#general
1 messages · Page 336 of 1
nano2 to compare
Pineapple Inc
this one looks good but i can tell its ai.
the gpt? nah
💀
We 100% in a good checkpoint rn
I think we in duct tape 2
imagine tomorrow's checkpoint
threw another one
@silent tree
thats crazy
Next is u
AHH
U haven't throwed our powerful weapon yet..
The Banana 🍌
dude my generation stuck in one last tweak
Skill Issue
did oyu add the caption? or came with caption?
(tiki tiki anime sigma pro 100% 2026 approved background song playing after u said that)
i told it to add it
wym
is gpt image 2.0 in the arena yet?
no
apparently it's even better than nano pro
@echo aurora approve?
It's better in some ways and worse in others
even got our main title
its crazy how its so consistent with the text
cherry looks so chopped dawg😭
thats what im saying
the two heads ruined it all
damn i ran out of ideas
I have some ideas
It’s worth hoping that DeepSeek V4 comes out this week regardless of Kimi k2.6 and GPT 5.5's benchmark performance. Creative writing in models like Kimi k2.6 is still poor right now. DeepSeek's priority is real-world performance over benchmarks
If fortnite collabed with openai (fortnite shop)
If fortnite collabed with Rockstar Games
Introducing Los Santos In Fortnite as a map for limited time.
That's a great idea lemme do it
dead chat
what the hell
fortnite in gta 5 map
wild consistency
it got trained on fake iran propaganda what do u expect
not good 😔
Gpt image 2.0?
why u do this to me?
We're definitely in a good checkpoint
its not that good tho
if the choppers were hovering a bit lower it would be better
but the text consistency of thsi thing is actually nuts
it just doent fail
image edit sucks lol
BREAKING NEWS!! ⚠
PEOPLE IN THE VEGETABLE COUNTRY HAD SPOTTED @echo aurora PINEAPPLE'S MISSILES, AND TOOK PICTURES, THIS MEANS WAR BETWEEN THE VEGETABLES AND FRUITS!
If u look closely those are pineapples lol
how
gpt image 2
You fooled me
which one
gpt image 2
chatgpt
it dropped there already damn ?
yeah
bet bett
oh god it mesesd up ai studio
still looks good imo
what is nano banan even doing
Is this what I pay 200$ and can’t use it?
meanwhile gpt users enjoying
I have the small plan but u think I’ll take the 200$ one
it uses web search
openai hq looks good and clean today
I think It isn't nerfed
it squantized imo
Well today at morning was bad but that was the worst checkpoint
Is Slop Now
i doubt it, it probably just has very good and recent image gen world knowledge
it should have web search though because the llm has the intelligence of a 35b model
Gpt image 2 is stupid
Is Nerfed Slop
It makes good images but its not smart
I've been saying this ever since it first landed on arena, it's honestly stupid
All images I've generated today:
in fact, it is so stupid it can generate things the llm doesn't know about
someone needa explain to me why duct tape 3 was better than the official release
Bro is obsessed witht the pineapple
ts tragic the sharpness is on the same level with wan2.6
gpt image 1 mini had the same thing going on with it, but it's more extreme here as the llm is dumber than 4o-mini
its 10x smarter than any other model
Nah its not
im actually so surprised
Gabriel Goh, Kenji Hata, Kiwhan Song, Alex Yu, Boyuan Chen, and Nithanth Kudige join host Sam Altman to introduce and demo ChatGPT Images 2.
the quality was so good in the testing models
Even the tiny versions of qwen are smarter
এই ছবির মাঝখানে এই লেখাগুলো বসিয়ে দেও,,,
“Opportunities lie in the place where the complaints are.”
👉 যেখানে অভিযোগ আছে, সেখানেই সুযোগ লুকিয়ে থাকে।
it's only a matter of time before qwen overtakes gpt
flashing it in the skkkkyyyy
hi guys
wsp teto
yeah the 35b is almost same thing of qwen 3.6 max
wsp beanbag
the sky
🔥 w response
the 3.6 27b will be the same thing of qwen 3.6 max
codex lab is out too
AHHHHHHH
Is it on API now?
Text rendering already good
claude release an image model and save us
gpt image 2 unfortunately does not know the super mario world nes bootleg. gave me this instead
gemini would !
@echo aurora Good evening, Moderator. Would it be possible for you to share any information regarding whether GPT Image 2 is scheduled for release following the conclusion of this live stream? Any clarification on the matter would be greatly appreciated. Thank you.
that's flux's dude lol
Gabriel Goh, Kenji Hata, Kiwhan Song, Alex Yu, Boyuan Chen, and Nithanth Kudige join host Sam Altman to introduce and demo ChatGPT Images 2.
Its crazy to me that deepseek is so good at math but like average in coding. Coding is all about math
this was gemini's personal attempt when i first tried it @polar horizon
not the worse but its bad
YESSSS
idk anything about mario but this is pretty good
They arent going to because they cant, he'd probably lose his job if he actually said something
Oood eveninO, Moderator. Would it be possible for you to share any information reOardinO whether OPT Image 2 is scheduled for release followinO the conclusion of this live stream? Any clarification on the matter would be Oreatly appreciated. Thank you.
this is how its supposed to look
I have a chance to be banned in qwen? lol, in a one message I send to qwen create 39 images
whatever who knows
Now please release it on Arena AI in Direct Chat
Pls
😭
I want this Bruh.
SPUD IS TYPING
anime boys
how do you not know anything about Super phockin' mario
EVERYONE SHUT UP
no
if these dudes release it on battle mode only im killing someone
im a newgen sorry😢
Where can i currently access gpt image 2?
bro im crine
Its likely taking longer to release because they are probably putting heavy limits on it, also they released opus 4.7 legit like 2 hours after it was announced
bro doesnt know abt all night nippon super mario bros
yes
So it takes time
to this day ive only used that 5 times
tragic
Its in battle mode only and recently released so I'm not surprised
they starving us
qwen3.6-35b is so much smarter than gpt image 2's llm that it can correct its mistakes
Because one model is a general purpose model and the other is an image generation
finna tap into ts
Two completely different purposes
WAIT TOMORROW DUD
Gabriel Goh, Kenji Hata, Kiwhan Song, Alex Yu, Boyuan Chen, and Nithanth Kudige join host Sam Altman to introduce and demo ChatGPT Images 2.
tomorrow?
I cant wait till seedream is good enough to make full on animes at high quality
I meant like for arena
why would anyone want ts
see dance you mean
Renatherena
prob 2 hours after the annoucement
Will GPT-5.5 be released this week?
so not tomorrow
prob
every day i get reminded that there's a difference between casual ai users and pro ai users
GUYS. GPT IMAGE 2 DISCORD BOT SOON
Neither.
UNLIMITED
it's an image gen model that is supposed to be a direct replacement for its predecessor, gpt image 1.5, and a competitor to nano banana pro
it should have at least the same, if not better, world knowledge than at least gpt image 1.5 but it's clearly much worse
What are you even saying? Millions would want it, you can make animes based on your own ideas thats great
accepters get the old duct tape.
You're clearly very casual
thank GOD mane
I pick Banana
Just a Normie
we gon kill the value of anime so hard😢
watermelon 🍉 or banana 🍌
2
2
2
🍌 Banana 🍌
i noticed that ever since they removed the flagships, a ton of image models appeared in direct
Thats not really true, it depends on how you train it
67
tragic
Just like everything else generative AI touches, this would be nothing new
that's why the quality of things going down now
mane i need to rethink my life
As in what exactly
i might go anti-ai chat🥹
Bro's just in Denial Mode.
no
AI is the industrial revolution
its the same thing
Good or Bad is Inevitable
Okay good for you, hop outta here because you clearly have no idea what you are saying right now 😭 please get educated a bit
are we serious
illustrations and shii
Pinterest is filled with slop now imo
Bro it's obviously the last one
big dawg
its getting better
gpt image 2 is the big hand
we haven't even used gpt image 2 yet bro
the start is hell
Slop Better then you could ever create with your Hands.
i agree lmaooo
I can't draw for shi
if you cant draw then dont post ai slop in pinterest
Images would be on pinterest not illustrations but I mean who cares, you cant put emotion into a writing, all you need to do is to be creative
simple
imo all the Chinese models are gonna overtake these guys
bound to happen
when they stop distilling from claujde
i will belive
unfortunately people don't abide by that
I don't post slop
New pfp is it tuff
I be making slop for fun
w chiikawa
Yes, but thats due to distilling from US companies models, but open source models will win eventually and its already insanely close
guys im making a refugee server where you can use image 2 on a discord bot
do u think its gonna get it this time
Don't forget we get it for Cheaper because they distille
no.
Absolutely not as long as claude is better keep doing it lol. Anthropic is a greedy ass company anything that comes is deserving to them
yall dont realise this can generate transparent images. free asset generator
Exactly
GO COMMIT DIE
I SPENT 15 MINUTES WRITING THAT PROMPT 😭
GPT has always been better at doing transparent stuff too, it was never perfect but it was better than the other models
LMAOOOOOOO
Make it copyright Free
no.9 by Carti hard ash bro
hit the image gen limit
💀
new account buddy
pro users having infinite generations 🤣
copy it again then send it into another ai
They don't.
😭
i do
here's the prompt:
Generate an image based on the following instructions:
- The image itself should be based on Super Mario World, but NOT the SNES game. It should be based on the NES/FAMICOM BOOTLEG of the same name released in 1995.
- Should only showcase the title screen itself. Low-quality 8-bit pixel-art graphics with limited color palette.
- Image elements: Blue sky, lime background hills and grass with blue outlines matching the sky's color, light skin-colored dirt. Two white clouds with a thin black outline and blue eyes. Top of the screen's logo should say "SUPER MARIO WORLD" in the game's usual logo font. Red, yellow, green and white logo palette with a solid black drop shadow oriented towards the bottom-right. At the bottom left of the screen, should be Mario riding Yoshi based on their sprite from Super Mario World, but with a more primitive color palette and generally lower quality graphics.
- Surrounding border should be based on an early version of FCEUX. All top-bar options, FPS and frame counters, and the ROM name (Super Mario World [Unl] (!).nes). FPS and frame counter should be at the bottom of the border instead of the top. The border itself is in the style of Windows 7, retaining its classic aero aesthetic.
- Additional settings: 480p reminiscent of early-2010's YouTube videos. At the middle-top, the old watermark for recording software Bandicam. 4:3 resolution, yet with visible poor cropping around the edges.
Not my friend
i do not care enough to pay for chatgpt
does gpt image 2 have thinking?
no he's right
i use it for the love of the game
it uses 5.4 thinking rn
so if it doesnt have thinking, its not gpt image 2?
IT DOES HAVE THINKING
The lion does not concern himself with paying for artificial intelligence.
thanks for the ideas
if you not making money off it why tf are u paying lmaooo
i hope you read what i've said, cus i was just trying to clarify
you gotta be dumb bro
Then why you spent 15 minutes writing a prompt
cuz i wanted it to look good??
it uses 5.4 thinking to write the prompt, then sends it to imagev2
look at bro's pfp
fahhh i needa pee
what in the actual heck is going on in this chat
hey
solve this: Solve for f(x) ∈ L²(ℝ):
f''(x) - 2x·f'(x) + (λ-1)·f(x) = 0
where λ = 2n+1 for some non-negative integer n.
Then compute:
I(n) = ∫₋∞^∞ [f(x)]² · e^(-x²) dx
and show that:
∑_{n=0}^{∞} I(n) · r^n = 1/√(1-r²) for |r| < 1
Finally, verify that the generating function satisfies:
∂²G/∂r² + (1/r)·∂G/∂r = G · x²/(1-r²)²
you would sell your soul to openai if it had contracts for that
I'm homeless but i have Gemini ultra subscription
If you get your money back yes, because any money you make from it is going to get you 10x more than what you pay lol, its like an investment
who needs a house when you can vibe code a house
actually the best use of money is buying chips 🤤
🧠
not worth
you could've spent that money.. on a house
nah dude the new deep think got me a gf
its comparable to 5.4 xhigh, not even 5.4 pro
Sigma of The Year is @waxen grove
welllll
now that's what i like to hear
specify that it's an ai girlfriend next time dawg
racism to clankers fr
*σ of the year
it's a girlfriend
agi is meaningless word
mythos sucks
what's dis
mfs who GENUINELY (Unironically) (No Joke) (On God) pay for ai when there's a guide on how to spend things wisely upstairs
Can't you read?
ok what's the current best model
now I'm legally blind
Slopus 4.7
how
Guys i found the best bypass method for GPT Image 2 in gatekeeping (u didnt even invent it lil bro)
it's not that good
That's why i call it Slopus
slideeeeeeee
no.
DMs
It was forgotten
they literally
bad
Ik
BY 50%
I use it on Ai Studio
same dude
And Nerf Again
the app is genuinely trash
Ai Studio is not Nerfed
so is gpt 6o spud
Whats this gang
I'm going to wait for Gemini 4
Nick Wilde
fortnite bundle
wait for gpt 6o spud
what the hell is that
gpt 6o on top
Gpt 6 will be ass
nah its omnimodal like 4o
nah itll solo mythos even more
So its gonna be good
Mythos is too op bro
5.4 pro xhigh already beats mythos
mythos is 5.4 pro standard tier
It ill smash gpt 6
based on their own benchmarks
mythos preview*
I call cap
"preview" = fp16 so before nerf to make it usable
👍
i call do basic research dummy
Insane ngl
guys guys
Nah but mythos literally beats gpt 6
mythos loses to 5.4 pro
6o will mog mythos
Mythos is a Myth
Gpt 2 image not released yet in arena??
Spud Mogs All
I do
is mythos in the room with us rn?
It doesnt
Wait.
Mythos destroys 5.4 without even trying bruh
Bruh What!?
joku
Good One
aiku
LMAO
5.4 pro is the assest model ever even grok is better dawg
no
not that far
It is tho
Its not a myth its real
mythos
What are you guys on about
🧠
its a joke to please investors
bro pineapple u gotta chill out
Bro is obsessed with the pineapple lmao
lmao
pineapple, add new GPT Image 2
He can't.
ah yes pineapple will add it 💀
Thats femini
banana
What
?
Are you a pineapple copycat
look at those pineapple missiles
no im a part of his club
Product
Prouduce*
Produce *
Cap
no cap
Yes cap
How did you do thid
it's a website
called your ai slop bores me
instead of ai
a human replies to u
How did you get thst role dawg
Idk
what did bro produce
Cap
no way u dont know whats produce
You just lucky
produce
uh
How are you guys produce role
uh
Me banning spam bots
Yea and it got my cousin's cousin's daughter's cousin's son's cousin Brocolli house destructed 💀
hmm
ANOTHER MISSILE HERE PLS
IM CRINE
BOOOMMM
BOOOM
Crypto scsm in the big 26
Lol
I genuinely think that wasn't a bot
The Pineapple Missiles are REAL talk.
and just a joke
Lol welp
which bot would start a conversation by
hi guys
bro.
I think it was a hacked account
Hello 
probably
Are you a potato chips
no i am spud
nice to see you pineapple
You as well
this chat turned from ai to fruits
AND vegetables for some reason
AND spuds
@jade topaz
yes
and a Nigerian Einstein
damn
wow
what a difference
Js now
nooo he's going to burn my gpu
What is yappingAI streaming
I guess yapping?
@surreal zephyr
look
its an anthro
"yappingAI"
like your ai doesnt want money every single time
💀
Anthropic 💀
I m Arena Champion?
@echo aurora How do u even get arena champion
You need to be a six seven fan
i think i finally got access to gpt image 2 in chatgpt website
Wdym finally
Its been out singe like 10 am
i tried it earlier today and it was still gpt image 1.5
like 1 hour ago for me i still had 1.5
i guess it's gpt image 2 because it supports 21:9
Are you in belarus
nah im in norway
Thats crazy
Sure are
Will find the announcement, one sec
Livestream starting...

k dms
OMG
link?
It released rn
gpt image 2 is either cheaper or more expensive than 1.5, depending on resolution
also btw image 2 can do 4k, up to 3840x3840
not bad
available in chatgpt web?
instant and thinking? can't wait to force instant to give me studio ghibli art since the thinking version is nerfed
it's been available on chatgpt
yeah
not for me
all the models disappeared tf
they're probably adding image 2 right now
also fal has it now (or at least it should pop up in the next few minutes)
actually in arena
Ooo
Is it just medium for direct chat?
where is high
looks like it
refresh, it should be in direct chat
also gpt image 2 medium is duct-tape-2
where u found this
GPT IMAGE 2?
yeah
and btw @echo aurora openai just called arena LMArena in their livestream
OMG
where u found price
openai api docs
goat
link pls
Oh we know
"LM"Arena though 😭
they deadnamed you
Hey does lm arena removed ai video option 😭
It's not showing in my website
it's like twitter changing its name to x, people are still going to call it by its old name whether you like it or not
WE ARE COOKED
Should be able to find Video Arena here: https://arena.ai/video
True true
mb ts was corny
Thx now it's showing
what about high
@echo aurora images generated by gpt image 2 are called "duct-tape-2" in the file names
'something went wrong' 🥀
guys any difference?
gpt 2 is more realistic
?
😊
prob rate limiting
Likely going to be rate limit, lots of demand at the moment
alot of people using it
oh
almost the same, the duct-tape 3 have some shaders on
i forget about rate limit
gpt 2 really cool
not nerfed at the moment.
I sent my application
Xd
Wish me luck guys
I think the original image 2 is better xd
same
i mean i think its arena
Maybe they'll add a high version cuz it's medium rn
we serious
like why arent some models no longer aviable in direct chat
look into it
its because of cost
left is gpt image 2 arena?
@cedar citrus
Yeah
ok
well it looks better
which one is which
fk you gpt keep on saying there's an issue
Left is arena right is original chatgpt image 2
actually i like arena better
Same lemme see others that ive generated using it
is left one arena or chatgpt? for me left one is much better
oh
Left one is arena
yeah i figured out already
"Handheld picture taken from a spaceship, zoomed into moon. Close to moon"
Left is Arena, right is Chatgpt
Which ones better
both are great
right better
both are cool but i kinda think right is a bit better
"You have reached your rate limit for gpt-image-2 (medium). Please try again in 51 minutes"
I didn't even use it... it was just giving me an error
I am crine
OMG
bro it's so good model
send this to israel
first
dming it
no way
Lets avoid this subject please
the arena model is better than the site apparently
how we ALL be rn
alright
gpt image 2 is nuts
ok
How is gpt image 2 so much better than anything else
@echo aurora stop baiting me! Every time I have pop-up from this discord I'm excited for the new usage system and the models coming back 🤣
Seem we had the medium version
Okay buddy
Very funny
- 252 elo points, that is massive
damn
Left is Arena, right is Chatgpt
"Handheld picture taken from a balcony in night zoomed in camera, there's 5 pineapple missiles going in the sky.", which one looks better?
PINEAPPLE MISSILES AYYY
Why does the arena one look kinda better
God damn
We're still going to post announcements lol 
its using low
It doesnt look like a pineapple tho
Chatgpt looks better
On arena
I mean those are definitely real missiles but I asked it for pineapple 😭
gpt images 2 sucks, still can't generate a postal 3 screenshot
They look like real pineapples
Is it erroring
what is that
oh cmon
yeah i can't generate anything
Okay guys stop with politics
havent tried on arena, but tried on chatgpt, it sucks
wow is nano banana rly obselete now????
someone should gen half life 3 title screen with gpt image 2
Gpt is better at generating, but Banana better at editing
a bad game
it makes postal 2 screenshot instead of postal 3
infinite inference for free? because if so gpt img 2, otherwise nano banana just because i get more usage
not according to the new arena leaderboard benchmarks
ok thats actually good, their is no refrence image behind it?
YO SANTA?
typa of stuff i see in my history class ah
Why gpt 2 in arena limited
someone do half life 3 title screen
use websim bot
How can I get unlimited
web sim bot
I asked claude to review if it's ai or human, But Claude said it was ai, BC it's too perfect
thats an insane gap wtf
How?
im DM you
now what was it?
Yeah pretty wild
gpt image 2 can perfectly replicate the 4o-style ghibli outputs, you can guess what i'll try to do with this information...
what about image high??
Thanks for deleting that inappropriate image, also, why does the arena gpt image look way better 😭
Like Pineapple, this is CRAZY.
People tried so much stuff to make a image like this years ago
time to wait for nano banana 3 i guess
mind, this didnt error, first try
whats the prompt?
quick question, when using direct chat with Max, can this still include models that are normally restricted (like the Opus models)? Because you can still get some anonymous votes/data from it, using direct or indirect signals; which you can then compare
nope
Handheld picture taken from a balcony in night zoomed in camera, there's a Santa Clause on his Sleigh spotted but u can't see him properly
what I thought
chose wrong size by accident
thanks
xd
OH NO PINEAPPLE SANTA SAW ME
W
Creepy vibe
looks like a fever dream
Gn guys ima go sleep
gn
Fr
Its very likey opus and gemini pro will return in the credits update, which could be within weeks maybe.
however, their is always the chance they wont come back at all
I'm fine with that, honestly
DISCORD BOT FOR IMAGE 2 RELEASED OMG
Deleting what
then again, grok is good enough for most stuff
Completely understand why you would take the good models away, it's a money burning machine after all
I am fine using minimax or else for coding, it's good enough for me, and I have a brain I can use to do the actual thinking
also true
though I won't use it because of obvious reasons
which reasons?
its 2097. openai released gpt-image-12, officially beating micro banana 3's 6 year reign. ai can now make images metaphorically indistinguishable from reality in a lapse of ~462.87761ms. nothing is factcheckable. nothing is recognizable. we are done.
Love to see it
its because its medium
well 1st because (and I know this is indirect) it's from elon, which sure he's had great ideas in the past, but some actions are questionable; the "recent" explicit photos generation thing, and generally I don't like their philosophy
Huhhh in what world is this nerfed lmao
Right thats bound to happen no matter what with so much happening in the picture
tbh that stil looks good if you would have shown it to anyone 2 years ago and said "ai made that", you would be called crazy
yeah thats fair
Go to the article where it says when u get an error do this or whatever
I used claude in the past, now my account was affected by their "accidental" mass ban, I guess I won't use them too, my current option is just using duck.ai, arena when it's not a context-rich query or other free inference
i dont follow. what is that?
whats websim
it will be years before that ever happens, the largest open image model is 80b parameters, image 2 could be in the trillions
websim i know, but idk what bot
i will dm you both
I would say 1-2 years if you're optimistic, and the boom hasn't died out by then, which I think will happen sometime this year
How to get mercury ai by inception for freeeee
dmed
Arena has it for free I think
does anyone know?
i think high isnt their due to high end models being taken off
(i may be wrong, just a guess)
official site
Ok thanks
how different is training a text model vs a image model? I guess it takes a much longer process and much more computing for image models, and that's why open weights ones can't keep up
Does anyone have an ai that generates unlimited pics for free
it's not really that, it's the fact that:
- image models can make more money than text, so big companies like bytedance keep theirs closed
- beating openai, google, etc. in llm is more important to chinese companies than making image models
- training is a bit expensive, so they tend to make smaller models (which is good for us since that means we can run them locally)
- the companies that tend to make image models also make video models and they don't want to open up either to boost their own platforms (minimax, kling, etc.)
qwen hsa a really good image gen model but they didn't want to open it up, thankfully baidu gave us ernie image
W
lol
GPT 2 / Nano Banana Pro
Which one is better? (same prompt)
what prompt
Hey guys
wth
How good is chatgpt new image model
no
How
The Video Arena is currently accessible only through: https://arena.ai/video. More information on how to use Video Arena can be found in this article.
The Discord bot was removed from the server and is no longer available. This announcement has more information about the removal.
Wtf is ts
yes buy
what bot is this?
LM Arena Tracker
how can i access it
I forgor
i can dm you a link to the server which has it
ok please
why did it cover the username
cant properly do microsoft chicago
img or txt
txt
this keeps happening
hm for me ilium_1 was saying simmilar to that but also said that it was founded by a team of engineers from openai 3 times
GPT couldnt do this.... very censored
this is nano banan pro
hm really
If you could only use 1, and never use the other again.
10
11
2
GPT Image 2 Series
Man I remember I played the game with that space guy
I forgot it's name though
Failed Generations aside (as nano banana often is broken)
Only consider the final output
13
14
2
GPT Image 2 is better anyway
Bro 😭😭
good times
Season 3 and 4
gpt image 2+seedance 2.0 = cinema
I think the cube is a little.. obese?
Yoo tell me the name of this game
I played it but I forgot it's name
Kerbal space program
Ty man
"not even gpt4o could do this" 😭
fr lol
gpt image 2 actually so impressive
lol
there's thinking and web-search ones
what are those names lol
I'm not gonna be surprised if they add something like named sigma
might be grok 4.3
yeah and it would be grok 5
Another one coming named “Bitler “
😭
H really said to b to go replace it
Trust me it works
gpt image 2 is pretty insane