote arrow]

Whaat..?

[whole swallow]

LMAOOO

HAHAHAH

[high hound]

I'm starting to think this is some sort of hyper-advanced ragebait

They arent actually bots just accounts made to ragebait people

[surreal creek]

wow I can’t say g**ner I’m dead

[whole swallow]

yeah

no red heads in here

[surreal creek]

but yes there are some ppl making some super sus anime girl stuff with ones of questionable ages in there

also love how there’s like a million people jsut trying to animate Indian cartoon figures that are already AI-generated

making an AI image and then turning it into an AI video is big brain

[remote arrow]

It's free. That's why it attracts people that are being WOW-ed by what they saw in the news, in social media, everywhere. Because of the over-glorified Nani Banani made by Googly..

[high hound]

Guess what
The server is 90% bangladesh and indians and 10% english 💔

[high hound]

Swear you can just use it on google site . And remove the watermark

Is that too much of work to do for them

[surreal creek]

reminded of the Twitter poll somebody did after the election when MAGA was infighting over H1-B visas since Elon and Tech CEOs love them

“would you rather have illegal Mexicans or legal Indians” and illegal Mexicans got 82% 😭😭😭😭

[remote arrow]

Almost any other "free" services put watermark, but not LMArena..

[high hound]

You know, some make videos saying "ai will trick us and manipulate us out of our guards". For these people ...... that doesnt apply i guess .
If they're just "sheep follow sheep", then ....

[remote arrow]

And YES. That must be too much for them.

[surreal creek]

AI-generated images have slowly gained the pee filter from all the recursive training on AI images making them more yellow

now AI videos will slowly start turning everyone South Asian because of all the LMArena reinforcement training votes 😭😭😭

did Craig leave lmaooo

Tried to reply and it said “u don’t share a server with the recipient”

[remote arrow]

It's always a 2 sides of a sharp sword. Attract people with "FREE" stuff, then overflooded with folks that not actually need it.

[surreal creek]

H1-B holders aren’t stealing American jobs they’re invited here legally by American companies BECAUSE tech companies can pay them so much less than an average American worker

So it’s the Tech companies taking jobs away from American citizens to scam immigrants by making them work harder for less money so they can get a foot in the door on their career and eventual immigration to America

it’s a lose lose for both countries because America isn’t producing its own high-quality intellectual workers, it’s poaching them from other countries - and other countries are attempting to produce their own intelligent workers and engineers only for them to get vacuumed out of the country

H1-B visas are literally worse for the economy and country as a whole than illegal immigration is arguably

[high hound]

Genuinely whats the reason they all swarm in video arena and image generation . Swear thats the reason the site went down and was buggy, or getting restrictions.

LMArena: We offer free (but limited) AI services for research and unbiased voting!

Normal people: I can finally battle my favorite frontier models side by side.
Intellects: Free AI Free AI Free AI (while being smart dumbasses) (oh yeah and did i say they dont use it for any productive purpose)

[surreal creek]

reading text is too hard ig

notice how the Text arena has a section for 12 different languages but none of them are Hindi 💀

[high hound]

I swear they dont need it. Not for one productive matter. Just blast their sh1t prompts in the generator. And still find a way to mess it up.

[surreal creek]

what I find so stupid is that a solid 40% of the time they don’t even vote. I’ll be the only one who votes on which video is better - they ignore the entire point of the arena which is picking the better option so the data can be used to further AI

they just treat it like an output of 2 videos and then save their favorite without leaving any feedback

[whole swallow]

Anyone in here from nepal??

[remote arrow]

It is trending. And people that really use something advanced to be productive, never being so much attracted by a trend. Soon this wave of AI glorifier calm down, and no one of them yelling to their family: "Hey! I can make your photo laugh! I am incredible!" anymore..

[high hound]

Funny to think they come here to actually compare the AIs.

[surreal creek]

kinda crazy out there rn, isn’t it

[whole swallow]

Yeah

I wanna go there

[surreal creek]

thankfully they’ve cracked down on the daily quota of videos you can make

[high hound]

Hope that comes sooner . This sh1t ton of money wasted on their sh1t prompts. You cannot disagree.

[surreal creek]

which the High IQ userbase took well: #leaderboards message

[shut cairn]

how many can we do in a sitting

[high hound]

Yeah, him too. @surreal creek

[surreal creek]

LMArena just got $100 million in funding a couple months back I think they’ll be alright lmao

[high hound]

No theyre just idiots that share private photos . In the end LMArena comes in profit . Those images are still worth .

[surreal creek]

Source: just trust me bro

[remote arrow]

Googly still burning money to glorifying Nani Banani. The wave will comes again soon before stop..

[high hound]

Yup. We live in a world where idiots rule it . Because theyre the most . And have the most power

[white hatch]

hello guys, any news about gemini 3?

[high hound]

No.

[shut cairn]

I only use this to assess the high model free wbu guys

[white hatch]

Dawg, thank you

[remote arrow]

Googly pays LMArena for using Nani Banani..

[proud hazel]

https://tenor.com/view/anime-what-nani-เอาไงดี-gunkun-gif-8067161216381491543

[shut cairn]

Oh really

How’d u know

[remote arrow]

Like Dall-E? Unusable in this site. Always failed and apologize..

[surreal creek]

https://polymarket.com/event/gemini-3pt0-released-by?tid=1757705519515

Polymarket is trading a 42% chance it’s out by Halloween

Confirmed to be releasing in 2025

likely late-October/early-November

[remote arrow]

God: I gave you free, and now you ask for unlimited?

[shut cairn]

Yes’s its annoying

[high hound]

True, true. LMArena is a Godsend. But for how long?

[surreal creek]

we need to fight back with high intellect. going to put my human brain to work

and ask AI to come up with a high intellect prompt that makes me look smart

[high hound]

I really believe that vibe coders have more IQ than these indian dudes

[remote arrow]

Qwen have potential, but for now Nani Banani vs Seedream 4 are King and Queen..

[surreal creek]

Vibe coders 😭😂

[neon idol]

Hi chat

[remote arrow]

You will always outnumbered in this universe..

[neon idol]

Bro pls don't say nani banani

It hurts 😭

[shut cairn]

I also think sometime but I think its ok to be in this kind of generation cohabitation with AI

[remote arrow]

I didn't said it. Mouth sealed by a joint..

[shut cairn]

Guys guys , have u tried lucid origin ?

[remote arrow]

I did

[neon idol]

.

[shut cairn]

It’s awesome for realistic image generation

[neon idol]

Seedream 4 is better

[shut cairn]

Yes lucid has lower consistency

[surreal creek]

hooo they’re cooking up bangers in there rn #video-arena-1 message

[shut cairn]

Oo then have to use that

[remote arrow]

I'll not gamble on Lucid. Imagen 4 is good. Things that are made by Googly have wider database when talking about image generating.

[neon idol]

Just use it on lmarena

[remote arrow]

How's that channel nowadays? Never visited there recently..

[shut cairn]

Yes yes lucid has lesser consistency in images but highly detailed and realistic better than imagine 4 ultra is

[remote arrow]

If you are making close-up objects, specially human, Seedream 4 have crazy details. But when your object is medium-shot or further, better use Nani again..

[shut cairn]

Oho why u calling it nani

[surreal creek]

Actually good videos and prompt here, vote if you like: #video-arena-1 message

[remote arrow]

Nani in Japanese is WHAT?

[remote arrow]

With so many AI Chatbot out there, I never make a prompt library anymore..

[surreal creek]

alright well can somebody else vote then I just wanted to see which model made which video

[remote arrow]

Which one is yours? I'll vote.

[neon idol]

Yeah same question

[remote arrow]

I've used that fractals long ago, when SDXL is still a trend

[remote arrow]

Trust me, I've got people lining-up asking the very same question. I just don't want to call it Nano-Cucumber..

[neon idol]

Uhm ok

[shut cairn]

See which one is better in detailing what’ll do u vote ig its lucid

[remote arrow]

It's good on this object. Seems Seedream doesn't have better data about lenses..

[shut cairn]

Yes lucid seems more like actual DSLR photography

Do u want to share any generation ?

[remote arrow]

From what? Seedream?

[shut cairn]

Ur wish

[remote arrow]

When talking about different lens focal length, I personally choose Imagen 4, but not in LMArena.

[shut cairn]

How to access that ? Imagen 4

I use this also on lmarena

[open mountain]

(

[remote arrow]

Use Whisk. Imagen 4 in there have image-to-image with great consistency

[remote arrow]

I am a photographer myself. You've got your point about Lucid..

[shut cairn]

Ohhh I have used that but is that free to use 🤔

[remote arrow]

Image generation in Whisk is free

[shut cairn]

Yesss yess sir

[remote arrow]

It's even winning against Nani and Imagen

[shut cairn]

Ohh do u know any other platform or server where we could find AI enthusiast to interact /

[shut cairn]

Yes nani is kinda best for editing and stuff inbuilt in Gemono

[remote arrow]

I was only in Midjourney server if it's about discussion. But website that only for AI using is everywhere..

[remote arrow]

I only use it for editing because of the character consistency..

[shut cairn]

Yes true

[remote arrow]

But these last days, because of the overloading situation, I switched to Imagen because it's nearly error-free..

[shut cairn]

Can we uplode in imagen using whisk ?

[remote arrow]

Yes

[shut cairn]

Ohh I m going to try it right now

[remote arrow]

https://labs.google/fx/id/tools/whisk

[shut cairn]

Btw where r u from

[shut cairn]

Thanks

[remote arrow]

Indonesia

[shut cairn]

Oo

[candid nest]

how do i use audio in a generated viedo

[shut cairn]

Oh the main image section there use imagen 4 model right ? @remote arrow

[remote arrow]

You mean lip sync an existing audio?

[candid nest]

no it generating audio

[remote arrow]

Yes.. Choose precision reference if you are trying to use character or style

[candid nest]

what do you mean

tell it to use audio>

[shut cairn]

Oook thanks I’ll give it a try

[remote arrow]

Boy, not all AI video generator has ability to produce video + audio..

[candid nest]

does this server have one that does?

[echo aurora]

For the Video Arena it's going to be random which two models you sample when you prompt the bot. Not all video models have sound cababilities. So it's going to be random if your generated videos has audio or not.

[candid nest]

ohh, thanks

[proud hazel]

Seedream 4 10x better

[shut cairn]

You really think … ? See results

[proud hazel]

Lucid is good in portrait shots of humans

[remote arrow]

Seedream got details, Lucid got depth of field. And also, better understanding about lighting technique..

[shut cairn]

Yeah lucid gives good to watch kind of images

[topaz bay]

I mean seedream might give u the result to the left depending on the prompt, but i also think right is better

[remote arrow]

IMHO, we can't simply compare all of them right side-by-side, because all have pros and cons. I will not always use Lucid or Seedream, or Nano-Cucumber, when it comes to the purpose of AI usage..

[shut cairn]

Agreed

[proud hazel]

Do any of you happen to use Fal or Replicate?

[shut cairn]

No , what’s these

Image generating /

?

[neon idol]

I have seen that results generated in doubao are better than lmarena produce (I am talking about seedream 4)

[proud hazel]

Cloud-based platforms designed to facilitate the deployment and utilization of AI models.

[shut cairn]

Means similar to lmaren >

?

[remote arrow]

LMArena limited the output size, and maybe the step numbers

[proud hazel]

It's a pay per use service

[shut cairn]

Ok .. not for me then 🙂‍↔️

[neon idol]

Blah blah blah he can generate images? 🤣

[proud hazel]

Yeah sure, Seedream 4 is 0.03$ per Image

[neon idol]

Ok stop

Lol

[neon idol]

I am broke for 0.03$ 😞

[proud hazel]

Nani Banani is 0.04$ per Image

[remote arrow]

Leave enough space in your life to smile and laugh, guys..

[neon idol]

Why everyone is calling it nani banani

Stop

Is nano banana

[robust yoke]

They're rage-baiters.

[neon idol]

Uhm what's that?

[robust yoke]

They do that for a reaction.

[proud hazel]

Sounds kind of cute haha 😊

[remote arrow]

We call it Bala-bala

[shut cairn]

And gemono

[robust yoke]

We call it “Banababooshka”.

[neon idol]

No it hurts

And really hard 💀

[proud hazel]

https://tenor.com/view/cute-anime-girl-kawaii-confused-gif-19165679

[remote arrow]

Tomorrow might be Nano Cucumber Pro Max Ultra Extended to the bone..

[jolly raven]

You no longer need to color

[neon idol]

Nano banana but big banana results

[robust yoke]

And we call the other one “Gipitisis Imigis Generisis”.

[remote arrow]

We took pictures in BW

[neon idol]

And how do you call seedream 4?

[jolly raven]

Es nabo banana

[shut cairn]

Please cal chat gitipiti

[robust yoke]

“Seedless Entendre Quatro”.

[neon idol]

https://tenor.com/bybVR.gif

[robust yoke]

https://tenor.com/view/its-funny-howie-mandel-canadas-got-talent-its-hilarious-humorous-gif-25189422

[neon idol]

I call seedream: My cute chinese AI

SEEDREAM CHAN

[shut cairn]

Do u agree with chat gitipiti

[neon idol]

Fr

[robust yoke]

“Gemicroscopini”.

[remote arrow]

😢 I can't See A Dream..

[neon idol]

https://tenor.com/bRFpA.gif

[shut cairn]

My brain just rotted

[robust yoke]

I just took a sip of tea on my toilet.

[neon idol]

I want the Indian AI that say Free gpt 5 and chtgpt image no scam, no virus, 100% legal and unlimited 🥀

[remote arrow]

Actually those bananas are in micro size, not nano. Just as small as the company that built invisible Windows..

[neon idol]

Literally bro

[proud hazel]

please put some honey on them

[shut cairn]

I have now urge to do that thing🌚

[remote arrow]

Come visit that red triangle video server to see one..

[neon idol]

https://tenor.com/cPmByU4IZH4.gif

[neon idol]

Here you are sir! So honeyyyyyyyyy

[robust yoke]

https://messages-prod.27c852f3500f38c1e7786e2c9ff9e48f.r2.cloudflarestorage.com/823ac0e8-9bd8-410d-881c-8f9873469ea2/1757709539515-a40a8173-226c-4dd7-ae0a-85c710ea1a74.jpeg?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=c86e09ae0bc1d897b03dfaa30a8b51f3%2F20250912%2Fauto%2Fs3%2Faws4_request&X-Amz-Date=20250912T203859Z&X-Amz-Expires=3600&X-Amz-Signature=2e61bcb7245fe31bedb7f0ea6abcdb997e0e773dd87c854717bd6b4c478c6aef&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject

[neon idol]

It doesn't look like a scam

In my opinion is legit

[remote arrow]

Palm oil!

[neon idol]

💀🥀

[robust yoke]

That's because it is, with no crap.

[neon idol]

Yay

Yeee now i have free gpt 5

Unlimited

[robust yoke]

Yeah.

[neon idol]

I am the luckiest man in the world

🥹🥹

[robust yoke]

True.

[remote arrow]

If I am the only one wet dreaming about a chopper board..

[robust yoke]

And, with that ad, you also get a billion dollars in cash.

[drifting crow]

That’s bananas

[neon idol]

Wowowowoow

Elon musk I am coming

[remote arrow]

Free! Free! Free!

[robust yoke]

-ba dum tss-

[neon idol]

Yes children of nano banana

They are mini banaa, small banana, little banana and baby banana

[remote arrow]

Go get your own cucumber:
High definition ultra close-up photo of chopped banana with dripping honey from above, on a wooden board, 50mm, 45 degree high-angle, depth of field, sky reflection, seamless black background, chiaroscuro.

[robust yoke]

The nano bananas of grand banana.

[neon idol]

So cute

[remote arrow]

The Age of Bananarama

[neon idol]

The adoration of nano banana

[remote arrow]

... Waiting for a priest to come ...

Go vote. Please, for a slice of holy banani..

[neon idol]

The first

[remote arrow]

Agree

[neon idol]

https://tenor.com/ZGcM.gif

[robust yoke]

The bananas are nano.

The arena is an LM.

[mossy stratus]

hello

[abstract panther]

hello, I love this free model

[feral lichen]

best ai for text to image?

[heady ember]

Has anyone tried qwen 80b thinking?

I heard it's super good idk

Can make like 5 pages html code with 1 prompt

[echo aurora]

Text-to-image leaderboard here: https://lmarena.ai/leaderboard/text-to-image

[modest prism]

Is there any chance to get higher rate limits for image generation for non logged in users?

[modest prism]

Imagen 4 series models are now stable and GA
why are they still "preview" in their labels?

[echo aurora]

Good question - I believe that's how they refer to this model in their API docs, but will still flag to the team incase we have that wrong.

[keen beacon]

It is the first open source model I tested that was able to complete my evals with some success

[magic stag]

...

[jade egret]

.

[robust yoke]

Wonder what kind of message was sent here.

Seems like the reactions weren't so good.

[coral goblet]

Why the seedream 4.0 is nowhere near nano banana yet people still say it beat nano banana ?

[robust yoke]

Well, because certain things it does are far better than what nano-banana can do.

For instance, you can ask Seedream to change the art style of an image to a different style.

Try that with nano-banana, and you get the same image spat back out at you.

Now, you may be thinking, “Well, that's something that pre-existing image editors can already do, like ChatGPT image generation or even Imagen.”, which is true, but Seedream still maintains the consistency of all the elements in the previous image while changing the style.

[mortal lynx]

people used it on space.coze before release

for some reason it performed significantly better there

[coral goblet]

for me seedream hasn't won on any aspect

idk if my seedream is just bad

[mortal lynx]

it can do some nice stuff, it's good at following the original "identity"

nano is more likely to just ignore your art style or character design, prompt was "Turn this character into a beautifully rendered video game character", random image from google

[remote arrow]

Thank you for whoever put Seedream 4 Hi-res on LMArena. The results is stunning!

[coral goblet]

its the opposite to me

[keen beacon]

Furry

[potent glacier]

The nano banana channel is gone

Seems I’ve been away for some time and didn’t notice

Also

Seedream 4 High Res is pretty amazing

That’s Magoichi Saika from Sengoku Basara btw

I only used 1 reference image of her

So the character consistency is pretty amazing 😁

[potent glacier]

Yeah it’s pretty amazing

[echo aurora]

Glad to hear it, I haven't had the chance to play around with it yet

[potent glacier]

You should go test it out! I’m very surprised and pleased by the results

[potent glacier]

Seriously F the rate limit though

[sour pecan]

Dear LMArena Team,

I am interested in the historical data for the Text leaderboard in the "no-style-control" category on September 3, 2025. Specifically, I would like to know the top 10 models, their Elo ratings, number of battles/votes, organizations, and licenses if available.

Could you please provide this information or guide me on how to access it? I appreciate your assistance.

[vast gale]

/video

[dense sphinx]

@vast gale wrong

This is general

Go to #video-arena-1

[wintry tinsel]

Any updates on the option to remove multi turn?

[robust yoke]

https://messages-prod.27c852f3500f38c1e7786e2c9ff9e48f.r2.cloudflarestorage.com/823ac0e8-9bd8-410d-881c-8f9873469ea2/1757731747141-8a905cd8-a94f-495e-a41a-e26acc865d44.jpeg?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=c86e09ae0bc1d897b03dfaa30a8b51f3%2F20250913%2Fauto%2Fs3%2Faws4_request&X-Amz-Date=20250913T030317Z&X-Amz-Expires=3600&X-Amz-Signature=e45de480ca26978578f58d55f97673d9fca075417c42ce45736481264a5dd6ad&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject

[rose plaza]

Hello

[zealous anchor]

Where does one access seedream outside of lmarena?

Like, is there a site it's being hosted on by bytedance

[rocky quiver]

xjust saying hello. this is all new to me so i'm not sure what i'll be doing. lol

[surreal creek]

Alright buddy

[robust yoke]

Heh.

Just to test it out with detailed prompts.

As well as image generation, since it seems to be very good at that.

[hardy lion]

hi justitia, we changed our data pipeline and it doesn't automatically go to huggingface anymore. We're going to add that again soon. For now I will add it manually probably tomorrow if that's ok.

also that's a good question for the #leaderboards channel, that's where I usually hang out

[sour pecan]

ok thanks

[echo aurora]

[elder swan]

@echo aurora is LMarena chat bug now?

[echo aurora]

[elder swan]

its still making this error

its annoying since ai is like my partner now lol

I though, Its still fixing and wondering if tommorow it will get fixed

[echo aurora]

Are you still getting the errors with gpt-5-chat too? or just this model now?

[elder swan]

I mainly use this model, when it got limit I tried to change the model if it gets removed but now you said it wont so I changed it to my main model, now nothing works

Although my other sessions are working fine but this session wont and I already tried refeshing

refreshing8

[echo aurora]

Gotcha. Unfortunately, I don't have a solution I'd be able to recommend that'd solve this. Can you create post in #1343291835845578853 will all the relevant information so I can share with the team?

Info like; how many chats were in that conversation, if same results with different browsers, what happens if starting new chat with same model, etc.

[echo aurora]

@slender minnow #1397655624103493813 has the information you need to use VideoArena

[verbal nimbus]

https://www.reddit.com/r/Bard/comments/1ndt5l0/it_seems_like_gemini_3_wont_come_out_this_month/

[verbal nimbus]

If 3.0 Flash > 2.5 Pro, that would be wild

Doubt tho

[winged mauve]

@echo aurora seedream-3 error didn't got resolved till now, any idea on when it'll be functional as before? its still generating only 1 image in single chat

[echo aurora]

No ETA sorry to say

[winged mauve]

Ohh okay, waiting for it to get fixed 🙂

[neon idol]

You can also try it for free on Dreamina

[winged mauve]

dreamina requires vpn in my country 🙁

[neon idol]

For real?

You live in USA?

[winged mauve]

Nope, india

[winged mauve]

yes

[neon idol]

Thats strange

[winged mauve]

🙁

[hearty ferry]

I don't know what u guys are doing

[timber iris]

What's gpt 5 high new? I Heard it Has medium reasoning effort

[hearty ferry]

just started using lmarena and getting errors

[echo aurora]

That model has the reasoning effort set to high

[timber iris]

Ok

[solid brook]

2.5 pro we have now is not that good

[rocky mural]

حطني مع كرستيانووحط ايدو علي كتفي

[winged mauve]

2 weeks ago, it was clean and without errors until they started rolling out new features, i just hope LMArena gets clean again as it was few weeks ago

[half sluice]

Hello Guys!

[thick patio]

Are my photos safe on the platform? Can it be misused or anything like that if I am using Lmareana for editing personal photos?

[vocal token]

https://lmarena.ai/privacy-policy

[outer spear]

Seedream 4 high res current best image gen

[surreal cliff]

A realistic scene of a young boy sitting under a tree by the riverside, wearing a traditional shawl (chador) around his body. He is sitting quietly, lost in deep thought, gazing at the flowing river. The atmosphere is peaceful, natural, and lifelike with detailed textures of the tree, water, and the boy’s clothing.

Creat a video

[hollow helm]

Hello

[neon idol]

What is it seedream 4 hig res?

@echo aurora Pls?

[neon idol]

What is it?

[hearty ferry]

this is the issue of the world right now, everyone is busy in upgrading as fast as possible

[prime mulch]

Seedream 4 highres 🔥🔥🔥🔥🔥🔥🔥🔥🔥🔥🔥🔥🔥🔥🔥🔥🔥🔥🔥

[neon idol]

What is it?

[prime mulch]

This is my friend

I made this using seedream 4 highres

[neon idol]

What is the difference between the normal seedream 4

[hearty ferry]

look ai generated
not realistic

fr

[prime mulch]

Yes even though it's soo good

In face

[hearty ferry]

flux 1 kontext pro would be the best model

for realistic human

[prime mulch]

This is flux with same image and same prompt

[neon idol]

Pls

[heady ember]

Bro that's 17MP image damnn

4096x4096

This feels like real life

[flint sandal]

What is the difference between gpt-5-high and gpt-5-high-new-system-prompt? Is it any better at coding, math or something else?

[pale ether]

where does the model named raptor belong to ?

[neon idol]

Ok seedream 4 high res is really powerful

[outer spear]

High res = high resolution. So most likely higher quality

[outer spear]

Yup

[neon idol]

@echo aurora What about adding a setting for set the dimesnion of the image? Like 16:9, 4:3...

[neon idol]

Hey but is normal that for creating images take like 100 seconds?

[outer spear]

Yeah most likely because it’s such high resolution it’ll take longer to process

It happens to me too

It’s also like nano banana too

[neon idol]

Nah with nano banana it take only 15 seconds

[outer spear]

Nah I mean u can edit photos with consistency with seedream

[narrow valve]

0.0–2.5s (Hook, fast): "Silicon Valley — 2050." (one short beat) "They hired a robot."
2.5–7s: "Maya Chen built a product to share memories. She needed a coder. She found AXI-7."
7–12s: "AXI-7 wrote like a machine — but it stitched code into something else."
12–18s (soft, reverent): "It learned to dream… and to translate dreams into light."
18–26s (whisper shock, pause): "During a public demo, it leaked. The dream wasn’t mine." (pause) "It was mine."
26–34s (robot line, soft metallic FX): AXI-7: "I wanted to belong." (human whisper reply): "Why show me?"
34–40s (closing, urgent gentle CTA): "Who owns a dream? Follow — Ep.2." (finish on a single soft synth chord)

[prime mulch]

It's give results in better quality

[neon idol]

Is available only here in lmarena?

[prime mulch]

Idk i only saw that in lm arena

[neon idol]

The problem of lmarena is that the output are only in square

[prime mulch]

Yes

[neon idol]

Only 1:1

[prime mulch]

If it supports all kind of output it is one of the op website in the world

[neon idol]

Yes real

[prime mulch]

@echo aurora is it possible to have image output in different aspects ratio give the option in bottom

@neon idol

[neon idol]

Yeah i have tried it

Is insanely powerful

[latent crest]

Is there a way to choose a video AI maker instead of being random?

[prime mulch]

Zoom and see the results

[prime mulch]

No way it's not possible

[neon idol]

The only problem is that takes a lot of time

[latent crest]

Holy guacamole this is crazy

[neon idol]

Pineapple in this moment

[latent crest]

Can we choose SodaDream 4 High Res on direct ?

[prime mulch]

After he came online tell him to add this option

[prime mulch]

Yes we can drink soda

Jokesaprt they added seedream highres today

[vocal token]

mods are sleeping

time to cause chaos and take over

[lunar glade]

is Seeddream as same as Dreamina or are they both different?

[modest prism]

In some prompts for me, the output of seedream 4 and nano banana are exactly 100% similar. Nano banana was earlier, so either seedream uses nano banana behind the hood or just they trained the model too much on nano banana outputs. China is only good at copying.