#general
1 messages · Page 393 of 1
Good evening everyone
Please how can I generate a video here 😢
I think they removed that feature
Omorr 😔😭
I think I have it in my server (but it's like private) if you want I can give it to you
try generate video in their official website arena.ai
well pretty much very limited
I think he/she hasnt access to the site but that works too.
well, he/she didn't, lol
you may need to sign up to arena.ai to generate video
@oak flume check your dm
@rigid copper just signed in
Not seen where to generate video. It's literally acting like chatgpt
With more thinking budget
It looks
(The code
really bad
XSS in Error Page
Let me do a pull request so you can update
nah no way it is
o1 and o3 pro were different
maybe more thinking budget but
context recall is something else
Its literally same model
Just better quanitzation + more thinking budget
Information on how to use Video Arena can be found in this article.
They updated it a bit after 5.5 came out
job
Qwen is better than codex at coding now?
🤔
uhh cuz 3.5 flash is so much more agentic and better at coding lol
top 35
oh
I would like to know daily Limit of chat gpt 5.5 in arena
???
We're going to be doing this for a bit. Once 24/7 mod protection is in, we'll bring it back during off hours.
No, I understand no worries
Hello pineapple
Currently, with all of the limitations there isn't good visibility into what that is. The new usage system we're planning to add will provide that visibility into the limits.

Hi @echo aurora I made a simple app for arena i sent it here and got deleted somehow is it a rule that i dont know or i should send it somewhere else?
Can you tell me a bit about it in DMs? It's possible it wouldn't be something we're able to endorce and therefore wouldn't want it being shared in this server.
sure!
Yo pineapple have u ever played cod
@echo aurora I sent the app in your dm
Battle Mode on Arena is the only game I play 
😭
Best img model rn?
Gpt Image 2
Would encourage you to check out the leaderboards https://arena.ai/leaderboard/text-to-image 
oh u meant in arena
you may be able to use it through max router in image gen
Hi folks. Has Claude Opus disappeared?
Some models have been removed from Direct and Side by Side mode. You can find these models in Battle.
Is possible select these models in side by side mode?
I'm hating qwen-image-2.0-pro-2026-04-22 rn
no
they're only in battle
that way users can't select them themselves
because they're hella expensive to serve
No, the Opus models were removed with Side by Side and Direct, you can find more in this announcement: #announcements message
ugh I hate the direction AI is going in
Guys, does anyone know which model in LMA Arena is best for programming other than and something with no limit?
I've concluded that both flux max and qwen image whatever are both terrible
TO EVERYONE, I BELIEVE THEY WILL ALL COME BACK BEFORE END OF 2026
HAVE FAITH AND LET @echo aurora DO HIS/HER WORK WITH THE CREW, ALSO SORRY FOR CAPS
ITS OKAY FOR THE CAPS ITS TOTALLY OKAY WHEN ITS DONE FOR POSITIVE MESSAGES LIKE THIS, THANK YOU!!!
WHY ARE WE ALL SCREAMING
AAAaaaaa
lol
damn gpt-image-2 be making hella fine gens
||jokes aside have fun man!||
its ai
playing minecraft
I did not look at the numbers long enough
Hi pineapple , i read that the team is also working on to make a way to share context of a thread which has run out of its limit to a fresh new thread . How's that going ? @echo aurora
is this ai
wtf are you prompting it
legit impressed
trying myselff
blud tried too hard
that looks AI
working on to make a way to share context
I can't say I'm aware of these efforts, where did you hear this? We are working on a new usage system we're hopeful will lead to us lowering the current context limits. I assume this is what you're referring to?
it's all so wrong
this isnt AI
y'all lying to me atp
?
Damn these models
So top model are top models because there top models ?
Idk why qwen doesnt keep qwen3.7 max as its default
oh this one isn't terrible!
@echo aurora dear it's them again
in all the channels
😭
Pineapple pfp is so cute!
yeah but $$$$$$$$$$$
theres a method to making images look so much better
chatgpt adds a watertmark btw
this one's chinese anyways
They're just trying to give us all free money, whats the problem?!?! (sarcasm this is clearly a scam and I wish they'd go away)
gpt-image-2 would nail it
I love your pfp very much!
Omg my favorite fruit is on
pineapple glaze moment
its like 2 cents to generate an image
"actival meter"
??
oh that got botched in the 16 turns of editing lmao
Its valid tho
Try watermelon
Thank you! Fun fact, it was made with google slides.
real
meh
Which watermark?
LMAO
Damnn
Better than ai
Under there
Ain't no way ur comparing pineapple and ai 😭 ✌️
Not my point
thats like seeing the Mona Lisa and saying "wow, better than ai"
synthid as well no?
Pineapple is more useful than ai
You can eat it
Deep dive into extracting and attacking Google DeepMind's SynthID image watermarking system using adversarial machine learning techniques. Learn how invisible watermarks work, explore LSB steganography, train a surrogate CNN detector model, and test regeneration attacks and C&W L2 adversarial perturbations against the real SynthID validator.
Same thing here
Holy chat flood
Open ai probably uses same methods of watermarking
Love it when Arena puts a hydrogen bomb against a coughing baby.
haven't seen my baby opus in so long :'(
4.7 opus is still there?
I think opus us gone right
Is*
I just don't use battle mode often
It’s really way to see the watermark in those images
Damn so battle mode still has opus in it?
Crazy
I do mostly side by side
real the noise
I do not see it in here honestly
That’s what the pattern is on those images that ChatGPT2 makes is
You can get rid of it and make it watermark free though without distorting the image
how though?
I can't say that wouldn't fool me
Bro text watermarking is even a crazier game
after you write your prompt, you can also set “no harsh artifacts”, to make the picture look more believable
I see. It is something we'll explore, but I don't have specifics to share in terms of updates towards this.
thanks for the response brathar
So are the limits initiated yet ?
There are limits that are currently in place, but it is not the new usage systme.
Aw hell na
Appreciate the response again
no problem. Don't hesitate to use #ask-here.
It can be helpful
So we'll get daily credits to use , lets say we get a 100 credits a day , and i didnt use any of em today , will they be 200 tomorrow ?
i hope the answer is good
got Opus on my battle mode but not in direct chat, why that?
Would you mind asking in #ask-here ?
oh
The daily credits will not roll over into following days.
Wait doesnt that means that side by side will take more credit points ?
Guys why the gemini 3.1 nano banana image generation model has been removed from lmarena
Looks like upscaling/compression issues. Is this a screenshot from I2V? prompt for testing? I want to check something.
Startup Idea: Prompt-expert website with rankings just like leetcode.
You have tasks, you are given a random AI (they give the name of the AI you're using), you have 5 minutes to make a prompt and then the website uses an LLM to evaluate the result after a single prompt, and ranks it.
Something that allows you to put that you're an AI expert on your linkedin
Prompts are anonymous by default but if the person who prompted and got the top result is ok, it can be unlocked for money and you get a fee, like 50% of what is paid while the person who made the prompt gets half the money too.
Price it $10/month for pro version and two challenges a day on freemium
make it super competitive
niche pfp
I'd be your first client btw if someone made such an saas website
Why did they remove the best image generation options "GPT Image 1.5 and 2.0"?
I'd say lets wait and see what the new system looks like. When rolled out, we'll be sure to have all of the questions answered.
We recently removed some models from Direct and Side by Side, you can find the announcement with the reasoning here: #announcements message cc @simple blade
Anyone interested in coding message me
Openai dropping massive thing in two weeks. Dont ask how i know 🤔
the spud?
if the spud doesnt top the arena it doesnt really matter TBH
in frontend and text
no way to reset the deleted email account from arena?
Sorry to say there is not a restore feature
Is it just only me or the models are randomly cutting off for everyone
за что
https://www.youtube.com/watch?v=ON9hzPCSoBc
Rats Survivors
I'm about to release this game. If you can help, add it to your wishlist, I'd be grateful.
Prepare for chaos.
In Rats Survivors, you play as a seasoned chicken fighting for survival against relentless hordes of rats. Hone your skills, unleash devastating powers, and face brutal bosses in a frantic, shoot-'em-up roguelite.
But behind the action... there's a real story.
https://store.steampowered.com/app/4613350/Rats_Survivors/
Enfrente hordas de ratos implacáveis em um survivor frenético, evolua suas habilidades e lute para sobreviver ao caos crescente.
@echo aurora seriously get better moderation than this 💀
3 scams + 1 ad
not deleted by mods
💀
its dangerous to keep such messages visible to everyone, people may fall for it
Hello, pro-hub owner
^ scam4
<@&1349916362595635286> cmon
literally its so simple to prevent those scams 😭
yeah.. A.. can do it better
@echo aurora wanna guess?
@surreal zephyr why is codex so bad
y'all do any of y'all have any gpt-image-2 prompts left
I don't wanna run a bajillion battle mode prompts until I get image-2
if it's even available
compaction is bugged as i said
what's the context window on codex?
tbh
codex is better for larger term projects
but CC literally has 5x the usage limits as xhigh fast
codex couldn't build a not too complex app
granted it had the first prompt and then a single followup
effectively 100k
but still
angst
yikes
it isnt really yikes, it has better recall than any other model
maybe 3 pro is better
at 100k? I'd hope so
do NOT use /fast bro
why
fast is honeypot scam
We're looking into adding more mods soon btw
is /fast actually bad 😭
wdym priority compute
"if noone is using the servers then we give you two servers at once"
oh yeah
this
keyword: IF
its "1.5x fast on average when servers are idle"
1.5x speed for 2.5x cost
apparently
no it literally depends on usage
it can be same speed or even 2x faster depending on time of day
oh
was quoting you 🐦
Automod bot would be much better and cheaper overall 🤔
anthropic borrowed money from google to pay extreme overprice to XAI collosus
xai has the most compute rn lol
openai actually has decent compute too
nah its google and meta
google wasted their compute on ai overview
oh yh
they have nearly none remaining
Those crypto scams aren't the main focus of the concern
well they are one of biggest issues with this server currently from what iv seen? unless theres something even worse going on?
I'd say they're the most common, but IMO not the biggest concern. Most people should be able to tell those are a scam.
well each one of those posting them fell of it, as its spread like virus
its real people's accounts stolen, not bot accounts
HAH
I managed to guess gpt-image-2 out of 20+ generations
so apparently a lot of ppl fall for it
I understand
and even then it got TON wrong
What is the main concern though? Maybe I could try to automate that too 🤔
thats cool
Goated wallpaper
also wrong I think
and it didn't do a wraparound 360° panorama
I should get unlimited access to gpt-image-2
who can @ sam
you have unlimited access in the sub
which sub
gpt sub
also that IS panorama
oh I hadn't enlarged it lmao
it looks wrong
also the seam doesn't seam properly
and rate limit
:(
@sonic badger @sleek urchin why there 2 aeros
And I got banned for thoose mr beast scam images
ts why I never send those images 🥀
Gime invite link
dm
Hello
That would be great as a wallpaper ngl
tyty
I can’t tell
So the dude writes in the article
He says
“From there, I took both images and decided to compare them to extract the delta for each pixel value and plot that. You can see the delta for the Red-Green-Blue (RGB) channels as well as a heat map of the combined deltas and some distribution graphs I’ll explain below”
still looks like compression/upscaling
Yeah not sure how it works exactly but the watermark is in there somewhere
Check that out
show prompt.
That was a screenshot from a video so might be a horrible example
But u do have something that could use your expertise actually
made it a minute ago. The request was zero black with an acid-colored letter G. NB PRO
This is great 👍 same pattern
I’ll be dammed
You ever see the text watermarking?
I saw that you wrote about them here, but I wasn't particularly interested.
You think you can fix that ?
I'll try it for fun. The first thing I noticed, to be honest, was a lamp that looked like a picture on the wall.
-# bump
@hollow ivy you need to add a "it works but i dont like the limits"
I was thinking of removing the lights and just putting diner window
unfortunately, i can't edit polls
can you create a new poll?
lamps can be moved outside the frame
very excited for gemini 3.5 pro
Like this but without the lights make the back wall a window
That way I can have an outside shot and see him thru the window too was my alternative
-# bump
@hollow ivy IMO is heavily depends here
for RP, which would you recommend?
nemotron/mistral/seed are out
y?
:(
its just really good at writing for some resaon
@hollow ivy u forgot tencent/hy3-preview
ah, right!
ty for the tip/heads-up
it sees a lot of tokens on openrouter
apparently so
that's really bad news
I also think Claude 4.7 opus and Claude 4.6 opus are great too. Have to give Kimi 2.6 a go
those are american models
i intentionally didn't put Claude/Gemini/GPT/Grok in that poll
(but i forgot about Muse Spark)
i have hopes for MS-2
yeah, because the first version was already that good
it's strange, that Opus-4.6 is not in DeepSWE
(and some others are also missing, Qwen, MS, etc)
because not funded by anthropic probably
muse spark not accessible
i think they must think their current model is bad if they arent exposing an API
or they just dont have the infra for API/platform etc
its likely both
which model is currently the most underrated one?
muse spark?
no lol
MiMo V2.5 Pro is now same price as deepseek v4 btw
they matched
qwen will prob do the same
yeah, heard that somewhere (in YT, iirc)
but DS V4 pro was still trained on nvda chips
theyre trying to blur that and convince people it was trained with huawei
Amazon, IBM and Nvidia seem to have no chance to catch up with models
yh they gave up i think
Who knows if they will ever come back
and what about other asian models? (outside china)
yeah, but i think, yandex has one
based on llama lol
LLaMa (=lame ^^)
what about that dark-matter model?
this is a joke benchmark
Gemini the least honest ai
One day Gemini will be in that list
When was the last time Google innovated anything?
Everything cool they have they just bought not innovated
You can try deleting the right part of the image and edit image with the same prompt, asking to restore the deleted part based on the prompt and slightly increase the aspect ratio. something like this. As I understand it, this was done in META with a certain style, like seventies animation, right?
Ohh that nice nice
What is the secret sauce to your work?
Finally found this model
a huge amount of trial and error
Yeah, you’re very talented. I’m very good at what you do. 😊
RIP ampro 😭 💔
What happened to him
check the ai server 😭
Is it just me or is may-alpha actually really good
You have some of the best work/skill I’ve seen
Hare what do you think. What will be the future of ai
Will it get shutdown or it will continue to grow
What do you think?
Well I think it'll reach a point it can't grow anymore
Like a limit to knowledge
Like what are we talking about like long in the future or?
Yeah ig
How much ai will last
Before getting forgotten
Or just discontinued
Me personally I think that western civilization is doomed to fail and I think the technology I’ll be the means of our own destruction. I based this on a few historical philosophies.
so... we should never build a big project or keep updating it in the same chat? so far that chat had the best understanding of my project which has come so far 💀
You got a point honestly
Well arena wasn't probably designed for long stuff
What model do you generally use
its agent mode
I heard people say qwen 3.7 max is good with code check and stuff
Oh
one week of chat ig
You should check out those books if you ever have time if you really wanna learn about some really cool stuff in history
I don’t see technology as a way of progress. I see it as it means to an end. What goes up must come down.
I have plenty of books stored somewhere probably even older than me because my grandma had them gotta check them out later
I wanna show you something really quick
What is it?
@vapid berry
Wrong guy
<@&1349916362595635286>
Some people really gotta stop putting moderator in their name. I keep pinging them on accident
Yup
I think I heard a little about it
It’s a Roman soldier of Pompeii
guys is gemini image is not available now?
The paintings look good though. I feel bad for the people there that couldn't escape the volcanic eruption
It’s very symbolic because his civilization the Roman is no longer around
And there was no hope for him because the volcano blew up
But he still held his ground. That is to say that even if the future with technology and AI sometimes made it look bleak.
It’s hard to explain, but there’s a lot of symbolism in it
Mean about people’s behavior
Cause I think dude technology and AI is driving people crazy social media all of it I think it’s not healthy for society. I’m not against it, but I just think that there’s consequences of that we pay the price for.
I personally think that this technologies are extremely addictive and it designed this way
Yeah I guess
The creativity is disappearing
Well, he writes about that in the book too Oswald Spangler about creativity
Though not that I had any creativity to begin with.
Like I genuinely struggle to create something creative enough. Even in sandbox games like Minecraft
All I can think of is building a wooden box to live in
Well, one example I could give you is if you look at a lot of AI generated stuff it’s usually people generating like IPS and characters that were all familiar with you know show celebrities, etc. blah blah blah
And it’s kind of ironic, isn’t it ?that we have this tool that could generate anything your mind could think of?
Though you know what would be disastrous
If somehow ai devs manage to give ai actual creativity
Creativity subjective
It’s all on the beholder. It doesn’t matter if it’s AI generated or if it’s somebody that painted it.
Think about it like this AI could beat the worlds best chess player
But people don’t tune in watching two AI’s duke it out to see which robot is the worlds greatest chess player
Why is that?
Gemma 3 27b Best gpt-oss 120b
Because robots I guess?
Who wants to see robots fight each other
That feels lifeless
not me
But anyway Back to the topic.
I'd say qwen 3.7 max non preview is rather good
It managed to replicate Roblox page quite well
welp someone got hacked lol
<@&1349916362595635286>
Hi! Is Nano Banana / Gemini 2.5 Flash Image Preview no longer available in Direct Chat? I can’t find it in the model selector anymore. Is this expected or temporary?
There’s nothing more lifeless than a robot my guy.
This is expected. May be temporary for when the new usage system rolls out.
ah ok, so you see the same?
This is real lol 😂
Well I won't lie those seem fun
That’s what I’m saying, though bro you’re not gonna pay $300 a ticket to go watch two robots play chess
Fr
Which you might end up paying $300 for some front row tickets to watch two robots destroy themselves
Yeah, a recent announcement announced this change.
Can say it very much is, went to a Battle Bots event once, 10/10 experience would recommend
Bro, you remember this game this was before ai 😂
This is one of the greatest contradiction of our modern age.
That’s why I say that it’s important to understand stupidity, because we’ll never know what intelligence really means s until we first have a concept of stupidity meaning
why i cant upload HTML file in Arena
@echo aurora
We haven't yet added that capability. We do plan to expand this.
Please add it, I really need it
Hey pineapple I have a question
Do you have a favorite ai model
Like what ai models you prefer or just have expectations on
Which Chinese model do you prefer?
Qwen seems to have a lot of potential
Especially qwen 3.7 max non preview
Yup
I’ve always used to think that stupidity was a disadvantage for the human race, but it turns out that stupidity has been extremely advantageous for humanity 🤣
Each new model from Alibaba brings a bigger surprise than the previous one.
So that why people keep saying F students are the inventors
I personally use gpt 5.5 a lot
I don’t know what the deal is, but I can tell you that it’s extremely ironic and contradicting, especially the fact that we live in the information age where any piece of information could be accessed on infinite number of topics or subjects at the whim of your fingers
But the more we have like the dumber we become lol
It’s almost like universal poetic justice.. think about this ancient man knew how to travel by stars… try to find a school near you or instructor within a 100 mile radius lol who can teach u
is it fixed ?
The bug we were seeing over the weekend? Yes
thanks but no limits ?
No there are still limits. Having limits wasn't a bug
i never had a limit before tho
Limits can adjust over time, this is why we want to move to a new usage system
ok bru , ty
No problem! We're hopeful this new system will be way more clear about what these limits are
if anyone is experienced with prompting in general, more spesifically veo 3.1, and your willing to help me out, please send a dm
Muse spark is over 😭😭
what are the ratelimits for battlemode?
Those limits aren't displayed at the moment. We do plan to make a change where this is visible to the user.
please help. I have been using Max-search mode, but after the second or third prompts, it stucks forever with "Searching..." for 1-3 days without stopping. There is no stop button
Can you ask in #ask-here that you're getting an infinite generation bug?
wait wait wait what happened 😭
Heads up I've turned off the ability to upload media for now. This is a temporary measure.
their is a pretty good chance it will be back within the new credits system update, however their is a chance it wont.
ok
wait... new credit system?
whats it about?
the way it works currently as of today, is every single model has its own amount of credits in a way, which is pretty confusing.
the new system will gather all those credits, and let you use them however you like, on 1 model perhaps. or a few you like.
oh ok!
the biggest diffrence is the limits will be daily, not hourly (unless this changes)
ok.
Is there any timeline now on when this change might be happening or is that still to be determined?
We cant send images anymore?
We are currently seeking a reliable partner to collaborate with.
Applicants must be European nationals and possess knowledge of AI.
This recruitment is a great opportunity for entry-level developers to learn and grow.
If you are interested, let's talk