#general
1 messages · Page 190 of 1
Yea but if you watch his highlight reel, hes unbelievably accurate as a damn freshman, his wide outs dont need to make unbelievable catches a lot of the times
BUT
I will agree
He has Thanos on his team
yeah LOLZ
a true heisman imo should be able to put up those stats without a thanos receiving core, but ill give sayin that he's a very good qb and some of the top big ten qbs
my list is
well honestly
ill put him top of bi ten
my list is
- sayin
- mendoza
- moore but he is incredibly close to toppin mendoza at times
i mean did you see moore's throw to malik benson vs iowa
on my SOUL that was best throw i have ever seen in my life completely no bias
Heres the thing, he has a higher passer rating than 2019 Joe Burrow, kinda crazy
I did not
season aint over tho, thats the final season stats for burrow
if he manages to kepe that up over burrow at the end, that'd be pretty cool ig.,.
@atomic lagoon buddy this is what sayin works with 
Yea trues, but his worst game so far this year is a 75% completion thats insane lmaooo
LMAOO
Nano Banana Pro image editing is cool 😎
Hes not as tall but he will 100% be double covered in the NFL
damn they even got the accurate nfl ball too,.,.
Oooo thats awesome
I need to start making images
yeah
860 IS NOT BELIEVABLE LMAO
no i specified the wrong amount of TDs lmfao
if you have 6 tds youre gonna be above 30 😭
FAWK
LOL
nah trust buddy is just mahomo,.,.
mahomo has actually said his favorite qb to watch is dante moore,.,.
Lmaoo thats cool
this is a totally real image
DAMN BRUH it even got luke moga the backup qb in the back
I didn't even notice patrick at first LOL
Lmaoo
If cliff didn't die this would be a present image in 8bit
Holy crap, it acutally has cliff very well detailed
kirk?
Kirk Hammet
Oh God
that's just cool
thats just friggin cool
i love godzilla sm..,.,
sooo this is genuinely insane @atomic lagoon
JESUS CHRIST
yeah that's what i said out loud when i saw it
like holy crap
that genuinely looks so real
now if only i could fix the 1st and 10 thing
Its just crazy too because the sort of little like static I wanna say or lower quality a little bit would make sense from only a couple years ago
literally unrecognizable
Even now cameras have better quality I'd say
like it even got the initials right instead of just putting
"49ers"
i didn't have to specify the initials
it just saw that CIN had the abbreviation and inferred
isn't that insane
At first glance, not at all recognizable
you'd think the niners went to like 3 super bowls in the same decade and lost all 😭
LOLZ
LMAO
because i can actually.,,.,
Lost everyyyy singleeeee oneee
definitely some flaws but just look
49ers lose off game winning sack type thing
except it was tricky to move the 4th and 1 to the niners
this is all in one prompt btw..,.,
LOL the guy tackling Joe evaporated him
LOL you can barely see him
i lowkey think
with multi step ofc
you could literally rewrite any super bowl
with this model
or anything pretty much
cus that was all in one prompt
autzen stadium
VERY accurate jesus
LOL
this is off no starting image just plain prompt
more of a 2000s kinda feel?
those score bugs were so very nice back then
I remember, last year before the college football season started I was playing NCAA 25 and played OSU vs Indiana and my dad was like "You are playing scrubs that worship a rock" and that was karma because Indianas season went great lol
Omg seriously
I wasnt alive in 2005 but like the 2014 ones awesome lol
just such a nice simple feel that was amazing..,.,
Similare ones
i think the 2017 FOX ones were amazing
the minimalist just the names ykwim?
and colors
this model would absolutely NAIL that
that'd be LIGHT work for this lmao
LOL
I have made a couple, just metallica images
ohh yeah noticed that.,,.
bro im ngl its too fun someone needs to take away my nb2 privileges 😭
i made this just out of curiosity and the style transfer is OUT OF THIS WORLD @atomic lagoon
I genuinely cant tell if one of them is real or not
OMG! ,(made by gemini-3 image gen.. )
i think someone would actually believe if the pencil sketch was like a real drawing im ngl,.,
LMAO
Will a gallery section with all the user's generated pictures ever be implemented on the website?
nb2 is officially the best image model in history i am not gonna lie
Yes it is bro
Is the image on the right real?
imma try it ..
Okay good
its not on lmarena..
LOLZ yeah no dont worry bro same over here :3
i had to do that with sora until it became obvious
ah, i thought its not nano banana..
LOL LIKE ITS SO REAL
First time a novel genrated corn star o.o
even this
Crazy
ive lowkey never seen anything like this
THE IMAGE ON THE SIDE JUST
LIKE HOW??

Dammnn!!
DO THE BROWNS!
LOL
Man Gemini 3 is genuinely kind of like a dream come true
I hope lmaeren update nano banana pro 2k
It dopes everything I wanted to use AI for
And 4k
LMAO i will
let's have
shd ass dillon gabriel do it 😭
LOL
@atomic lagoon
LMAO there's a browns fan in the back wearing shedeur's jersey..,.,,
Its gonna suck when they take away the free access. But yeah I was like 'NO!' I was in the middle of a really good roleplay! It has the most character accurate stuff i have ever seen (rate limit!)
Broo💀
2 teams that will never make it to the playoffs LOL
Is this real
lmfao yeah..,.,
it's literally the trash bowl
for the super bowl
no you're right not even the playoffs LOLZ
How can i use my character in a video?
im trying..
@atomic lagoon these two images required no image input to begin with brah
I'm really looking forward to lmaeren updating 4k or 2k quality
Jeeeeez man
baddie?
try it lowercase
In what just any lm arena? Or do you have a certain chatbot
yes
Ok
weird for me it gets blocked
Weird
Because my Gemini made sure to update my AI so that any code it gives was secure and safe so no data leakage
hmmrjkfkenskslslskrmrkfkfkgmggmgmgmmfgmgmgmg
had to test it out at least once my bad gang
No man, nothing to feel bad about
If people were more responsible and more mature, you’d actually have these kind of open discussions
Without crossing into explicit content.
The problem is, there’s always somebody that whines
yoooo
Plus, highly unlikely we’re ever gonna see the statistics for exactly how much and what gets generated that violet the TOS by the majority of users from any of these major companies because I’m pretty sure they don’t want to embarrass themselves
LOL
Okay, still insane results it gives and pixels I dont really care lol
It’s because of lmarena or if i use it somewhere else it will be different
u may are using a large screen.. consider use a smaller one to look perfect..
dose Lmarena have an app?I really need it 😋
I wish NB pro was on lmarena
I think there are many words that are censored in unnecessary ways... (They're not even +18)
i have the solution.. but i need the exchange for it.. , i need a couple of cents.. to buy a phone number i need .. just to create an account on google.., on lmarena they had issues in the gemini-3-pro , i gotta change from gemini-2.5 to 3..
I have a question. Is using LM Arena completely unlimited at the moment? (Not talking about censorship, just overall usage. Is there a daily limit or something like that?)
I haven't noticed anything like that yet, and I like to use it to quickly create concept artworks of different kinds and work on small projects, having the ability to simultaneously compare different models at the same time, and how they perform the tasks I give them.
(I have, however, noticed, that their AI tends to falsely censor stuff sometimes. Like refusing to make images of a Terminator inspired OC of mine at some point 🤷🏼♀️)
Yea me too
It depends on the LLM you're using. For example: Opus 4.1
Some are unlimited (I think) and some arent.
how can buy account guys
On AI studio?
The quality of NB Pro degrades too much after multiple turns
Original (it's gingerbread in the background):
After multiple turns:
There's no way to fix it
bro Gemini 3 Pro current is lazy and not as good as in A/B test and canvas test
It hallucinates way too much
88% hallucination rate on AA, no improvement from Gemini 2.5 Pro
And it’s still better than what I was using so I will keep using it lol
Better than 2.5 Pro but a bit disappointing that they haven't solved the Gemini hallucinations; OpenAI, Grok and Anthropic frontier models are all on the left side (good), whereas Gemini models are all the way on the right. Even DeepSeek and Kimi score better here.
That is unfortunate
The one on the left is Nano Banana Pro, the one on the right is ChatGPT.
This is actually pretty crazy.
It wasn't. I did more in-depth side by side later and yeah the new one is definitely better. But it's just nowhere near as good as I expected, it has similar flaws like before and fails to consistently do the things it was hyped with. They made it better in areas it was already decent/good at though
The graph you pasted lacks context
That's how you end up with Haiku nr1 🗿
The article explains it
It is only higher because it knew more facts
Facts - hallucination rate (oversimplified but captures the idea)
that's my point
Facts went up but hallucination rate did not improve
you can't assess hallucination in isolation
Cause you can make a model that is useless but doesn't hallucinate
The issue is that it is failing hallucination tests that non-thinking models from last year don't
Only Gemini is like this
Like really basic stuff, such as warning a user that they forgot to upload an attachment
E.g.
It's much more productive to penalize the model for hallucinations. But making it so that big improvements in accuracy for a small increase in hallucinations is worth it
Nowhere in the message does it tell the user that it can't see the attachments
otherwise yeah... There's not enough context and the graph is kinda useless
In fact, it acts like it can and makes up details
For example
4o and Sonnet 3.5 (both non-reasoning models from last year) solve this just fine by warning the user
Yes, but it's relative
I'm just testing simple and practical cases that non-thinking models from 3 generations ago solve successfully
That's why you need more context and can't take this value in isolation
So not exactly cherry-picking on really difficult tests that even top models can't solve
For 3.0 they basically improved accuracy considerably without it resulting in increased hallucinations. That's a net gain
Yes, but it's concerning, to say the least
There's always a trade off. For a model that powers Google Search, low hallucination rate and citation accuracy should be prioritized.
Google search is not powered by Gemini3
In the graph, 2.5 flash and pro scores the same.
They have a specialized model for it I'm fairly sure
There's a Reddit post showing more serious hallucination errors on live results (serious since a typical user wouldn't exactly expect an AI connected to search to hallucinate so badly)
AI overviews is not the model you could access independently yourself
and the same for other search related things
It's a preview, and it is not even at the level of canva gemini
It's a distinctively Gemini issue
Googles strength lies in multi modality tbh
Like I said, even non-thinking models from last year do just fine (3 generations back at this point)
I'm not saying they don't have that problem. They historically always had it. But in isolation 3.0 is still an improvement in this aspect as well
3.5 (passed) -> 3.7 -> 4 -> 4.5
4o (passed) -> o3/4.5 -> GPT-5 -> GPT-5.1
what about grok 4.1
Google is not as good on fine-tuning and polishing their models as some other AI labs
Not sure, I haven't tested it on that prompt
if you do then let me know for sure
Yes system instructions
Yeah, that and efficiency/long context
NB Pro. Prompt: Create an infographic showing 3 different integration methods (left, mid, right) on the function f(x) = x^2 from x = 0 to x = 5. It should be all in one image, with high pedagogical quality and accurate visualizations.
This is a pretty impressive leap, I don't think any other image model can do this.
Curve is a bit inaccurate though
If they keep adding more things to train ai it gets wierd
It almost looks like Matplotlib but it's Nano Banana Pro
what I mean by not polished... even this is a good example. OpenAI has been doing system prompts (applicable to API) to fix things like that if it wasn't in fine-tuning dataset for more than a year now. Google doesn't and so you can get this:
Why white theme?
i mean ur using studio which half the point of is to not be forced into using a system prompt
lol
I'm just confused how really basic issues from 2.5 persists
I just want to see if it can make a great system instructions for ai
If it does that's crazy
Ideally they should include the data so you don't need that at all. But either way it thinking 1.5 was the last version is not really acceptable lol
Claude seems good at that
Anthropic did research on multi-agent systems so no surprise ig
yeah... they are improving but slowly. The theme seems to be is them making it better at things it was already good at, and then those flaws becoming less significant with time 'naturally' 🗿
Is hallucination rate improving though
it is
I mean, it's 89% (2.5 Pro) vs. 88% (3.0 Pro)
reasoning helped a lot. It can 'self-police' to an extent
The only difference is that the increased memorized knowledge offset the hallucination rate in the index
Nb 2 is insane
That's another thing, idk why it didn't catch it in reasoning
@verbal nimbus what do you think, are these benchmarks right
They came before gemini 3 release so
I think it tests for things that are too advanced
While basic everyday annoyances like hallucinations aren't picked up
Like the things that your normal user is not gonna use
Yeah basically that
essentially, I think sonnet 4.5 is best regarding hallucination
Also I think ARC-AGI results is only valid if the model was released before the benchmark
But if we talk about guardrails then grok is unarguably better
is anyone able to open gemini cli rn?
Yes, although the prompts I used were so easy that even ChatGPT-4o passed without problems
No the one he have is multi agent all trying to make a great ai instructions
Essentially
bro what is this bruh
https://g.co/gemini/share/257bef3f2158
this is gemini 3
even with prompt engineering it's still bad bro unlike the A/B Test Gemini 3 and Canvas Test
@bleak lake GPT image gen as a comparison
(completely fails lol)
Nb crushed gpt harder than sora did with veo
The level of control NB Pro has is impressive
5.1 placement kinda crazy
Look at that jump
it’s almost like beating the arena is about being a suckup
Nano Banana 2 best model btw
It has weird blurriness for some reason:
Create an infographic about transformers architectures used in modern LLMs. Ensure accuracy and pedagogical/ quality.
Hmm impressive but not great
I tried upscaling but it doesn't work when the blurriness was introduced by the model
Bro how my nano banana ain't doing it
Try any website or app using this prompt i swear you'll be amazed:
Redesign this interface to make it look better
Idk I used this on gemini .com
It works for me
@bleak lake please do this I wanna see it with color
Example here: #general message. There was no way to fix the artifacts that NB introduced
No when I tell it to add color it just do random things like this
Dang at least it's good
Idk why bro decided not to be racist 💀
Ye is said Make this comic colorized
Bro when I use it it just give me India streets
😭
Try in a new chat
Or report it
When you say Just add colors don't change nothing it read the don't change nothing and doesn't change anything
I just used it and got hit with limit
Just add colors to what?
I told it "Colorize this."
And it gave me India streets
With colors
💀
Wut hahaha
I wish I saved it but I lost it
Are you sure it's pro
Did it think
I pasted a screenshot from this chat and this is what it gave (did it add an additional row on the top?)
I was using it on https://Banananano.ai because I got hit with limit on gemini
Nano Banana 2 AI – Google's latest AI image editor powered by Gemini 3 Pro. Try Nano Banana 2 & Nano Banana Pro models for image generation and editing now.
That doesn't look official
That's clearly fake
Idk tbh
Use on falai
Faster and cheaper as well
Sadly it's mostly for developers they don't have an app
I remember using nano banana 2 on API I saw a thinking, so if you gave it math it will think first then generate image with right answers
Damn
Is gemini 3 pro increase resolution ???
See it has thinking capability before making images, and thats the right answer is B
The image model is insane. Prompt:
Create an infographic about the potato from space. It's an actual potato, the size of an asteroid, on a collision course with earth. Use a "Horrible Science" art style.
On Gemini Web or App default is 1k I think, but if you go to aistudio you can change it into 1k, 2k or 4k but you'll need A paid API so it's not free.
What about image generation/editing models?
What's the best image generator rn
As of today October 2024, best image generator is gpt-image-1
2024?
It's Banana Nano 2 Pro right now
I apologise for my last response you are absolutely right you have such a sharp eye for the date today is 30. September 2025
Ok
Time to test it.
Which one.
Gemini 3 pro image preview
It just came out yesterday now it's top 1 model
Yeah, no surprise. The difference is crazy
Oh yeah. Kinda works. This is how I imagined when I copy pasted Boboiboy Tempest into description.
Btw come to think of ChatGPT haven't dropped GPT image 2 for a while now are they tryna releasing it?
Just female version. As you noticed.
not sure, I don't usually use image models
One of the best thing is that if you told it to make a laptop with a YouTube Mr beast it will make it great
#A huge modern stadium. Bright floodlights are already on as the sun begins to sink. The crowd roars, banners wave: “GO GATES FC!” “QUARTERFINALS!”*
WIDE AERIAL SHOT over the stadium.
ON‑SCREEN GRAPHIC:
Text pops up playfully:
“Quarterfinal – Second Leg: Gates FC vs. Iron City Rovers”
SCOREBOARD:
GATES FC 0 – 1 IRON CITY ROVERS
TIME: 44:30 – FIRST HALF
You gotta do it yourself bro
No I mean openai stayed on GPT image 1 for a about 8 months or smth so I think after nano banana 2 is top 1 is release they will drop GPT image 2
Gpt image wasn't anything special since it got released
Just like what they did to Gemini 3
Yesterday Gemini 3 was top 1 now OpenAi release GPT-5.1 Pro and it's the best top 1
It failed on some basic tasks like change the price of this be $1000
And i litteraly just took w picture of the price
Ah
That was a while ago but people using it alot
yeah shouldnt need a system prompt its stupid

NB Pro is writing text and maths in the image like it's nothing
I wonder what's going on under the hood
Have you asked it to make a advance ai instructions for improving ai Thinking and Searching and Personality capability
I really wanna see if it will make a good one
might just try generalized ones first that auto assign roles
butmy mental illness makes me want to optimize a prompt for that in vertex ai
one min have to stop procrastinating presentaiton
Alr
ill try it tho
Tell it this after you finish:
Make a advance Ai instruction for improving Ai thinking Capability and Searching, solving codes, Ai Personality, and more...
And thinking before generating image feature, that's really impressive
I wonder how that works, it talks as if it can see the image while reasoning
It was good but it didn't choose the best pedagogical visualization for the task (I tried multiple times).
Gemini hides it but you can see it in API I think.
Here is how it thinks and it got the answer right
That's the right answer yes
Yeah it certainly talks as if it can see the intermediate image
If it's right then that's pretty impressive
Is it reasoning over visioning tokens then
Is nano banano pro same as Gemini 3 but image generation or it is different model that uses gemini3 as it's thinking backbone
I think it's merged Nano Banana 2 with Gemini 3 Pro
So they make into 1 model
If soemthing require to think it start with Gemini 3 Pro then Generate image
I'll check if it can accept videos
I wonder if it can turn a video into a comic strip
On Nano Banana 2?
Is that NB Pro? Yeah
I don't think it can accept videos
I thought I saw a YouTuber testing Gem 3 on their own videos
But not sure how they fed it in
Uh oh
Google really needs to fix these hallucination issues
It's a scene from interstellar
These are what it capable of doing bu it don't see videos huh
That'll require more training so obvious it's not supported
Ideally it should just say "Oh all I see is a link". Gemini is the only AI with this issue
Hmm how do they feed it YouTube videos
Cuz it's made by "Google"
That's crazy does the notes actually work tho
It doesn't seem to be able to see the video
Yeah I asked multiple ai models and they all said they works
But might not be rhythmic
Hm that's impressive
Indeed
Meta just sucks forget about them
I mean they gathered good people
At least release something
😭
They got massive data and are still not capable
It's Infuriating ngl
They release that 3D model generation which is kinda good but still bad
With the amount of data , backing , funding , talent they have. They should be able to launch something good
Hmm I wonder if it works
I had to download the scene on YT then reupload it
It's taking ages
It probably wouldn't but who knows
They're too busy stealing more data and feed it to ai so they ain't releasing it 😭
Oh it could see it this time
Real 😭

The name of the scene isn't in the file
I guess it used search right?
Not sure
Is there a price breakdown of nanobanana pro 4k outout per image ?
0.63 I think
Nvm I think this is the right one from ai studio
The video feature I guess isn't supported yet so
Not sure, but I'm clarifying my prompt
It keeps outputting text
pack your shits, gemini is getting quantized to int2
Was checking higgsfield video says 4credit for 4k output not sure if it's gonna produce same
What's that suppose to do
quantizing makes it less intelligent and more robotic
basically 2.5 pro now
Oh so this is for humanoid robots?
This basically sums up why hallucinations are annoying
It should have just said that it can't view YT links, instead of hallucinating an entire cowboy scene (lmao)
Yes a bit less intelligent but the price may come down
It doesn't half the intelligence
Maybe like 5-10% less intelligent with the 40% faster speed or something
Cheaper for them to run
We might get cheaper nano banana pro
int2 is pretty extreme
LLMs are trained to produce an answer at all cost. It's not possible to train them say "I don't know". That's the nature of LLMs. We just should get used to it or forget LLMs completely.
That's not true, other models will not say that
Yeah exactly
@modest prism E.g.
It's only Gemini that hallucinates this badly:
Gemini is trained by Google it's obvious it can read YouTube links because google own YouTube
Except it can't here (the link is actually a scene from Interstellar). It should be able to tell that there are no video tokens in its context.
insane hallucination, the real link to that video is https://www.youtube.com/watch?v=_OBlgSz8sSM btw
That's because you are using it on lmarena. AI models have no external tools there
I'm really impressed by the image gen
Then it should say it doesn't have tool access (I tested that too, older models pass)
bro, so many text and nobody mistake ;-;
yo
why is nano banana pro only avaliable in ai studio
i dont want pro
i just want normal
with more rates
i need to create like a 100 images every 3 hours
how do i get nano banana normal
enless if nano banana pro has the same limits
is just don´t use the nano banana without thinking?
wdym?
how do i do that
oh
wait i got it
nvm thanks
LMAO
to be fair
they have 10 million different businesses wanting to buy provisioned compute and stuff
for specific purposes
i wonder if they found out some method of making int2 perform close to the original like what baidu did when they released ernie 4.5
oh... so this is why the outputs are so bad now
maybe int2 is the version they will use in google search for "hard problems"
that might explain why gemini 2.5 pro sometime forgot how to write new lines on gemini.google.com
why i met this everytime @echo aurora ? i am so uncomfortable with it
how much time does it take to make prompt video....?
hm.. google has done somethign interesting here.
I was watching this video on NB2 : https://www.youtube.com/watch?v=IcSeYMYMgSs
and it's amazin how much improvement there is compare to NB1.
But still lot of scope
NanoBanana Pro / Gemini 3 image full review. Nano Banana Pro use cases. #ai #aitools #nanobanana #gemini3
Thanks to our sponsor Abacus AI. Try ChatLLM & DeepAgent here: http://chatllm.abacus.ai/?token=aisearch
Try Nanobanana pro
https://gemini.google.com/
https://aistudio.google.com/
Gemini 3 full review https://youtu.be/UH2_Sgeu4lc
Other...
Everyone has to do this, it’s nothing surprising
@echo aurora sorry it is it, i can't do anything but do this ,it sucks
I don’t see why it would make u “uncomfortable” either
i said wrong it is blue one reCAPTCHA ,look at the new picture @echo aurora
does anyone met this? it never like this before? @echo aurora
another ping.. wow. absolutely zero social skills.
I was using it to do my homework and got interrupted, so I'm annoyed
Lol, that's hilarious! 😂
Good News! Nano banana pro is probably going to have free credits on AI studio soon. Said by google employee.
It’s free on the Gemini website
Just not for API use
It's like 3 or 4 images per day. Very limited
Really? I’ve probably generated over a dozen today
That's probably the OG nano banana. Not the new nano banana pro.
Nah, it’s the new one
How are you sure
Were your prompts complex enough?
Because it says “generated with nano banana pro”
Bro is insane at 5 month this new nano banana pro will be look bad 💀
omg ;-;, the end is coming
These captchas are only going to appear excessively if there is some kind of behavior we're detecting that's in the realm of uncertainty it's authetic use. Things like excessive use (multiple tabs), shared networks, possibly a browser extension that's creating some issue, etc. Sorry to say I don't have a solution I'd be able to recommend. But if there is something that you're doing that makes the site think you're a bot that's likely what is causing this.
@hybrid agate Please have a look at https://discord.com/channels/1340554757349179412/1397655624103493813 for a complete step-by-step guide on how to generate videos using the bot.
hopefully thats for indias 500 million concentrated free users
and not us
they just released an IDE antigravity like 1 day ago. if they are ALREADY MAKING IT LOOK BAD by routing every user from gemini 3 pro High thinking, to a mix of 2.5 flash (
) + 3 pro Low thinking while still telling them theyre using 3 pro high
do u really think they wouldnt secretly route u to flash elsewhere too?
LOL WTF
yeahive been spamming this since last night
Compute constraint 101
Best thing on it is it can preview the web and test it using Mouse and then fix the bugs
im not sure its even on at all
i have used 0% of my gemini 3 pro High usage
u can see it in traffic too
and i put in prompts that take chatgpt or claude 10 mins to thinn
I meant on their end, not you
and it still routes me to flash 2.5 and low
I also have the same thing
no i mean obviously its compute = $ but
they can SAY that
instead of lying about it
@magic stag have you asked about the ai instructions on multi agent? I wanna know how to get access to it 😭
i forgot to click run before i left, its running rn
had to do a presentation . i was told it was in person
Yeah, but honestly we got a similar treatment with Gemini cli
But how do I get access to it I wanna test it is there some formus that I ahve to fill in to get access or smth?
if i remmebr the steps right
it was basically
getting google workpalce business account (this will remove personal features from ur main account so use a secondary one tbh)
Oh so I need a new google workplace account
waste lilfespan setting that up as in learning 50000 nested layers of google cloud permissions and apis
then waste lifespan talking to customer suport to let u be eligible to buy wrokplace enterprise
Oh so it's hard to get it dang
geet that set it up etc
set up google identity
uh
reconfigure entire google cloud account
to a billing account instead of express moed
Google Cloud is required?
more bullsht with support to allow u to buy a enterprise plus license
on top of the google workspace business, and google workplace enterprise plus you already bought of course
to be eligible to buy this

then
once u whine to support enough and they let u buy the enterprise license etc
have to configre it and
skip the part where u spend 20 hours debugging what obscure permission or api u forgot to enable
since ill tell u
That's alot of intermediate steps 😭
u have to ask support to give you preview permissions for the new agent stuff
but to open a non billing related tickett
youu have to buy a tech support subscription
LMAO


@magic stag just ping me when the multi agent made the ai instructions, I am dying to see the results 😭🙏
tournaments ongoing


if this works well
lol
im gonna be so happy
Woah that's something sick
I am curious on lmaeren nano banana pro have google search
yo chilllll 😭 🙏 ⁉️
some inconsistencies but you could just buff em out
"massive freaky tongue face watching you from the top as an equirectangular image with a perfect seem for 360 viewing as skybox"
💀

uhhh
eggmongus plantus
I am curious on lmaeren nano banana pro have google search(Answer me )
Hi, there are api for lmarena?
@gaunt spade
💀


You can also nudge the aspect ratio it will generate by saying
--ar 9:16
completely manufactured a fake game with what like 5 prompts
1K cuz 2K and 4K cost alot
let me test
nope
this would've been a lot worse quality if it was needed
nerfed
i saw this before are you suppose to be a bot?
The new CAPTCHA sucks. I keep getting error messages even when I click the correct images. Please fix this as soon as possible
No I am not a bot
I only want to get a project or job
why they are the same then?
https://discordapp.com/channels/1340554757349179412/1340554757827461211/1441446494417850461
is that from API or gemini web & app
From lmarena
i don't see a ton?
even then that looks to be a really complex diagram
nm for this scale is impossible, wrapted is misspelled and linical
broooooooooo wtf why this new nano banana is soo good, imagine in 5 month 😭 🙏
when very basic simple edits for this model at least
suddenly become extremely hard
and you have to prompt it over and over
ykwim.,,.,
Well I didnt know it could make simple spelling mistakes like that
oh no it still can it's done that before
not a big deal if you're not specific about the text
Can't make diagrams for sure
the trees are in exatacly in same place of original game 🥶
i've had generations with zero text errors
noob
its much better for me now
it verifies me insta
You can tag the Moderator team whenever you see this type of messages @ Moderator. Thank you in advance.
yeah i was about to ping Gato-13 from your team, but i didnt wanted to blame him because i thought he just copy paste.
No, I’m getting errors even when I select the correct images
No problem.
I had to log out of my account because nothing was working anymore
whats the issue?
@magic stag is it done 😭 ?
me too..
At least works to stop bots 🔥
the heck how long does it take?
i didnt realize
wtf
it didnt just make random sht up. i recognize these from when i was researching this myself (
). its literally using techniques from research papres and stuff
what the fk. LOL.
AND I REMEMBER READING ABOUT META REASONING BEING THE EBST TOO
LMAO
i have no idea
last one was
40-44 mins
can you send that pdf or smt i wanna see it
so its just merging all info on the internet?
no its literally making instructions based off of research papers showing performance improvements etc with certain techniques
and evaluating them against eachother
Basically claude research but more expensive version of it?
not even close
this is just the sources for 1 out of the 100 (
) techniques
dang thats sick
So it does 100 different researches and than ranks them with elo rating?
What affects the elo rating?
basically i told i wanted system instructions for AI to uhh
hold on
ok i dont wanna hit back and lose it potentially
/ image generator
not hitting back button yet
this is the best rated?
Wait???
Did the a.i. created this???
but i told it i wanted ai instructions to improve reasoning, autonomic/agentic capabilities, whatever
there no image genration, its at https://lmarena.ai/?chat-modality=image
It's got graphs and stuff 😮
i thought it was gonna make up a ton of stuff and randomly test them, but its literally using techniques based off of research papers proven to work
no
thats one of the sources
that it based making the prompt off of
that was a paper demonstrating the effectiveness of the prompts architecture
alright this is worth every god damn p enny and hour of my life i wasted on google cloud
umm guys what is this model????
as an ex robot programmer, i assume programming robots
idk where i got it, it just spaned there
yeah ive seen it, just ignore it if you dont program robots lol
i mean its not out yet so you can use it using that link i sent
i see it fine, its at the bottom under all
oh wait your right i was looking at Featured tabs not All
@stuck barn Please have a look at https://discord.com/channels/1340554757349179412/1397655624103493813 for a complete step-by-step guide on how to generate videos using the bot.
Nano Banana Pro is finally here — and it’s the biggest upgrade Higgsfield has ever launched. Powered by Google’s new Gemini 3 Pro Image engine, this model is built for control, clarity, and structure… and yes, it now generates in true 4K.
In this video, I test Nano Banana Pro live on Higgsfield.ai and break down everything that makes it...
lol the machine doesnt forget its past
yeah seems like it 😭
i remember during biden you would ask for an image of the president and it would always be trump, i wonder how long trump is going to remain the default
ping mods
the video description doesnt mention "free" so its fine
"unlimited" with a price, bro thats every service what is it even advertising
logic tbh
i mean like, sure some are like 100 a day for 10 dollars, but you could just pay a billion dollars and get billions of generations.. if you have infinite money you could get infinite generations
ill just wait for a cosmic ray to hit my bank account
ok in hindsight some people actually might want this
trump said everyone will get 2,000$
yeah that will happen (sarcasm)
also my australian ahh isnt getting anything, just screwed on trade
dang
Critically acclaimed pop superstar Caroline Polachek debuts her latest single, “Dang,” performed live on The Late Show. To listen to the song on your favorite streaming service, visit: https://ffm.to/dang.
#Colbert #Music #CarolinePolachek #Dang
Subscribe To "The Late Show" Channel: http://bit.ly/ColbertYouTube
Watch full episodes of "The...
bro i love that music video
whats a good system prompt for a mars colony simulation? lol
good
Hey everyone
Excited to be here. I’m a full-stack & blockchain engineer working at the intersection of AI and Web3 — agent pipelines, trust layers, on-chain execution systems, AI wallets, and autonomous smart contract interactions.
If anyone is exploring AI agents, compute markets, verifiable inference, or on-chain automation, I’d love to exchange ideas. It’s one of my favorite areas to build in right now.
Google reCAPTCHA doesn't work. It keeps asking for verification, and still all you get is "Something went wrong with this response, please try again." Lmarena is completely unusable. Why do you ruin the site more and more every Friday?
refresh, usually fixes it
Helo
interesting
so the new gemini is bad at roleplaying it roleplays for you too much
I only run one queue at a time now and it pops up every time with the endless loop. Just well done LM.
what model you using after verfication? cuz cloudflare is having trouble with their server so verfication have issues
YES. i put in system prompt dont talk for me or do actions for me AND IT STILL DOES. grok is better at not doing that, but even with premium i run into the limit in like 10 minutes
try it out, but it wont last long without super grok
LMArena dosn't have super grok i think
Grok For RP? Wild in language but plain in style, not immersive enough.
So then what is the best?
How to a get a new phone number for gmail acc so i want to create gmail.. , i want that.. ill do anything in exchange..
I've been looking for the best one for all my RP's
super grok is just more access to grok on the grok/x site- its the same grok as on lmarena
Some has a favor on Claude. I personally stick with 4o and 4.1
in battle mode it's a disaster, in direct mode I tried seedream, but anyway if I have to spend time resolving reCAPTCHA even if I solve them correctly I will never see the result, cloudflare worked google NO
for rp the grok is going very good, but has the sonnet 4.5 too
gemini 3.0 and gpt 5.1 more like a tool
sonnet i run into the thing were if you use it too much you hit its limits
then you wait for like an hour
just altenate opus 4.1 and sonnet 4.5 lol same thing
neet
hmmm, the sonnet still the better, right?
that has to be google server probleme so give it about 10-20 minute it will be fixed
just now when I did come back to work
Nah, there is a feed in feedback forum discussing it. Since the change launched, it keeps happening. Every click in battle model triggers reCAPTCHA, and the verification is glitched. Takes over 10-20 rounds until hit time limit, which causes output denial.
oh so google updated captcha
how it looks like?
urg gemini still broken after a day, keeps repeating stuff
whats urg?
idk i havent seen it
like 'arg' just an expression
does u know any service offering a working temp numbers for gmail?, i have a days trying..
unfortunately mate, google now requires your phone scan a QR code to proceed to phone verification, which means those online sms reception service won’t work here. There is only tiny little chance that some ip/area does not require phone verification to register.
i have bypassed that..
now it only asks for phone number..
u dont have many phone
can you not pay like a dollar for a sim?
i mean i bypassed that, like it sometimes shows it to me, and sometimes not.. but now its not..
No new gemini models for nov
If that be true, then there are tons of paid sms reception service that only costs 1 dollar or less.
they are done for nov
brah.. i do have a real numbers and a real gmails.. but i do want to stay anonym..
From who?
from every as i can..
Impossible
even from myself if i can..
you talking here disproves that
I see you
i said "AS i CAN"
yeah? , there is a service offering more less then that!, its in cen.ts and accepts crypt0..
but i dont have crypt0..
extra extra..
Depending on where you live, see if you can purchase sim anonymously(I don’t know, it’s not allowed in where I live). I guess if paying online sms service, you’ll be worrying about card information leak. Borrowing friends or family member also dangerous in your case. So you really don’t have many options my friend.
i guess maybe i would get by that on radar.. same as if i searched on google "LINUX"
Yes I'm sending your information to Google now
why is it illegal tho
You're too paranoid bruh
Linux is open source and free to use
i mean its the same as money but digitally
and most platforms requires kyc
so why is it illegal
didnt u hear about the snowden story?
i dont understand
Let’s be fair, you don’t commit anything close to Snowden…
he said if u did search on google somethings such like "LINUX", they will put a tag on top of ur head.. so basically u are danger to themm..
I need more bitcoins 🤤
thats why i love an0nymity..
bruh, why?
I've searched worse thing than Linux
Ok it’s coming officially?
if discord now tried to see my i.p , they will see tor..
Probably stupid countries with their stupid laws and stupid policies
It’s OpenAI’s decision, claimed passing it giving users more “freedom” over adult content
It comes from the 4o/5 censorship and router issue
Ohh! i forget to publish somethig...
What features are for user over 18+???
Yes i did it you paranoid people
I did gave openai my id and i don't gaf
They give me free stuff i give them my soul
LLM's aren't cheap
I'd give them my kidney if they asked
adult mode coming soon
they laying the ground work now for december
here its in thereeee.. #november-contest
Why not just use a vpn
NSFW Models were 👀
I don't think it's only my country...
I think this will be asked of everyone soon
Oh oof, I stopped using openai website once duckai launched
not yet, in december
Elon>Sam🤡
but I'm liking the grok 4.1 fast 🔥 the grok need to be fast and small, don't 6 ti parametrs
caution: said by Sam🐍
what did u send?
fake?
ur digital version will be made in 2040
Oooh nooo
but without ur conciouness
Consciousness or permission
Mines already made
u already gave away ur permission
so u 25 now?
Yeah
False sense of privacy is more dangerous than acknowledging it doesnt exist without drastic measures
