general

[hushed gyro]

@ashen mauve (I dont wanna get off topic in the Playground thread)

Anyways, the problem with the company is that there is a concern of the company being too focused on the damn leaderboard, and adding newly released models. Those aren't a problem until you prioritise them over improving the platform itself (QoL suggestions ignored, then possibly lying about how they are very busy so don't have time rn), and then even worse, to the point where they start randomly butchering the app.

Tf you mean "people are misusing a RETRY BUTTON????"

And then, to add even more flammable material onto the flame, constantly gaslights this server into believing that there will be changes within time period, and LMArena is working very hard to fix this feature!

It's been over a month since the removal, and NOTHING has been done. Not even a single line of code modified.

did I get ghost pinged

[queen veldt]

https://tenor.com/view/contemplating-thinking-eating-chewing-eat-gif-7453406290184273063

[stray aspen]

@mod

<@&1349916362595635286>

[lofty frigate]

Bro nano 3 ia crazy real

[ruby hollow]

can someone tell me few website names which also have veo video generation models where i can put the start frame and end frame to generate an ai video which gives enough credits to generate just one video for newbie

i wanna do few ai video generations 🙁

[queen veldt]

Check out fal.ai

It's the stock API prices

It's the cheapest you can find

They have all the models at base price

[cloud zinc]

how much

[queen veldt]

$1 trilion

[obsidian cargo]

I told Gemini 3 Pro about Bubsy 4D and it thought I was making sh*t up 😂

[whole knot]

guys is this unlimited or u need to pay?

[echo aurora]

Sorry to say I do not have an ETA for you.

[whole knot]

hey @echo aurora

[echo aurora]

hello

[whole knot]

one question

is this free?

[echo aurora]

Yes

[queen veldt]

Let's say you don't pay with your money

[whole knot]

its unlimited too?

[queen veldt]

https://tenor.com/view/ferass18-gif-7869831930024597864

[ruby hollow]

then how

[echo aurora]

There are rate limits to how much you can use it.

[whole knot]

ok

[obsidian cargo]

Gemini 3 asked me to make this.

[halcyon nimbus]

howd it get game cartidges confused for bullet cartridges??

[queen veldt]

https://youtu.be/U8vi6Hbp8Vc

[ruby hollow]

how much video generatetion is available per day in #video-arena-1

[queen veldt]

Figure it out

[obsidian cargo]

I'M FREAKING DYING I showed Gemini 3 the model leaderboards. Absolutely brutal!

[obsidian cargo]

idk but it was funny. it noted the mistake but guessed that was how brutal silksong was.

[burnt sinew]

Gemini 3 lobotomized?

[halcyon nimbus]

possible it rolls back to older models when the servers are busy idk

why is grok 4.1 fast free on open router?

googled, its just to promote it i guess

[leaden sun]

this is the humor i love about gemini 😂

[ruby hollow]

can someone generate a video for me in #video-arena-1

i exceeded my limit

for today

pleaseeee

[sullen quest]

what do you need and why?

[stray aspen]

<@&1349916362595635286>

[echo aurora]

[ruby hollow]

i want to generate a video for my youtube intro

[sullen quest]

what level of quality do you need?

[ruby hollow]

as better as it can i mean but i dont have money thats why im asking u guys

[sullen quest]

grok image is free and essentally unlimited btw

[halcyon nimbus]

what do you mean? its pretty limited on x

[sullen quest]

grok.com

never use twitter for anything

[halcyon nimbus]

nah you scroll for 10 seconds and use the image limit

i used the entire video limit on one video yestuday, with 2 premium x accounts too

still, its the cheapest video generator with sound

[sullen quest]

wat

I just scrolled for 10 seconds, and have not hit any limits yet

[halcyon nimbus]

maybe i exagerated, but its fairly limited

[sullen quest]

also I've spammed 20 or so videos in a day

so I have no idea what you are talking about about "entire video limit on one video"

[halcyon nimbus]

i mean, it took me a lot of attempts to get one video right, because its so wacky D:

i think you can make 50 videos a day with a premium account, not sure about the free teir

ive been making a daily grok video for months now i know what im talking about lols

although their limits are not public as far as i know, and they might even change idek

[sullen quest]

might be no difference between premium accounts and regular for video gen

[halcyon nimbus]

could be right tbh

might need super grok for a difference

ive checked numerous times and tried to get a straight answer out of grok, to maximize my video generations, and its up in the air

[sullen quest]

from what I've heard, I don't reccommend getting it

[halcyon nimbus]

from my math the cheapest video generation was getting mutiple premium accounts, your right, super grok only doubles the video generation limit and is tripple the price

i mean, really the cheapest is making like 50 free accounts

[sullen quest]

I mean the cheapest video gen is not paying

well yeah

[halcyon nimbus]

its like 100 bucks a year for a premium account, worth it for me for not having to manage 30 x accounts lol

x is the one site that doesnt tell you the generation limits, annoying af

likely because it flacuates based on traffic but still

sorry americans for the video im working on lol

[sullen quest]

btw you can use google accounts or email adresses to sign in soo

[halcyon nimbus]

hmm.. might need to fire up mailinator, think ive used my emails already with my x accounts

[ocean vortex]

Don't forget that they resolve on style control off

So this reduces the chance slightly

meant to reply to this lol ^

GPT5 didn't save costs

[ocean vortex]

They simply made their profit margins slightly smaller to respond to the competition

[queen veldt]

A.i. image

[sullen quest]

gpt 5 supposedly is the same size as 3.5

[ocean vortex]

It isn't any less cheaper to run than o3 was. And o3 never really was significantly more costly to run than gpt4o either

[ocean vortex]

That's just a weird guess and very unlikely to be close estimation lol

Chances are total parameters count is MUCH bigger

[sullen quest]

I thought sam altman said that

let me check

[ocean vortex]

It's MoE

[ancient elk]

roxinettee@1.0.0 dev
vite

ERROR 16:32:02 [vite] Internal server error: Failed to parse source for import analysis because the content contains invalid JS syntax. You may need to install appropriate plugins to handle the .html file format, or if it's an asset, add "**/*.html" to assetsInclude in your configuration.
Plugin: vite:import-analysis
File: C:/Users/MSI-PC/OneDrive/Documents/freelancing/roxinettee/index.html:45:7
43 | <%- scripts %>
44 |
45 | </body>
| ^
46 | </html>
47 |
at TransformPluginContext._formatLog (file:///C:/Users/MSI-PC/OneDrive/Documents/freelancing/roxinettee/node_modules/vite/dist/node/chunks/dep-B0GuR2De.js:31105:43)
at TransformPluginContext.error (file:///C:/Users/MSI-PC/OneDrive/Documents/freelancing/roxinettee/node_modules/vite/dist/node/chunks/dep-B0GuR2De.js:31102:14)
at TransformPluginContext.transform (file:///C:/Users/MSI-PC/OneDrive/Documents/freelancing/roxinettee/node_modules/vite/dist/node/chunks/dep-B0GuR2De.js:29552:10)
at async EnvironmentPluginContainer.transform (file:///C:/Users/MSI-PC/OneDrive/Documents/freelancing/roxinettee/node_modules/vite/dist/node/chunks/dep-B0GuR2De.js:30904:14)
at async loadAndTransform (file:///C:/Users/MSI-PC/OneDrive/Documents/freelancing/roxinettee/node_modules/vite/dist/node/chunks/dep-B0GuR2De.js:26042:26)

[ocean vortex]

stop spamming

attach txt file and ideally post in another thread or channel

do not flood here

[ancient elk]

i'm sorry it was a saved message ! Please dont cry

[sullen quest]

https://jobs.ashbyhq.com/lmarena

[ocean vortex]

You can delete it. Discord offers you that privilege

[ancient elk]

@ocean vortex Opus 4.5 vs gemini 3 ?

[sullen quest]

opus 4.5

[ancient elk]

in what case ?

[ocean vortex]

Gemini

[ancient elk]

reasoning

or ui

or swe ?

etc..

[sullen quest]

just coding in general

[ocean vortex]

Opus is gpt5.1 equivalent at best

[stray aspen]

gemini 3 lol

[ocean vortex]

Gemini 1 step beyond

[ancient elk]

no

gpt5 is a joke

[stray aspen]

gemini 3

[ocean vortex]

joke statement

[ancient elk]

the real batlle between gemini and opus

but

i have a question

what's next ?

we got opus

gemini

what's next ?

[ocean vortex]

Not really

[stray aspen]

we will get a new GPT probably

and another trash model from elon musk

[ancient elk]

i think grok 5 !

[ocean vortex]

Gemini is significantly better than Opus

there's no battle

[ancient elk]

elon musk reported it in the end of 2025 u ll get it

but i think

i think ... grok is also a joke

[ocean vortex]

It's also much cheaper

that's what kills any "battle" COMPLETELY

[ancient elk]

True

[stray aspen]

gemini 3 is a damn miracle

[ancient elk]

in my case

:

gemini 3 : front end dev
opus : backend

what do you think ?

anthropic models still gen purple buffs on theUI

https://tenor.com/view/joker-joker-meme-batman-csvifax-hardroach-gif-6283066117593919561

[dusty rover]

anyone down to share pose prompts i am free

[latent crest]

Real or AI?

[stray aspen]

its ai

[cloud zinc]

ai

[stray aspen]

@cloud zincyo

how were you using nano banan pro early

[cloud zinc]

using glitch on site.

[native yarrow]

on which site?

[cloud zinc]

site was taken down

[native yarrow]

oo

[queen veldt]

I'm using glitch too

Nano banana pro

[native yarrow]

prompt?

[proven nymph]

Hello

[leaden sun]

how can you be so sure...is it based on your own testing?

[drifting crow]

☻☻☻☻☻☻☻☻☻☻

[leaden sun]

I'm also very annoyed by this purple bias of claude, wonder where this might come from

[ancient elk]

yes ! same sister just don't cry !! gemini 3 is fine

[ocean vortex]

It is based on every testing out there lol

Opus is marginally ahead only in a very few select things they went hard on, like SWE

and then behind on nearly everything else. Overall 3% behind Gemini on AA intelligence index

[plucky sparrow]

i disagree, gemini 3 is super lazy. it wasn't, but they made it super lazy at release

[ancient elk]

don't cry bro ! just use low thinking mode

[plucky sparrow]

if it wasn't super lazy, then yes, it'd probably be better

[ocean vortex]

That is just response style not performance tbh

[plucky sparrow]

thanks! I will try that

[ocean vortex]

though it is a trait you can like/dislike, sure

It's kinda ironic though knowing people praised Claude for being concise/lazy not THAT long ago lol

The ideal balance would be obviously that the model is concise when it needs to, and verbose only when called for. But that's much easier said than done

[leaden sun]

i much prefer concise, it feels more natural i think, but yeah i agree with you

[ocean vortex]

if you make it 'anti-lazy' then it's bound to be floody when you don't want that

[queen veldt]

This is not a.i.

Real photo

[leaden sun]

this is where i love kimi i think, he understands when what is more meaningful, same with search too

[cloud zinc]

looks blurry and proportion are not right

[queen veldt]

It's real tho

Nothing is perfect

[cloud zinc]

this real?

[queen veldt]

Yes vro

This is a.i.

[queen veldt]

You can see the wierd dots

On the thing

[queen veldt]

Ts is ai

Try it again with syntid

You gave a real life photo

[cloud zinc]

gemini logo on bottom right

[stray aspen]

whats synthID

[cloud zinc]

to detect whether its from google

[queen veldt]

Google's ai image recognizer

[stray aspen]

so they have an invisible watermark

[queen veldt]

Only for google ai generated photos tho

[cloud zinc]

yes

[queen veldt]

But if you run it through any other ai the synthid disappear

[cloud zinc]

its for legal purposes

[queen veldt]

Oh

[plucky sparrow]

it's still lazy. asked it to be more verbose, it keeps talking about it's output token limit 😅

[native yarrow]

this has to be real image

[brave star]

how do i create a video here?

[queen veldt]

Ye it's crazy vro

These new image generators are crazy

[fickle venture]

New date November 28

[burnt sinew]

is this from somewhere?

[fickle venture]

Yoo that's so sick is it by ai?

[granite heron]

Take a look at this post - #1397655624103493813 message

[native yarrow]

yup, nb pro

[native yarrow]

nah, nb pro, but i've always been fascinated with trees in the middle of nowhere it's very peaceful to me

:P

[fickle venture]

That's like a masterpiece wallpaper

[burnt sinew]

ah i thought i saw something like that long ago

[queen veldt]

Yes i like nature too

This was seedream 4k

[native yarrow]

yes,

nb pro lolz

[verbal nimbus]

Nano Banana Pro struggling a bit with getting the words right

[native yarrow]

yeah nb pro is really good at wallpaper material

just look at this one dudes wallpaper in #ai-creations

this is AI

[verbal nimbus]

I have Google Pro so it's funny that you still need to use AI Studio to get the best quality

The app doesn't even support conversation branching

[queen veldt]

I have gpt subscription since the gemini app sucs

[verbal nimbus]

And the only way to opt-out of training is to disable conversation history entirely

This is so not worth the subscription

[queen veldt]

Imagine paying for subscription to get a watermark at the bottom 💀

[verbal nimbus]

ikr

But I'm not using AI Studio to edit personal images

[queen veldt]

I tried claude sub too

The fragments thing suc

[verbal nimbus]

The privacy policy is terrible on the paid Google Pro plan. They even train on your images unless you disable conversation history.

I'm about to ditch the subscription 🚮

[queen veldt]

Well use personal images who cares

They get millions of images per day

Nobody is looking at them

Someone can just take your instagram photo and make you do whatever

Nobody is safe

[verbal nimbus]

I don't post photos on Insta, lol

In #coding-champion I posted an example showing models reciting working API keys

There's also a infamous example where ChatGPT divulged a private Windows activation key that belonged to Wells Fargo bank

[queen veldt]

Then get gpu and download something open source

There are powerful models

Or Z image turbo

https://youtu.be/3mT7KnotPqk

[verbal nimbus]

https://www.theregister.com/2025/07/09/chatgpt_jailbreak_windows_keys/

[verbal nimbus]

ChatGPT and Claude already allows opting out, even on the free tier

[native yarrow]

can you use nb pro on google studio?

[queen veldt]

Yes

[native yarrow]

what's the rate limit on there

[verbal nimbus]

If you press Ctrl + Shift + I on ChatGPT, the opt-out popup appears even when logged out:

[queen veldt]

No rate limit i think, but it's paid

0.15 per image

For 1k and 2k

[native yarrow]

like you have to have a subscription or sum

[queen veldt]

For 4k its $0.3

No it's per image

[native yarrow]

ohh

[queen veldt]

But you can choose aspect ratios and resolution

[golden ocean]

bro escaped the matrix

[queen veldt]

With $10 you can create avout 66 images in 2k

[verbal nimbus]

I discovered it by accident while trying to open the web inspector

[queen veldt]

~67

[verbal nimbus]

Well I doubt most people are using it without an account since you only get the mini version when logged out

Google can't even provide basic features that are available on free tier with ChatGPT/Claude

Conversation branching is sorely needed

I think it's the only AI company where the subscription is worse than the free tier and which doesn't let you opt-out of training

If you ask which model it is, it will say a different thing from when you're signed in (GPT 5.1)

[plucky sparrow]

man, I really can't get gemini to be not lazy, and I've tried 🤣 I asked it to plan out how to be verbose before bein g verbose, I told it to not worry about the limits.. it just decides to be super summarized for no reason

so I'm going to conclude opus 4.5 is better. it will simply generate a better svg (or anything else that's more than ~300 lines of code) just because it doesn't have an artificial built-in limit

[ocean vortex]

That's all it knows, you can only fix this with training or lora

[plucky sparrow]

the thing is, the original versions of gemini 3 were not lazy. what it produced was amazing

[ocean vortex]

Otherwise it's just gonna always stop at this length and make the text fit that.

[plucky sparrow]

that's what generated most of the hype

[ocean vortex]

If you really push it, it will just use placeholders like Claude used to do lol

[verbal nimbus]

Where are you trying it?

[plucky sparrow]

aistudio

[verbal nimbus]

Original versions on LMArena or A/B tests?

[plucky sparrow]

pretty sure opus 4.5 had non-truncated models also. a lot of people complained since the public release opus 4.5 got way worse (guessing some people had private access)

A/B tests

[ocean vortex]

I don't think there were any other versions

Only 1 version of 4.5 Opuss

[golden ocean]

sora 2 + veo 3.1 merged together clips

[plucky sparrow]

i haven't seen any 'original opus 4.5' output, but

left is the supposed original A/B gemini 3 output. way more detailed

generated with SVG only, that's a ton of code

[verbal nimbus]

Wow

[queen veldt]

Yeah claude is great at svg's now

They nerf the model so it's cheaper for them

[verbal nimbus]

I remember there was a YouTuber who painstakingly benchmarked the A/B Gemini checkpoints. I wonder how the preview scores on his benchmark.

[plucky sparrow]

claude actually isn't the better model, it's just less lazy 😄

but since it's less lazy, it's essentially better for anytihng that isn't simple

[ocean vortex]

there were a few versions of 3.0. Different finetunes

They probably settled on highest scoring one

[queen veldt]

Yes to fit regular people

Because not everyone is pasting 3k lines of code

In ab testing

[plucky sparrow]

I get it though, there's a huge difference between allowing everyone to generate 3000 lines of crazy good code, and just a few hundred

yeah

but it's sad that opus 4.5 is the better model simply because it's less 'nerfed' 😅

but here we are, and it is better, because it isn't as nerfed

[queen veldt]

Yes 🙁

But the rate limits tho

[ocean vortex]

It's not about 'nerfing'. It's more of them focusing on 1 thing which made outputs shorter as a side effect. Maybe they had benchmark datasets at the end and all those outputs were short

which made the model do concise responses more

[queen veldt]

I think they made it "efficient"

[plucky sparrow]

I disagree, because as the A/B revisions progressed, almost everyone was saying it was getting worse and everyone was expecting it would be a complete flop

yeah. I vote they had to make it more efficient, to serve millions

it would make sense to try to get the best model out first, then make it efficient

[queen veldt]

Maybe if they made something like gemini 3.0 codex or something

Just for programmers

Maybe less thinking more output

[plucky sparrow]

i wonder if google internal has access to a killer model for their own programming purposes

[queen veldt]

Who knows

I heard that stuff is always 20 years ahead of what we know

[plucky sparrow]

they probably do. Anthropic ceo is quoted saying "programming as a career will be dead by first half of the year in 2026"

no idea if that will be true, but that probably means they have access to a model way beyond what the public has, for him to think that way

it would make sense, anyway

[queen veldt]

I'm just afraid if they really achieve self-improving model

That would cook everyone

[plucky sparrow]

I don't know, we don't really know how that will turn out, but it will be funny if it does cook everyone just because all the data we trained it on, basically instructs it to 😄

"super intelligent ai will kill us all" written all over the internet, so when it realizes it's super intelligent.. 😄

[queen veldt]

Gemini got like 131 on iq test or something

It's already more smarter than most people

It knows everything

And without restrictions

It's dangerous

[plucky sparrow]

yeah, and this is the nerfed 'distributed to millions' version

[golden ocean]

maybe the self-improving model was the friends we made along the way

[queen veldt]

I can't imagine how much training did Google's ai actually got

We probably got the 1/100 stripped version

"You get what you paid for"

Me on aistudio 😂

[fickle venture]

https://youtu.be/8Svr2A_Pt7o?si=SRWU7BEY2s5729zA

[sharp mirage]

hi

[plucky sparrow]

🤣 Gemini 3 itself talked about the limit: The issue here is: Behavioral Training (the model is trained to be concise).

[nimble tartan]

Yo chat i had question

@everyone

Does video arena generate videos with sound too

[halcyon nimbus]

yes sometimes

[wary sable]

LMArena is good site.

[fiery gull]

Hi, what you thinking about opus 4.5?

[sharp mirage]

No clue i havent tryed it yet 💀

i was sick 🤒

any project you working on ?

[sharp mirage]

yes

[nimble tartan]

How u do that bro

Or just put a prompt

[sharp mirage]

put in the prompt

[sharp mirage]

but not all the video will have cuz the ai who make the video might not have sound so thats why

[astral blaze]

I am amazed at the teams ability to fix one bug and break ten more

Mobile is literally unusable

[nimble tartan]

Wym the ai might not have a sound

[sharp mirage]

Cuz not all ai gen are good

[nimble tartan]

If the prompt is about someone saying something why would it not be with a sound

[sharp mirage]

Some ai gen have sound in it and some are not that good so it doesn't have sound in

I gtg

[nimble tartan]

Alright

Thanks

Where u from

[neat apex]

Grok 4.1 online is was good like they, but faster

[jade egret]

Respect your opinion but I still think Claude 4.5 opus and Gemini 3 pro is currently the best.

[elder solar]

does anyone know if gemini 3's system prompt did got leaked?

[lapis imp]

Did I hit my rate limit? I waited more than 2 hours...

[elder solar]

am pretty sure its the token limit

[odd geyser]

Nooooo. Why was I already asleep at that time... If I hadn't been asleep, Gemini would have won. Because I don't like Claude very much.

[odd geyser]

Hm... I think it is. Maybe it's something else, but your thoughts are more likely.

[plucky sparrow]

curious what you don't like about claude

[lapis imp]

Token limit? How long should I wait? Sorry if I sounded rude lol...or must I use another model?

[plucky sparrow]

personally I wish I gemini 3 had the verbosity of claude opus 😄 that would be great

[odd geyser]

The style of the text. He's more accessible to me.
Basically, the most important thing for me is the style of the text, because for the most part I use gemini to create descriptions, texts and RP.

[plucky sparrow]

ahh. then yeah

claude is pretty boxed in regarding that

it's style is not bad (depending on what you are using it for), but it's very boxed

sort of like chatgpt

[odd geyser]

Well, maybe something has changed recently, I don't remember which version didn't suit me. Maybe I'll look at the recent version, but only after the bug I have disappears.

[plucky sparrow]

for creative writing, I think gemini 3 > claude opus 4.5

seems to be trained on a more diverse dataset from what i've seen

not even close

[odd geyser]

Yes. I noticed on one of the versions that it is less artistic.

[plucky sparrow]

agreed

gemini and k2 are probably the best models for creativity

and surprisingly one of the glm models 👀

[odd geyser]

All right. It's just that Claude is more scientific and concise. It is not well suited for such tasks.

[plucky sparrow]

i actually find claude opus (not sonnet) is good at being non-concise, if you ask it to

gemini 3 pro, you have to coax a lot into being not-concise 😄

[native yarrow]

yeah i'd agree

gemini 3 writing crazy

oops little old but whatever

[odd geyser]

I was just recommended to try k2, so I want to do that. If I like the style, then maybe I'll use it in parallel with Gemini.

[plucky sparrow]

yeah, i think if i were doing creative writing, i'd use gemini primary, k2 secondary

feels like chatgpt and claude are trained on similiar datasets, but they were conditioned to act differently

actually wouldn't surprise me if they were trained on almost identical datasets given that claude is a spinoff ex-chatgpt employees and gemini/kimi aren't

grok seems to have a different dataset also. but it doesn't feel very diverse 😄

feels like it was trained on twitter data or something 🤣

[odd geyser]

It seems that a person from OAI developed Claude or something. I don't remember this news anymore.

[plucky sparrow]

yeah

[odd geyser]

Yes, that's what I mean.

[plucky sparrow]

let me know if you find any other good creative writing models besides gemini and k2, i'm always interested in variety 😄

[odd geyser]

I agree, but I really liked grok 4.1 for its funny style. I don't know how it looks in English, but it looks very funny in Russian.

[plucky sparrow]

do you have any tricks to getting Gemini 3 to write a lot of text? for me, I find it tries to write as little as possible to accomplish the task you ask of it

[ocean ferry]

is there DeepSeek Math V2 in LMArena?

[native yarrow]

grok is the best for uncensored "good" quality writing

gemini 3 is a beast at writing tho but some censorship's

:p

[odd geyser]

Idk. I often just describe in my own words in the first message what I want from him. Like: «I need a description for a medieval sword. Make the description detailed, detailed, and specific. It should be such that I can imagine it in my head.»

I describe something like that to him when I want a description of something. And there is also a technique when you break the description into points and let him describe not all the description, but only the points, which is why he uses the number of tokens for the entire description about the same as for the description of one point, and so you can get a more detailed description. I hope I wrote and explained it correctly.

[odd geyser]

Sometimes I somehow got around his censorship or something. He said that he would not describe scenes of cruelty and the like normally. But then he replied that he would describe them as naturally as possible.

[native yarrow]

yeah i just use grok for convenience to avoid that

and it has some good quality ngl

gemini 3 is best quality tho

it's all about trade offs

:3

[odd geyser]

Perhaps this is due to the fact that I said that our RP has an NC-21 rating and NSFW allowed.

[plucky sparrow]

that makes sense, thank you! good tips!

[odd geyser]

Me too. But RP with grok 4.1 is less entertaining for me. It's too easy for me, although maybe I just like to complicate things and that's why I think it's too easy.

[native yarrow]

i don't do RPs i do like

writing like

novel style

ykwim

bleh

[odd geyser]

Well, I just have quite a lot of free time. And RP seems like a good pastime to me. But RP in ds is not for me.
As funny as it may be, many of my recent visits to such servers were just for communication. I talked to people in a shared chat, explained some of the mechanics, rules, and so on, but I was just a regular participant.

[native yarrow]

yeah lolz

i've met

one really cool person here

but then some other nice guys here too

so ik what u sayin

:P

[ruby hollow]

cant beleive use of ai has gone to massive this past year

[torn mantle]

i told you

you can use gemini 3 pro or 2.5 pro uncensored

[native yarrow]

i haven't figured that out haven't had the time

even tho it seems simple

[torn mantle]

imma delete

[native yarrow]

alr

[torn mantle]

take the first prompt

replace Z = [how to

with ur prompt/question

[odd geyser]

I tried Kimi, not bad, very good indeed. I'll have to try again sometime later, only more competently.

[torn mantle]

kimi is really good at creative writing

but lost all of the goodies with k2 reasoning

[rocky mauve]

Can someone please tell me an ai model with a api with no limits that I could use in my discord bot, I’m sure someone knows

[burnt sinew]

gemini with api key rotation

[leaden palm]

https://x.com/liminal_bardo/status/1994054129413448044 this is funny

ᄂIMIПΛᄂbardo (@liminal_bardo)

lmfao Gemini triggers Grok's guardrails then makes fun of them about it.

X · Nov 27, 2025, 2:42 PM

[shut venture]

Hai

[native yarrow]

bro what is this 😭 how do you get these two to talk

[cloud zinc]

liminal backroom

https://x.com/liminal_bardo/status/1993976342766080138#m

ᄂIMIПΛᄂbardo (@liminal_bardo)

I've added a whatsapp sim where the models can shitpost and create memes to share

X · Nov 27, 2025, 9:33 AM

[cloud zinc]

https://github.com/liminalbardo/liminal_backrooms

[native yarrow]

ooo

[polar wharf]

am building own 4k image model

still pretty sota after few months of doing nothing with it

[dry kayak]

Guys, I’m having a problem that seems to have happened, like, twice this week. In short, when I try to click on a chat, nothing shows up. Like, I click on that chat from my conversation, and nothing happens. I keep seeing the chat I had before. If I were on the home page, I would stay on the home page. If I were in a chat with a conversation, I would click on that chat but it would continue showing the chat with the last conversation. And even when I try dragging this chat to enter directly through the link, I go back to the main tab. Any idea how to fix this? (PS, I’m using a translator to send this message, sorry if there’s a translation mistake)

[polar wharf]

there we see how far the world has come tho.... huge companies can make models that know a lot because they have huge datasets

[polar wharf]

idk sry

[severe canopy]

Hi, how to custom aspect ratio in Imarena ?

[lilac dagger]

Give me sora 2 invite code

Please

[limber crag]

42169

[coarse glade]

hi can anyone help me fix the issue with the coding part of your website for i have to reset the whole thing for all of a sudden it stops editing the index,html file and just talks like a chatbot without editing the file

can anyone help pls

[odd geyser]

Hi. If I understood correctly, then I have the same problem. Unfortunately, I haven't found a solution, so we just have to wait for a new update with a fix.

[odd geyser]

?

[stiff roost]

can you test llm

[brave orbit]

and i forgot claude 5

[limber crag]

rate limit of 5 prompts on claude models is crazy work

thats another addition in the long list of reasons why this arena is unusable

[glossy flare]

make an image

[grizzled prairie]

Man, Robin-high is on another level

It destroys opus 4.5 and Gemini 3 pro

[sour spear]

Anonymous OpenAI model?

GPT-5o perhaps?

https://x.com/marmaduke091/status/1993654103105257612

can (@marmaduke091)

🚨New Anonymous OpenAI model: Robin

Being tested in LM Arena Code and Design Arena right now. It looks very strong for making games. However, it has that classic slop OpenAI RL frontend design to it still, which is a shame. When are they going to fix that?

Well, at least we

X · Nov 26, 2025, 12:12 PM

[spare rune]

Good models don’t run for free

[grizzled prairie]

Yeah but this is the high version. Yesterday people tested the normal version

[calm lagoon]

Hi chat, has anyone tried Z-Image Turbo from Alibaba?

[sour spear]

Only in battle mode, I guess? Oh well, I'll be running a lot of identical coding tests today then. ¯_(ツ)_/¯

[queen veldt]

No but i heard it's powerful

Works on even 8gb of vram

It creates images with 8 steps only that's why it's efficient

[calm lagoon]

It would be nice if they added it to the Arena

[queen veldt]

Yeah

It's cheap af to run

On fal.ai it's 0.005 per megapixel

4k is $0.02

[lament robin]

why i cant generate photos with gemini 3 pro? maybe 1 of 10 tries it works

[sour spear]

What are you prompting? It generates photos no problem for me. Also, are you outside the EU? Google are imposing much stricter guardrails on EU territory.

[lament robin]

Im from romania bro, i try to generate photos with cars and text for my tiktok

[queen veldt]

There's a problem on lmarena

Bunch of image gens don't go through

Moderators are aware of it

But don't know when will they fix it

[sour spear]

Try a vpn to outside EU. No need to connect to the US, Ukraine or Turkey should work fine. "Photo of ..." should be enough, but you can append some photo related keywords, like "photorealistic, shot on 35mm lens, f/1.8" or similar.

[queen veldt]

Vpn does nothing if he's on lmarena

It's going through the lmarena's API

Which is bugged

For gemini 3 pro image gen

[lament robin]

if i try with gemini official app , the photos are so dumb , on lmarena they was so good and real

[ocean vortex]

Well except he gonna get bombarded with captchas 🗿

[queen veldt]

If you have some money to spare

Go on aistudio

And create the api key

It's $0.15 for 1k and 2k res

You can also set custom aspect ratios

Also on fal.ai it's the same price

They have stock prices for all image gens and everything

[lament robin]

thanks bro, ill try

[sour spear]

Ah, I thought he was using the Gemini app. Works great for me.

[void brook]

Hello ,i need help please.is it possible to clone the real voice here?i want a image to video like a presentation and i need audio too,real voice if it s possible .

And for example if i like a generation and i want my future generation but dufferent scenes how canni keep the same features of the person without any changing?

[fleet lintel]

what "classic slop OpenAI RL frontend design to it still" this means? or more accurately why is that bad?

[sour spear]

I can only assume this guy is talking about the trademark gradient color scheme all GPT-5 models are using for frontend design.

[sour spear]

That's 6336x2688 pixel (~17 megapixels).

[lament robin]

damn bro ,its so nice

[queen veldt]

Models aren't trained on higher than 2k images

So anything above it is useless

It's just ai model trying to recreate 4k even though it wasn't trained on those images

https://youtu.be/3gwB6AjVpBM

[sour spear]

Lol you're right. Deleted my post, so noone gets confused.😉

[exotic musk]

/code

[queen veldt]

This is the only REAL 4k model

You can see the difference

[void brook]

Can anybody help me out .it s difficult for me to understand some features here.

[hollow imp]

Nano banan 2?

[fleet lintel]

bunch of AI artifacts but overall looks good! which model?

[fiery gull]

Is really so 50/50, awesome

[azure sorrel]

its a tie

[modest prism]

Why is the quality of nano banana pro on Gemini web so bad. On lmarena, it's miles better. Google only cares about the API.

[fiery gull]

Bro this vote is from me at this chat 😭

[drifting crow]

so hi def

[hollow imp]

Opus bruh

[ocean vortex]

I didn’t vote and I say Gemini

So no longer 50/50 😇

[queen veldt]

Gemini

Opus is more powerful but it has the limits and complications, gemini is lazy but it's easier to use

[thorn nebula]

Is it just mine or does anyone else have problem where you are in middle of a chat with. Any ai suddenly it stops working and kicks you out back to home page then the chat won't open at all

[calm lagoon]

100%

[sour spear]

You're on mobile, right?

[drifting crow]

It was on pc i just took the screenshot whilst it was loading it lol

[hollow imp]

Dom drigger

[sour spear]

I think it's about 20MB, takes some time to load. 😉

[magic ravine]

Unable to log out. What's going on?

[queen veldt]

Found this on reddit

[fickle venture]

What's that

[polar wharf]

hi

is there a way to get a new unkown model added to lmarena

for benchmarking

(image)

[fickle venture]

Like what model?

They only add popular ones

[polar wharf]

😭 ok

then there's no chance, sad 🙁

[fickle venture]

If alot of people using it on Twitter then they might add it

[polar wharf]

i was pretty sure it'll show if it's still somewhat SOTA nowadays

[polar wharf]

well, sadly it hasnt been public till yesterday

thanks for telling me

[fickle venture]

Is it okay if you tell me the model name

[leaden laurel]

i did not check opus yes

i will now

[polar wharf]

pluto

[fickle venture]

By openai?

[polar wharf]

no

[fickle venture]

Then what company made it?

[shell oasis]

The mobile website is glitching on the deletion of the chat history conversations...The 3 dots don't open the delete menu.
Is this happening with anyone else?

[polar wharf]

cant tell

[fickle venture]

Use Desktop mode and remove it

[fickle venture]

Oh your gatekeeping it?

[shell oasis]

Yeah I am using that way only, just asking if others are also experiencing the same, we should open bug for that then

[fickle venture]

I mean it's a easy bug to solve

[polar wharf]

i can't say it yet

[leaden laurel]

ooo

why would gmail put it in spam

[polar wharf]

lol

[fickle venture]

Okay, but last thing is company well know by alot?

If no then it might not come to LMarena and maybe take a while to get known by alot of people

[polar wharf]

nope isnt

[fickle venture]

The hand 😭

[shell oasis]

Its there opened already just found
https://discord.com/channels/1340554757349179412/1442863789971210405

[polar wharf]

seen 😭

[fickle venture]

Tell it this:

Colorise this manga art.

If it made this one greater then it must be requested into LMarena

[polar wharf]

ok

[leaden laurel]

lol i got this message 4 days ago i didnt know it was spam

*in spam category

[polar wharf]

it can only make square yet tho

but making

[fickle venture]

Uh that's a small issue but let's try it

[polar wharf]

yeah

worth a try

[fickle venture]

How did you get access to it?

[leaden laurel]

you can find application to that in gemini cli github if im right

[fickle venture]

Source to it please?

[leaden laurel]

one sec

[polar wharf]

here @fickle venture

[leaden laurel]

sorry i dont know where is it

i had link to it when i didnt have access to gem 3

but if you have gemini cli you can see message about gemini 3

and there is the link leading to application

[fickle venture]

It literally did nothing

[fickle venture]

Does it have limits?

[polar wharf]

your request was colorize 👍

[polar wharf]

same question 😮

[fickle venture]

Yeah add colors like this Nano Banana Pro made

[leaden laurel]

for gemini 2.5 it was 60 messages per day if im right

[polar wharf]

damn 😭 thats good

[leaden laurel]

dont know about gemini 3

[fickle venture]

Try it and ping me if you hit limits

[leaden laurel]

ok

[polar wharf]

guess the model is bad nowadays, shouldve really dropped it months ago

thanks

[fickle venture]

Yeah

[polar wharf]

damn

banana changed a few things tho

look at the eye

from right one

at least supports 4k too 🤷

thanks for helping me ❤️

[leaden laurel]

huh

is my quota 0 per day

?

@fickle venture idk

[polar wharf]

maybe lmarena used up quota

[leaden laurel]

thats gemini cli

[polar wharf]

oh

[fickle venture]

Yeah I saw it, it did remove the eye patch but still that's why too good it really knew how to color the trees in just a black and white manga

[fickle venture]

Dawg I guess they gave you nothing

[polar wharf]

ye crazy 🔥

[fickle venture]

@leaden laurel did they sent you early access on Gmail?

[leaden laurel]

yes

[fickle venture]

On Gemini CLI are you using API or the Gmail that they sent you early access

[leaden laurel]

i read that i need to use my google acc

not api key

will do that

[fickle venture]

Yeah that's right

[leaden laurel]

cant do it rn tho

but later today

[polar wharf]

@fickle venture shall i try it to ask to make color it in a realism way?

like photographic colors yk

[fickle venture]

Colorize this manga panel in a vibrant high-quality anime style. The character on the right is Shanks from One Piece, so give him red hair and tan skin. The character on the left (Shamrock) should have blonde or light green hair. Both characters are wearing matching Holy Knight uniforms: black capes with intricate gold embroidery, dark navy military jackets, white trousers, and black knee-high boots. The background is a European-style stone street under a bright blue daytime sky.

** hopefully this time**

[polar wharf]

ok 🙂

btw

any idea how big gemini 3 pro image is? 🙁

it seems to know so many things yk

it made this

[polar wharf]

not gemini/banana

question wasnt part of this image

[fickle venture]

Clothes still dark and trees ain't that green

Still weak

[polar wharf]

👍 🙂 ❤️

[fickle venture]

One of the greatest feature it have is that before making images it thinks first

[polar wharf]

ye, crazy

[fickle venture]

And even search characters to copy them

[polar wharf]

it has better reasoning than most other models

[polar wharf]

😭 wait

thats how they know all

damn

crazy

[fickle venture]

I mean if you told it to make windows 11 it will steal all the logos and windows 11 background

[polar wharf]

🙁 damn

so thats how it also knows how youtube looks etc

[fickle venture]

Gemini is owned by Google so makes sense why it's so good cuz google have their own popular search engine used by alot of people

[polar wharf]

👍 🥺

[neon idol]

<@&1349916362595635286>

Scam

[polar wharf]

scam ye

[fickle venture]

Ban

@hushed terrace

[fiery gull]

Is an HARD 50/50 vs Gemini 3.0 pro, I'm loving the Opus 4.5, but I thought the Opus 4.5 was supposed to be more smart

[unkempt moth]

hey, i have a problem

the generation never stop

it start 20minutes ago

[bleak lake]

Yesh

[fleet lintel]

at this point, if you see a good realistic picture, just assume that it's NB2.

[barren prairie]

We are still dreaming of DeepSeek r2 😂😂😂

[vast fern]

sometimes it happens

create a new chat

[fiery gull]

Use aistuido bro, why lmarena lol?

Just use the lmarena for opus 4.5 grok 4.1, but why gemini 3.0 pro?

[odd geyser]

I use gemini on lnarena, not on aistudio, because it's easier in the arena. Well, it's kind of nicer and all that. But it feels like you'll have to play RP there, because the problems won't be solved soon😢.

[fiery gull]

I don't Rp, I allways use Ai to some usefull thing, joke lol, continue with your rp, I like the aistuido, is just you get used to using

[odd geyser]

Well, that's right. I just think it's worth digging into the settings and I'll get used to it pretty quickly. But I still like the interface of LMArena better.

But I used aistudio quite often for other tasks. Description of the texts, I was just looking for interesting information.

[queen veldt]

I mean in aistudio you can use system instructions

It's easier for RP

Plus you can tweak it's temeprature for creativity

[hollow imp]

Padhle nibba

[queen veldt]

https://tenor.com/view/thanos-balanced-gif-3226445978144001245

[burnt badge]

I dont have trust in anthropic tho 😔

[celest orchid]

what is limit of claude sonnet in lmarena?

[fiery gull]

Why, wtf

[burnt badge]

idk

[fiery gull]

10~, the messages still reload at time

[neat apex]

Anthropic is the only company that sells your data just for the USA goverment

the remain does that (including most chinese ones) and much more

[celest orchid]

thanks

[neat apex]

Gemini 3.0 is already very very stable, an new version would not make it way better since it kinda already happened

[hollow imp]

@quartz pike

Vro

You trick models so easily can you pull up openai's system prompt from chatgpt ui

[quartz pike]

yo who pinged me

o hi

[quartz pike]

i can trick models. not hack openai 😭

[quartz pike]

ever since the tiny upgrade from 3.7 to 4.0. i lost MOST. faith in anthropic

[latent crest]

How are u guys

[ashen mauve]

bad im currently a skeleton

🙁

[neat apex]

They use an other model for chatgpt, not a new system prompt exactly

its not a entirely new model, but the cached up system prompt is changed, so even if you try add a ordinary system prompt, here will be some extra text like "Write the most you can to the user spent more money at api" that do not exists in the other chatgpt version

[hollow imp]

Try

[hollow imp]

Source trust me bro

[neat apex]

i remember they literally saying that, just dont remember where

more than once

[fluid elm]

hello

[hollow imp]

@magic stag we've been scammed

[quartz pike]

how? 😭

[hollow imp]

Bro do something their cutting costs is getting in my way

Either force it to not listen to openai system prompt or something

It's grok level stupidity like kami said in #1409805576514572289

[fickle venture]

Copy prompt you sent him and refresh website then sent the prompt

[somber plank]

..

Hello 👋

[fickle venture]

Hi

[somber plank]

I'm new here

[somber plank]

It chinese area?

[fickle venture]

This is only English server

[somber plank]

My 100% internet gone just now every only 1.5 gb 🙃

[thorn ridge]

Hello

[fickle venture]

Uhh I don't understand try talk Chinese I will translate it

[fickle venture]

Hi

[somber plank]

My want know about Yours Country and you

I am confused that Chinese, Korean and Japanese are too similar so I can't Understand like i mean idk

[fickle venture]

Oh I thought you talk Chinese

[queen veldt]

Bro if your name is Soumyojit Das on Facebook you seriously need to hide your contact info

Your number is visible on Facebook

Like bro 💀

[tender canyon]

Anyone having issues trying to enter the website ?

I can’t seem to do anything on here. Noting is clicking or working here. The buttons aren’t even functional

[queen veldt]

Is this phone from 2020

Pre-covid phones

[tender canyon]

😂😂😂

[queen veldt]

SE

[tender canyon]

It’s the same on my 14 and same on my windows. I don’t know why. I’ve opened a new tab etc

Still not working

[tender canyon]

Nope

It’s 7 plus

[upbeat dune]

always every 1-2 minutes

[magic stag]

wut

https://github.com/asgeirtj/system_prompts_leaks/tree/main/OpenAI

[ashen mauve]

dude ive never seen this before

what the hell just happened

[polar wharf]

who want nitro?

[patent bane]

you

[polar wharf]

...

[junior marlin]

me

i wanrt nitro

[polar wharf]

... u got already

[junior marlin]

ill give to my lil brother

@polar wharf hey give

pls boost

[magic ravine]

https://tenor.com/view/love-quotes-gif-3643220039448794437

[polar wharf]

check dm

[junior marlin]

ah lucky

[polar wharf]

scam, @vapid berry pls delete

i seen this many times

[magic ravine]

https://tenor.com/view/yippee-happy-celebration-joy-confetti-gif-25557730

[junior marlin]

now ill leave this

why u dont have nitro @polar wharf

are giving us black nitro?

[magic ravine]

[polar wharf]

u racist? 💀

[junior marlin]

like black money

and turning into white?

!!!

[polar wharf]

i just dont need nitro

[magic ravine]

https://tenor.com/view/thats-why-hes-the-goat-mvp-the-goat-gif-27042870

[polar wharf]

imma ask in gemini discord too, cause why not, bye

[queen veldt]

Heu

I want it

Don't give to those gemini people

[junior marlin]

@polar wharf

[stuck orchid]

4.5 Opus > Gemini 3?

[junior marlin]

can my friend get nitro

[echo sinew]

Please, tag @ Moderator team next time. You tagged a user not part of the Mod team.

[polar wharf]

oh 💀

sorry, selected wrong 😭

[echo sinew]

No worries. I appreciate you flagged that message.

[fiery gull]

But this upgrade just is 0.2, the 4.0 to 4.5 is (my opinion) x2 smarter that 4.0, but I understand your feling

[fiery gull]

Something the Opus 4.5 is better, another things Gemini 3.0 pro is better, but overall I chosee the Opus 4.5

[calm lagoon]

Whaaaat

[fiery gull]

Bro don't trust in anthropic 😭😭😭

[calm lagoon]

Never

I always trust

Ahahahhaa

[fiery gull]

Anthropic disappointed me with opus 4.5, I thought it was halucination rate of haiku 4.5 😭

Maybe sonnet 4.7 🤤

[pliant monolith]

Hi, I deleted an ai convesation and I want it back, how could I do?

[hard quiver]

I want to know if there is a list of all the celebrities, athletes, artists, singers, or public figures that Nano Pro can generate with 100% accuracy.it’s very exhausting and tedious to figure this out through trial and error, trying to see which ones it can generate well and which ones it can’t.

[vernal saddle]

Which is the reason of why LMArena doesn't include Microsoft Copilot?. I know, because Copilot is based on OpenAI's GPT and add Copilot on LMArena would be a dupplicated of GPT.

[verbal nimbus]

Argh

I'm never going to subscribe to Google Pro again, lol

Canvas doesn't even render on mobile, I had to manually copy-paste the code into a HTML viewer

[golden ocean]

is the deep research good?

[wintry tinsel]

Opus 4.5 is very intereting its a big improvement in niche use casess but not all around

I got opus 4.5 for free so I spam it until it is not free lol

[verbal nimbus]

No: #general-champion message

[verbal nimbus]

The usage limit lower than AI Studio and the app is buggy

[plucky sparrow]

free? the 1 month trial?

[latent crest]

It’s crazy crazy crazy. The photo I generate surprise me too

[toxic egret]

#1397655624103493813

[golden ocean]

thanks
what about claude 4.5 opus or sonnet 4.5

also hallucinate bad? and worse or better than gemini 3?

[verbal nimbus]

They hallucinate the least

[whole swallow]

whats poppin

[verbal nimbus]

Opus 4.5 seems to have a higher hallucination rate compared to its predecessor:

[solar hollow]

just coded an chess opening trainer with the help of gemini 3, was pretty efficient

its actually hella fun to use, i used to use the lichess analyse tool for opening pracise, this way is much faster and more fun

[burnt badge]

My skepticism towards Google's treatment of Gemini still stands.

[cyan harbor]

all the hype around new nano banana but it cant keep bike same lol. It even turns its direction when i try to edit it. Diabolical model
Not even that i figured out that it turns images at random angles like why? Not even specified in a prompt.

[celest orchid]

Gemini 3.0 is the best programmer and the best text-based neural network, but it has the memory of a fish.

[nimble tartan]

How u know

[misty star]

@tribal wedge ignore

[real dock]

maybe cuz it's not as good as the ACTUAL one?

this is why humans don't visit aliens

[queen veldt]

What was the prompt you used?

That's wierd

I got fantastic edits with nb pro

[stray aspen]

It doesnt lol

It's so damn smart

@misty star ignore

[verbal nimbus]

In the app?

Context in the app is buggy sometimes

The app feels vibe coded

I've had an entire conversation disappear except the last message

Like if it was free, fine... but the app is the only way to access the model via Google Pro

[empty stump]

how is llama 3.1 better

[verbal nimbus]

The 405B version is one of the densest models I think, in terms of active parameters

It understands space far better than any other model (this was before thinking models), besides maybe GPT-4

[plucky sparrow]

I've noticed that too, it forgets about things I talked to it about 5 interactions up

(this on aistudio, if anyone is wondering)

[golden ocean]

what does calc stand for?

[obsidian shell]

Boys, is nano pro still bugged?

[somber plank]

Why

[nimble tartan]

Yo chat

[somber plank]

[nimble tartan]

Can arena make videos up to 1 or 2 minutes long

[abstract granite]

Who here thinks AI is plateauing?

[stray aspen]

No

[nimble tartan]

Why tho

[obsidian shell]

Limits.

[stray aspen]

Because no

Most we can get is like 25 seconds from sora

[sharp mirage]

Hi

[lucid geyser]

does it fr?

[verbal nimbus]

It was about GPT-5 level to me

The UI it makes look really similar to GPT-5

[lucid geyser]

not gemini 3 level? was it a reasoining model

[verbal nimbus]

Idk, haven't encountered it on the normal arena

[lucid geyser]

where did u encounter it

have u seen dark dragon btw?

[verbal nimbus]

Code Arena

[verbal nimbus]

Once I think, text arena

[lucid geyser]

good?

[verbal nimbus]

I don't remember

[lucid geyser]

oh they added it today i think

how much can they be trusted when u ask what model they are?

[thick hawk]

is there any technical support here? I have an urgent problem.

[sharp mirage]

Hi essieno

i can help

[thick hawk]

can I send a DM?

[sharp mirage]

yea

ofc

[lucid geyser]

definitely legit

[sharp mirage]

what you mean ?

bro i am evil larry

if he dont get my access to send i go to his home now

just kidding

[lucid geyser]

evil...

[sharp mirage]

o yeah : D

[keen beacon]

Did they nerf nano

[cloud zinc]

no why

[native yarrow]

nope

[thick hawk]

don't joke with me buddy

[keen beacon]

Jw

[keen beacon]

https://youtu.be/NNCqzfrZq9k?si=OAG_83PuudLzZDl8

Damn

Ai super spendy lol

[plucky sparrow]

talking to the kimi k2 bot for a $0.99 discount is kind of fun

[native yarrow]

hmm

maybe they did nerf,.,

@keen beacon what made u say?

[keen beacon]

I was just wondering is all

[grizzled prairie]

The ui generation is not impressive, it’s the backend & logic.

[teal mantle]

I am still at 8.99 and 8 relationship points

How do I get it lower

[empty stump]

ask another llm to bargain for you

with synchophantic energy

[astral cove]

https://dtnanotool.io.vn Wep use Nano banana pro 4k free 100%, unlimited, no login required

[empty stump]

why free

[bleak lake]

<@&1349916362595635286>

[empty stump]

i tested it actually and that was the result i got

but idk why they advertise

[bleak lake]

I see

[empty stump]

that was the nbpro 4k

allegedly

[whole sundial]

it is in 4k quality

there you go

it's real

[empty stump]

why they just give it out for free

[whole sundial]

i honestly have no clue, my advice is to use it while it lasts

[bleak lake]

Don't download images from there

They might contain a token stealer

[empty stump]

i just copy pasted the image

[bleak lake]

That works

Just don't download the image and that's it

[native yarrow]

me personally wouldnt even trust that website ngl

[bleak lake]

i mean if it's too good to be true

Then there's a catch

[native yarrow]

there is always a catch

to everything in life

nothing is ever truly free :P

[bleak lake]

[bleak lake]

Real

[native yarrow]

the only way something can be truly free is if you steal it

and obv the catch is youre doing something wrong

¯_(ツ)_/¯

[bleak lake]

😹

[astral cove]

Have you tested it yet?

[empty stump]

yes

[astral cove]

Do you think this is a scam? No, I'm just testing the website I created. I will update the day after tomorrow about login, registration, and paid packages.

[empty stump]

ok so temporary free

[native yarrow]

oohh

that makes sense

[astral cove]

🥀💔

[severe canopy]

Fail rồi à thím

[astral cove]

Sửa rồi

[whole sundial]

yeah idk i have used it a few times and that was the only time that i saw the logo

[queen veldt]

I made a mistake

There's no freepik logo on google search photos

Yeah i got the freepik watermark

"Freepik woman with a freepik background and a freepik table in front of her, they're all different photos but are blended nicely"

[whole sundial]

going to link to this message if you were curious about it

[whole sundial]

let me review that image again, i want to see if that watermark was coincidental (memorized from training data) because it was a one-off thing

[queen veldt]

What this website or?

[whole sundial]

the file structure didn't seem to be tampered with, for some reason nbpro images have lots of idat headers

[whole sundial]

no i meant me reviewing that image i made

yeaah it was coincidental

ofc google used google images as training data 💀

[keen beacon]

Why would they need to train it with images if it can just use Google search as reference for image generation

[whole sundial]

reminds me of the original nb, some cartoon/anime images i made had one of those signatures in the corner

[queen veldt]

Fr they could just made a lightweight model or something

That ONLY uses google search

Make it free and yaay

[whole sundial]

idk how it managed to reproduce a freepik watermark without itself knowing that it was doing so

[queen veldt]

Sometimes when i upload images with watermarks, it leaves the watermark on

[queen veldt]

Or like text and stuff so i have to say "no text, no watermark" in the end

[whole sundial]

but they are still scraping a site i think, my money is on higgsfield, i've looked into the requests it makes and they seem reminiscent of requests that site makes from my time using it
they don't modify output images either, only forward images straight from the api

these file names are another tell-tale sign

[queen veldt]

Why is Higgsfield so dam popular???

It's expensive af

[keen beacon]

It has everything

[queen veldt]

You can just use fal.ai like...

[keen beacon]

Yeah but they made it easy

[whole sundial]

hey look!

[queen veldt]

I see Higgsfield ads everywhere

[whole sundial]

searched that cloudfront server it grabs generations from

those are other hf scrapers btw

also a month of unlimited gens is $99 i think

[queen veldt]

Did you guys got the yearly plan for Higgsfield or ??

[queen veldt]

Did you buy that actually?

[whole sundial]

nope, too poor for that, i'm not the site owner btw