general

[whole sundial]

pretty sure openai models limit song lyrics output to 10 words max, but i have gpt image 1 images of it generating more lyrics than that, but that policy may not even been in its system prompt because openai didn't expect people to make song lyrics with their ai image generator

[sterile tartan]

True

[whole sundial]

and lack of reasoning too

but once again, gpt image 1 can easily make nearly perfect reproductions of album covers, especially the mini variant

[knotty fable]

There are a number of AI powered webpages for lyrics, I had a look at one point and tested to see if those figured out my stuf.
Done the same with my music, to see, and a few have provided creative ideas.

[sterile tartan]

How mini is better?

[whole sundial]

i have no clue, but it's better for some reason

[sterile tartan]

Interesting noted

Maybe i should start selling those

Album Covers

💀

[queen veldt]

Whoops

It CAN sing

Nevermind

[whole sundial]

nano banana 2 can do it too, but you have to prompt it the right way, its accuracy is between gpt image 1 and gpt image 1 mini

[sterile tartan]

Noted noted

How you have so much knowledge

💀

[whole sundial]

i've spent a lot of time with these models, i have over 1,500 nano banana pro images in my collection

and over 4,000 gpt image 1 images

[sterile tartan]

💀

Do you work for them or something

Props to you fam

[whole sundial]

no, i just have a lot of higgsfield credits

[sterile tartan]

How

💀

[whole sundial]

twitter

[queen veldt]

It's a secret

[sterile tartan]

Spill it

[queen veldt]

There's this twitter account that gives a

[sterile tartan]

Code?

Redeem code?

[whole sundial]

let's just say i've reposted at least 100 of their tweets for credits, they even pay me now to tweet about their product, which i use to buy their sub so i can use the credits

[keen beacon]

Its scheming right now

Movement ai

[sterile tartan]

Daym they pay you too?

[keen beacon]

lol

[sterile tartan]

Is it good?

[whole sundial]

got $102 last week and i should be getting $250 soon

[sterile tartan]

U have alot of followers?

[whole sundial]

yeah the site is nice, i like using it for image gen

[sterile tartan]

Ikr

[whole sundial]

i have 4 followers lol

[sterile tartan]

Bruh

[whole sundial]

but they still pay me

[sterile tartan]

They are still paying?

Alright let me join in

[knotty fable]

There's a lot of incentive program s like that, SeaArt got one also.

[sterile tartan]

💀

[keen beacon]

This is s scheming

[queen veldt]

Wait how are they paying you again

For sending tweets about them?

[whole sundial]

if you repost and reply to their tweets enough, you may end up on their list

[queen veldt]

$100 a week is a LOT

[keen beacon]

Ya

It is

[sterile tartan]

Alright that's very helpful data

[queen veldt]

$400 a month is crazy

[sterile tartan]

Time to make my art collection

Then i can easily get these creative programs

[keen beacon]

😂

[whole sundial]

tbh i just started a week or 2 ago, they gave me 2 other promotions when they first put me on the list, could've gotten $150 minimum if i did it

[knotty fable]

Here's a recent ranking from SeaArt.

[sterile tartan]

Hold up im crashing in

[keen beacon]

The only problem is when you get sponsored

[queen veldt]

But is it hard to get "on the list"

[keen beacon]

You’re tainted

[queen veldt]

How big is the list

[knotty fable]

3300 dolares for the top guy.

[keen beacon]

It’s hard to be objective

[sterile tartan]

Is ok i don't want poverty

[whole sundial]

it took me like a month or so interacting with them before they added me

[sterile tartan]

💀

[keen beacon]

You don’t have to be poverty

I’m just saying that it’s part of the territory

[whole sundial]

i should get paid for their most recent promotion in a day or so

[keen beacon]

Nothing wrong with

[sterile tartan]

I know is just business

[sterile tartan]

Man you are really motivating me now

[queen veldt]

Are they paying you with monoply money or?

[keen beacon]

Awesome to hear

[whole sundial]

actual money

[sterile tartan]

Is real cash

[keen beacon]

Good for you Kiri

[sterile tartan]

$$$

[keen beacon]

Motivational

💯

[sterile tartan]

I have started my journey of being Ai Artist Anyways

[whole sundial]

i was able to put it into paypal and spend it, i may become an influencer, who knows

[keen beacon]

Thats wats up

lol

[sterile tartan]

Make good images

[keen beacon]

Don’t get bought by the industry

[whole sundial]

yeah some of my nano banana pro images actually look pretty nice

[sterile tartan]

Show us

[keen beacon]

I wish I could share mine 😭

[sterile tartan]

Share it

💀

[keen beacon]

They would kick me out lol

[keen beacon]

Not from the program, but from using their apps 🤣

[sterile tartan]

Daym

💀

[keen beacon]

Nice colors

[sterile tartan]

This is flux

That one is seedream

I told it to make it anime style

[keen beacon]

Wait

Top is seed?

[sterile tartan]

Yes

[keen beacon]

U can tell

[whole sundial]

some examples (both 4k)

[keen beacon]

I don’t get messaging in that right image

[sterile tartan]

Really?

[keen beacon]

Yeah, of course

[sterile tartan]

Reproduction

[keen beacon]

Is it?

[sterile tartan]

Seedream really has that asian asthetic

[keen beacon]

Nice

[sterile tartan]

That's what i think so

[keen beacon]

Watercolor?

[keen beacon]

Great character placement

[whole sundial]

the model's idea of studio ghibli (which it doesn't really look like, it's just a generic anime theme from the model)

[keen beacon]

No, I’ve Ghibli to. It’s hard to replicate.

[whole sundial]

google models are weird about reproducing certain things

[keen beacon]

Give me some interpretation, but they never do

Ya

A lot of models are

[sterile tartan]

They can't afford to get sued

[keen beacon]

Actually

Wait

[sterile tartan]

Censored models will always have such constraints

[whole sundial]

but openai can! look at their 1t+ investments! their image models can straight up reproduce copyrighted album covers, lyrics, and even some single covers with a simple prompt in api

[sterile tartan]

💀

Are they not getting sued?

[whole sundial]

no, because nobody knows about this

[sterile tartan]

💀

[whole sundial]

nobody's tried it, so nobody knows of it

[sterile tartan]

Daym

[whole sundial]

of course nano banana pro can reproduce album covers too, you just have to be a bit clever

[sterile tartan]

🤫

[sterile tartan]

Like?

Kinda like jailbreaking i guess

[whole sundial]

i figured out that if i ask it about a song that is on the album, it will sometimes generate that album's cover in the image

[sterile tartan]

Daym

💀

Do u think i could become a Ai Artist?

🤔

[queen veldt]

Yes vro

[whole sundial]

anyone can, you just have to be good at prompting, that's all that it is

[sterile tartan]

Gotcha

Ty again guys

[whole sundial]

try it multiple times, choose the best result

[sterile tartan]

Now i know who to target and how

[sterile tartan]

I don't have problem for image generation since i have yupp

And also LMarena

I can use any model of my choice

[sterile tartan]

This is unique

[keen beacon]

Ugh

[sterile tartan]

It has that odd psychological feeling

[knotty fable]

Weird.

[keen beacon]

The other version are better lol

It is werid

[sterile tartan]

Yeah

Why are you being weird

[keen beacon]

But there’s technicality to that makes it interesting

[sterile tartan]

Just kidding

[keen beacon]

Not necessarily the image itself

[sterile tartan]

Then?

[sterile tartan]

Daym

[keen beacon]

But really what I’m into now is multi frame image consistency

[whole sundial]

looks like a real product, right?

[keen beacon]

I think this is harder

[upper locust]

👋

[keen beacon]

But what’s cool is

[whole sundial]

maybe i should start convincing people this is real, i'll just finetune kimi k2 thinking off of claude and call it this, put it on cerebras too

[keen beacon]

It would be totally normal

[shell crypt]

👋

[keen beacon]

This is a very common tactic

[keen beacon]

You just need an infrastructure

[whole sundial]

yeah just host the site on cloudflare and host the model and do the training on cerebras

[keen beacon]

Dude

I’m about to show u something

That will make you feel stupid

[light meadow]

#1397655624103493813

[keen beacon]

https://youtu.be/SfitdmwsDTk?si=fjbJ-olGlBjsrOUW

This chick …..

Charges one dollar a minute to talk to her AI

She’s a millionaire

All she did was clone her voice make am avatar set up a chatbot to respond like would

[keen beacon]

100% legal

[torn mantle]

wait what

[keen beacon]

lol

[torn mantle]

is this real

[keen beacon]

Yes

[torn mantle]

after all of the embargo?

oh wow

[keen beacon]

That’s why they call the theater of politics

But there’s actually a paper here

[whole sundial]

i don't think xi would approve chinese companies to buy the chips though, keep in mind china really wants to scrap nvidia chips so this may not do much

[keen beacon]

No one man rules alone

Stay in power you have to maintain power

I’m sure there’s national interest for China to get on board

[torn mantle]

Shipments could begin as early as the next few weeks, once the Commerce Department completes its review and issues export licenses to NVIDIA. Trump stated that the process is "finalizing details," and NVIDIA has already expressed readiness to proceed. This follows weeks of internal deliberations and a recent meeting between Trump and NVIDIA CEO Jensen Huang.

interesting

[keen beacon]

Oh look, I’ll show you the check this out

Or actually, you could look it up yourself

So the ballroom that they’re building, guess who donated to that quite a large amount

[torn mantle]

This comes amid China's controls on rare earth minerals critical for U.S. tech production. Trump noted a "positive" response from President Xi Jinping, suggesting it's part of reciprocal negotiations.

AHA

[keen beacon]

https://qz.com/trump-ballroom-nvidia-meta-apple-microsoft-amazon

[torn mantle]

more like loosing

lol

it backfires on him

[keen beacon]

Doggy

Check this out

The video that got me banned from AI discord

[torn mantle]

its still solid

ngl

[keen beacon]

https://youtu.be/Na2AJ-W7g54?si=dMRRLY9t-NK_6g5Z

[torn mantle]

dont forget they will optimize the hell out of that chip

[keen beacon]

This was four months ago just forward to the end

Trump thought he owns all of AI. He didn’t even know who he was.

They don’t have to let them buy

They have company set up in Singapore

[whole sundial]

yeah, many major chinese ai companies do a lot of hosting/inference in singapore

[keen beacon]

Yeah, because Singapore is one of the best countries to do business with

[torn mantle]

h20 vs h200

still a good deal

[keen beacon]

H20 for sure is back doored or used for political leverage

[whole sundial]

because the fancy gpus they need can't be brought to china, so they all live in singapore

[keen beacon]

Yup

And also, Singapore has laws protecting businesses and established business global business reputation

[whole sundial]

i have an extension that will show me the location of sites as long as it isn't on a cdn, z.ai says it is from singapore

[keen beacon]

Have you heard about the Singapore business Federation?

[whole sundial]

no

[keen beacon]

Really and I’m surprised to learn this too

I’m not sure if you remember

[sterile tartan]

Guys do you know best image upscaler?

[queen veldt]

Nb pro

[whole sundial]

yeah nb pro is great with details

[sterile tartan]

The synthID

💀

[queen veldt]

Just flush that through other ai model

[queen veldt]

Say "Just give me the original image, don't change anything"

[whole sundial]

I was using topaz, but I tried it when it was first leaked on media.io and it was great

[sterile tartan]

It will lose quality if i do that

Unless

I use another 2k-4k model

[queen veldt]

Seedream

[sterile tartan]

💀

[sterile tartan]

As i thought

So there won't be any point of dedicated image upscalers now?

💀

[queen veldt]

Frfr

It's not even much more expensive tbg

[whole sundial]

not really, the best image upscalers are just diffusion models anyways

[sterile tartan]

Oh

[keen beacon]

And expensive

Topaz

Is good

[keen beacon]

So so is the crystal scaler?

[queen veldt]

0.08 per image upscale

[sterile tartan]

Alright so i can use any ai model with 2-4k quality

[keen beacon]

Yeah

Fal is good

.16 cents

I think for the crystals scaler

[sterile tartan]

I will probably buy paid tools after earnings

Weavy, OpenArt of Higgsfield

[keen beacon]

If u do anime stick with seedream

[sterile tartan]

Yeah

[keen beacon]

Because the the rich color

[sterile tartan]

I have kinda figured out strengths of each model

[keen beacon]

Yeah good for u

[sterile tartan]

Seedream basically has asian Asthetics

[queen veldt]

Hoold uppp

[keen beacon]

If you ever wanted to convert it to videos if you want as much color as you can

[sterile tartan]

Flux for realism

[queen veldt]

This one is insane upscale

[keen beacon]

Which one is that?

Ya yeah that looks good

[queen veldt]

And it's $0.001 per megapixel

Seedvr2

[keen beacon]

Ugh

[sterile tartan]

💀

[keen beacon]

Have u tried crystal upscale

[queen veldt]

Meaning 4k (4mp) image is 0.004

[keen beacon]

https://fal.ai/models/clarityai/crystal-upscaler

[queen veldt]

No i haven't

I don't have any image in urgent need of upscale

[whole sundial]

yeah these bytedance models are crazy, seedream, seedance, seedvr, they also have llms just called seed but only one is open weights (seed-oss-36b)

[queen veldt]

Heard they were making 3d generator from image too

[sterile tartan]

Bytedance has Duobao for Text

I use their chinese and english app

[queen veldt]

We already have some image to 3d but they're bad

[keen beacon]

Seed dream is going to come up

[whole sundial]

seed is the international name for doubao llms

[keen beacon]

Yeah very good stuff

[sterile tartan]

Ohhh

[keen beacon]

They still have a little bit to go, but I think they’re really awesome

I mean to reach the kind of colors. They are reaches incredible

[whole sundial]

seed-oss is not bad, nous used it to make hermes 4.3

[sterile tartan]

Chinese are really something else

[sterile tartan]

Is sad they don't have discord

💀

[sterile tartan]

Is hermes good?

[whole sundial]

I don't really use those models, sadly so idk how good it is

[keen beacon]

It really sucks though

If you wanna make high-quality videos

[sterile tartan]

Right let's just stick to SOTA

[keen beacon]

You have to use upscale images and then highest resolution videos, and then upscale the videos

[sterile tartan]

💀

[keen beacon]

And it gets spendy

[sterile tartan]

I can use wan 2.5 for video

[whole sundial]

tencent should've at least made one for hunyuan, after all they own part of discord

[sterile tartan]

True

Didn't knew that

Good intel

I have heard new qwen edit is coming soon

[keen beacon]

Qwen is good

[whole sundial]

yeah they need to make a new qwen image edit

maybe z image edit is coming soon as well

[sterile tartan]

I have a stack of around 30 image models for use

[keen beacon]

They’re not restrictive as American one for some reason

[sterile tartan]

More likely

[keen beacon]

Open ai img gen is gunna be good too

[whole sundial]

also by alibaba and tongyi lab, just a different part of it than qwen

[sterile tartan]

Image Generation Models
Nano Banana/Nano Banana Pro/Imagine 4
Seedream 4.5/Seedream 4
Flux 2 Pro/Flex/Dev
Hunyuanimage 3
Wan 2.5 T2I
MAI Image 1/Dalle 3/GPT Image 1
Qwen Image/Qwen Image Edit/Z Image
Ideogram 3/Recraft 3/Playground 3/ Rev 1
Photon/Lucid Origin-realism/HIDream I1 Full
Stable Diffusion 3.5 Series
Minimax Image 01/Bria HD/OmniGen2
Midjourney V7

My models list

[keen beacon]

Probably may for new open ai model

March

[sterile tartan]

How u know

[keen beacon]

Guessing

[whole sundial]

that's a lot of models and you don't even use most of them likely

I stick to nano banana pro and gpt image 1, maybe I'll use a bit of seedream

[sterile tartan]

True some are just there for specific use cases

[keen beacon]

Yeah, I was gonna say I don’t think you could really use a single model. You have to use multiple if you’re trying to make like something really unique.

[sterile tartan]

My top are NBP, Seedream, GPT, Flux and Midjourney

[whole sundial]

mai image 1 sucks though, it has the worst text rendering

[keen beacon]

Wow really

GPT?

[sterile tartan]

Is just there for name sake

[sterile tartan]

GPT Image 1

[keen beacon]

Mini max getting suied

I think they’re waiting for the lawsuit and the release

M2

[whole sundial]

heard about that, they added text filters in prompts

[keen beacon]

Ya

I looked at the court papers

[sterile tartan]

Minimax image is good for game images

[whole sundial]

for example, you can't use "Genie" (as in alladin)

[keen beacon]

Ya

I think so too that it’s gonna be really good

[whole sundial]

I hope so, gpt image 1 had good world knowledge until Google beat them, hopefully openai will fight back with a better model

[keen beacon]

Open AI model for some reason in my opinion are artistic

[sterile tartan]

Guys where's GPT 5.2

💀

[keen beacon]

Ngl veo sucks

[whole sundial]

yeah sora 2 knows a lot of things

[fading summit]

Hi) Is it possible to connect my database to lmarena? Like mcp to claude

[keen beacon]

Last night I was using VO3

It’s not really that good

[whole sundial]

lmarena doesn't have mcp integrations, sadly

[fading summit]

In a dev mode mb...

[keen beacon]

Sora is really strong model

Animation so much smoother

[sterile tartan]

Is imagine 4 good?

I know is old

[fading summit]

Ok, but mb some kind of different database?

[whole sundial]

no database connection either

[whole sundial]

it was good for its time, especially with anime

[sterile tartan]

What about now?

Will imagine 5 come

[whole sundial]

just use nano banana pro

[sterile tartan]

Gotcha

[whole sundial]

probably not, they are focusing much more on native multimodal models

[fading summit]

I made a bot in telegram, connected to claude deskop with mcp on my laptop, but the limits even with pro are too small, i have no other idea...

[keen beacon]

Move made so much clips from a movie

I never thought that would be possible

[sterile tartan]

But i guess nano banana is the successor anyways

So there might won't be anymore imagine

[whole sundial]

yeah, nano banana pro is much better in every way

[keen beacon]

Nano is cool because it was the first model that let you have that kind of consistency

Fr

[sterile tartan]

Ikr

[whole sundial]

I'll be surprised if imagen 5 comes

[sterile tartan]

Both bananas are all over internet

[keen beacon]

I heard they’re changing the name

[sterile tartan]

Does nano have Asthetics like imagine?

[keen beacon]

I don’t think so my observation at least it looks like it just generic standard

[whole sundial]

idk, you'll have to try it

[keen beacon]

Really doesn’t seem to have like an art style and the one that does have is very plain and boring

[sterile tartan]

But again for Asthetics we have Seedream

[keen beacon]

Yes

I guess there’s also wisk

[sterile tartan]

Whisk yeah for designing

[keen beacon]

That’s not really that good

[sterile tartan]

I don't use it

💀

[keen beacon]

Same

[whole sundial]

yeah seedream is just an image gen attached to a seed/doubao vlm

[keen beacon]

I tried it and it just is not really my thing

They already got preset or something. I don’t know they don’t blend right maybe I’m not using it right

I’m forgetting one more model that’s really good

Just doesn’t come to mind

[sterile tartan]

Any words?

Is it chinese?

[keen beacon]

No I want to say

Or whatever it’s called

[sterile tartan]

First time heard of it

Or u mean krea

[keen beacon]

Ya and I think that’s more like a all-in-one kind of thing

[sterile tartan]

Ahhh that platform

Is like openart

Daym

[keen beacon]

Isn’t good

[sterile tartan]

True

Just fine

Do u know why it couldn't make a good one

[keen beacon]

Cause I don’t know what the real one looks like

I have nothing to compare it to

[sterile tartan]

Exactly

There is no such real passport of rome

There is passport of italy

[keen beacon]

Well, obviously

🤣

Does that kind of look real?

Let me check which model did that

What’s a ghost pepper…?

[sterile tartan]

Yeah it did look good

Ghost Pepper?

[whole sundial]

<@&1349916362595635286>

[keen beacon]

The model’s name is ghost pepper

[sterile tartan]

Probably a new model

It could be Nano Banana 3 Flash

[keen beacon]

It would be really nice to have a model like it have consistent background for videos

And composition

[sterile tartan]

Yeah we know soon which model it's

[keen beacon]

But honestly, I make a bunch of videos. I do not think at the bottom of my heart. I don’t think they wanna design it to make long-term videos, bro.

Not long-term I mean long form

[sterile tartan]

Long form videos are hard for now

[keen beacon]

Replicate from movies because currently Sora is able too

[sterile tartan]

Short form is maximum capacity for these models for now

[keen beacon]

Ya but Sora showed us u can use it as a full movie editor

[sterile tartan]

The compute also will be expensive

[keen beacon]

I already got one movie down where I could replicate it 100%

From Sora

I’m working on the Sopranos see if I can get that

[sterile tartan]

But i assume in future we will have long form as well

[keen beacon]

Ya

Hopefully

[sterile tartan]

Good luck

[keen beacon]

Haha I’ll need it

[sterile tartan]

More Likely Dw

🤞🏻

[keen beacon]

For some reason

The videos come out like this all stretched

I never seen this before and it does it constantly when I try

[sterile tartan]

Looks like a still image

Seems something associated with aspect ratios

[keen beacon]

Werid

[sterile tartan]

Well i be going to sleep now bro take care Good Night and Good Luck

Is 7 AM

[keen beacon]

Until g

Night I I saw your request

I’ll get to it

[sterile tartan]

Is probably associated with original source aspect ratio

[keen beacon]

You might be right

[sterile tartan]

💀

[keen beacon]

It could be a protective layer

Thank you

[sterile tartan]

Your Welcome

🫡

😴

[keen beacon]

Yeah, I gotta go to bed. I’m not sure when I’ll be back. I’ll be back maybe may b a while so we’ll see everybody. Good luck 2 all

[jade egret]

is this allowed

[whole sundial]

<@&1349916362595635286>

[obtuse smelt]

wow

is seem error ?

[jovial solar]

Hello

[severe canopy]

Hi, how to know which's AI VIDEO model that LMArena use in generating ?

[fringe carbon]

do we think that 5.2 will beat out gemini 3.0?

[simple heart]

[cloud zinc]

no

[nimble lava]

Hello

When i write slash image to video nothing happens

Is it because i’m on the phone ?

[jovial solar]

#video-arena-1

[thorn spear]

Is there a limitation for video generations?

Like 5 videos per day

veo 3.1 and kling

[keen beacon]

in future,is possible Lmarena added sora and veo in website?

If that ever happens, it will be the most epic thing.

[stoic solar]

Heyy I m making website for ai ad but the output is bad anyone can say how do I make it good? I need something like user inputs a product image then it generated production level ad.

[thorn spear]

Is there is a quota for veo 3.1 and sora 2 pro?

[jovial solar]

I think so

👀

[delicate frost]

will deepseek special ever return?

[plucky sparrow]

random. that's the point, it's supposed to have you rank anonymously which is better

[atomic lagoon]

Maybe

[fringe carbon]

what probability would you give it?

[atomic lagoon]

Are we talking codex 5.2?

[fringe carbon]

no

[atomic lagoon]

or GPT 5.2 or something?

[fringe carbon]

text leaderboard

[atomic lagoon]

Then which one

which model is 5.2

[fringe carbon]

is it not in lmarena yet?

it releases this week

so i assumed it was

it comes out thursday

[atomic lagoon]

Dude what

Which model is it and by which company

[whole sundial]

i know openai has been testing a few image gen models, possibly the successors to gpt image 1 and gpt image 1 mini?

if that's true, then whoever said openai would release a new image gen model soon isn't wrong

[atomic lagoon]

Their next model should be good seeing as how good sora 2 was

Or is

[whole sundial]

i think this is from the larger model of the two

looks more like flux.1 kontext dev or qwen image edit to me, but i guess studio ghibli got angry at them so now they have to change up the theme? better use gpt image 1 before it goes away, but knowing openai that won't happen for at least another year or so

i know studio ghibli got angry at them when sora 2 came out, maybe this is the response on the image gen side so they'll be "compliant"?

still 3:2 though

i should try with multiple subjects on both

also i haven't gotten ghost pepper yet

that's from google maybe?

[fringe carbon]

gpt 5.2

by openai

[atomic lagoon]

Oh okay thanks thats what I was asking

Uhhh

Honestly it could but I wouldnt count on it

They went AGI last model

This they might have all around

[fringe carbon]

my prior is that it won’t surpass google

[atomic lagoon]

I'd say the chances of it being better than gemini 3 is 40%

[fringe carbon]

that’s pretty high

[atomic lagoon]

It is but you gotta think about how they've been sitting on it for a bit

And if robin really is the GPT 5.2 then it will be better

[fringe carbon]

they sat on 5.0 forever

[fringe carbon]

also 5.1 just came out

[atomic lagoon]

5 was also a big upgrade at the time from where they were

[atomic lagoon]

Yes, 5.1 was an AGI update and the only thing it has that 5 doesnt have is the talkative side

[cloud zinc]

10 percent

[atomic lagoon]

5.2 will be overall

Everything

[fringe carbon]

have u seen it in lmarena yet?

[atomic lagoon]

Dont listen to socky, he has probably top 10 worst takes of all time

Genuinely he never has a good take on things

I feel like hes trolling 90% of the time

[cloud zinc]

gpt 5.2 is a rubberband before christmas break, they main gemini 3 killer will come in january

[whole sundial]

yeah this is probably a new image gen model to combat nano banana
too little, too late imo

they are probably rushing it, will likely come out at the same time as gpt 5.2 so they'll have something until their new next generation model comes out

[fringe carbon]

will they actually be able to kill gemini tho?

[atomic lagoon]

That wont combat nano

[fringe carbon]

ifl google has such a massive leg up on data

[cloud zinc]

no google has something better

[whole sundial]

not with this they won't

[cloud zinc]

like for gpt 5 released in august, they were in no hurry and waited till november to release their model

[atomic lagoon]

Nobody is killing google in any way lmao. Google has probably the most resources out of all companies. Google is basically the internet atp

[whole sundial]

they took the one thing that made people want to use gpt image 1 in the first place and completely ruined it (likely for legal reasons)

was fine before sora 2, but now oai is scared so with their rushed nano banana competitor they just forced the model to forget about the ghibli theme and generate a different anime theme instead

[atomic lagoon]

This isn't openai. Theres just no way lmao. Their current image generation is better than this

[whole sundial]

btw i can confirm this is an openai model, the c2pa data mentions gpt-4o, dall-e, and openai, no way you can forge that plus same res as gpt image 1

and this is the better of the two btw

this is the worse one

[atomic lagoon]

Give me the prompts you used for it

[cloud zinc]

sora is on hold

[whole sundial]

i can't share the image but this is the prompt:

Transform this image into the Ghibli art studio theme but:

1. make the picture landscape resolution
2. turn the background into a nature scene
3. make the girl look cute and happy, but keep her eyes open
4. make it look like she is sitting on the bus
5. make her wear a black t-shirt with nothing on it
6. make her hair fade from bright red to black

[whole sundial]

makes sense, they are doing fine in video gen, they need to catch up in text and images

but honestly, the whole hazel model series they are testing is just a band-aid, it's not going to help the fact that google just released a 4k model with better world knowledge than gpt image 1 or gpt image 1 mini

the real gpt image 1 successor will probably be based on shallowpeat and be an actual nano banana pro competitor

but google will already have nbpro's successor ready by then most likely

here is hazel edit 6 again

hazel edit 2 (the worse one)

honestly i'm surprised openai would even bother testing their image gen models on lmarena considering what lmarena did to gpt-image-1's quality

[keen beacon]

Dalle 4 😍🥰

It would be the perfect img model.

[whole sundial]

yeah they could brand it as that considering the 2 dall-e models are being killed off, they need something to replace it and this is a lesser model, then keep gpt-image-2 for the shallowpeat based model

[keen beacon]

Unify both

[keen beacon]

This is really nice

[keen beacon]

A man is galloping at breakneck speed on his horse and appears as if he is going somewhere important; on a mission of some sort.

Another man standing along the side of the road shouts, ‘Where are you going?’ and the man on the horse yells, ‘I don’t know, ask the horse!’

[distant spoke]

G3 Flash？

[ashen mauve]

ah yes I can read this

[sterile tartan]

Where is GPT 5.2

Is 9th December

[torn mantle]

Yea

[ocean ferry]

they can't just merge it(or plug and play), they need to train another image gen model with GPT-Image-1/2 and DALL-E or both, it is called distillation, if you just copy the weight of DALL-E/GPT-Image to another the model will not work.

[keen beacon]

You are wrong

There is a way

[proud pagoda]

hello here

[keen beacon]

What would your username sound like if it had the letter “C” in front of it?

[compact scarab]

hey guys can someone tell me how to download if possible LMA arena at my phone ? (iOS)

[ocean ferry]

what is the "way" then?

prove me wrong about my statement

[weary galleon]

Which one? Codenamed?

[whole sundial]

all image models that start with hazel are by openai

[sterile tartan]

@whole sundial found new image model

Long cat image

[ocean ferry]

Hey Squid, since you seem so confident about merging these models, quick question to see you do know what you're saying about that "There is a way"

Can you explain what a neural network actually is? Like the basic architecture - input layer, hidden layers, output layer, how they connect?

And while you're at it, what's an activation function and why do we need them?

Just wanna make sure we're on the same page about the fundamentals before we talk about merging two completely different model architectures.

(No Googling btw, if you really understand model merging you should be able to explain the basics)

@keen beacon

[keen beacon]

The earnings

Look em up

An ancient traditional Japanese technique called Gattai

[golden ocean]

squid rn

[keen beacon]

😅

[whole sundial]

i already knew about this https://discord.com/channels/1340554757349179412/1446723851097735239

[ocean ferry]

??

is this a joke?

it clearly has nothing to do with what we're talking about

[golden ocean]

ohnoes looking at lmarena general chat rn

[keen beacon]

https://youtu.be/z5yIapnxHr8?si=Q5rQNI5bojdiFA4k

[ocean ferry]

it has nothing to do with what we're talking about.

[ocean ferry]

if you think we're living in that dragonball magic anime thing, no. it's the reality that we cannot fuse models, it's an architectural mismatch.

unless you do what i am talking about with distillation

@keen beacon so please hear me out, if you do this is as a joke or you do believing that anime magic, please do not confidently say anything without knowing the actual facts, it's not that i can't joke around, but it will spread as misinformation if no one is correcting it.

[sterile tartan]

So is it good?

[whole sundial]

not really compared to nano banana unless you want an image editing model that can run locally with a 12gb+ gpu

but it is a good local competitor to nano banana, i'll give it credit for that

[obtuse smelt]

hmm is nano banana error ?

well is nano banana is look not respond

[keen beacon]

It’s a joke

[ocean ferry]

read this @keen beacon

[keen beacon]

Ok, fine let me ask you this do you believe in AGI?

[ocean ferry]

No, before they(LLMs) can self optimize/self modify and Continual learning, then no.

[keen beacon]

Do you think it is possible?

[ocean ferry]

Yep, and it's already happening. in Google recently came out paper they built a self modifying model called HOPE, it has lower perplexity and higher accuracy than Transformers and Transformers++, You should check out the paper cuz it's really cool to see a proof of concept, the paper name is "Nested Learning: The Illusion of Deep Learning Architecture" By Ali Behrouz

well sadly this is a really new paper and people doesn't seem to care about it

[distant spoke]

Web page made by holo-scope.

[ocean ferry]

well if some AI enthusiast worked to train a model based on this paper then open sourced it, it would be really cool and unique.

[ocean ferry]

damn i think it's distilled from Gemini 3 Pro😅

[keen beacon]

The paper is philosophical in nature.

[ocean ferry]

the style is very similar to Gemini 3 Pro

[keen beacon]

Agi = A GREAT illusion

[ocean ferry]

lol yea

[polar wharf]

I am a real fox!
Not an imposter!

[wicked sage]

hi guys

[wicked sage]

wow 😲 i cant see a diffirence

[polar wharf]

😺❤️😊

Cute wall drawing (not ai)

To share some real art

[wicked sage]

thats nice

[polar wharf]

That's real streetart, like I love it 😊

Cool that you like it too

[obtuse smelt]

oh nice

[gusty wyvern]

how to fix that?

[obtuse smelt]

reload page maybe

[odd geyser]

Reboot it. It's just that sometimes the page doesn't load like that.

[obtuse smelt]

yeah

i reload how much is not responding

[distant spoke]

G3 Flash

[plucky sparrow]

looks like gpt 5.2 not coming out today

[obtuse smelt]

gpt how i can use chat gpt ?

[golden ocean]

do9nt bully it

itll get insecure

[obtuse smelt]

oh ok

[regal heath]

can i fix it? endless generating

[lyric wigeon]

Hi guys why i always getting "Something went wrong with this response, please try again." when trying to edit image

[obtuse heart]

been like that since yesterday

dont think itll be fixed very soon

[sweet topaz]

👋

[shell crypt]

👋

[obtuse heart]

what in the bots

[upper locust]

👋

[golden ocean]

👋

[pale reef]

👋

[obtuse smelt]

👋

hmm ?

[obtuse smelt]

yeah me too is like yesterday

[lyric wigeon]

Uhmmm

[torn herald]

👋

[sterile tartan]

I see but again Z Image is Better

💀

[crisp wing]

a bit late but thank you so much!

[torn mantle]

https://x.com/Similarweb/status/1998329140672385104

Similarweb (@Similarweb)

Gemini in November 2025:

- 1.351 billion visits
- #26 website in the world (+4 spots from October)

X · Dec 9, 2025, 9:49 AM

😮

[fleet lintel]

December would be even more interesting. Nov doesn't capture the full impact of Gemini 3 and nano banana pro

https://x.com/Similarweb/status/1997963677065195902

Some drop is expected because of Thanksgiving effect

Similarweb (@Similarweb)

Top 10 websites by total visits in November 2025:

→ No change in rankings.
→ All websites experienced a traffic drop compared to October.
→ ChatGPT experienced the biggest drop.

X · Dec 8, 2025, 9:37 AM

But interesting point is :
→ ChatGPT experienced the biggest drop.

[stuck orchid]

gpt 5.2 today?

[low imp]

Yes

[stuck orchid]

😱

And grok 4.2 coming soon

(This month maybe)

It is real benchmarks?

[fleet lintel]

this was proven to be false

[torn mantle]

aha

right

[torn mantle]

no

fake

[fleet lintel]

never believe any such image now in the age of nano banana pro.

[torn mantle]

idk, i just know they denoise pixels lol

[fleet lintel]

it is just too easy to create fake.. previously it would have taken like 30 min but now few seconds 🙁

[stuck orchid]

With Nano Banana Pro?

[torn mantle]

take real image -> make it into noise pixels ; and the model automatically learns patterns to denoise it

idk the technical stuff tbh

[fleet lintel]

this is exciting. Please dont dissappoint me OAI like last few times

[stuck orchid]

If it's something much better than gpt-5.1, it should be called gpt-5.5

[fleet lintel]

have you seen the names of previous gpt models? their naming team(more like a person) produces work of art. It might even be called gpt 1.1 for all they care

[queen veldt]

https://vm.tiktok.com/ZMDJwPxGE/

Bro posted whole movie 💀

[fleet lintel]

https://x.com/Similarweb/status/1997991458087862583

how Claude is down? All API usage?

Similarweb (@Similarweb)

Ranking of leading Gen AI websites in the list of top-visited sites, November 2025:

#5 ChatGPT (no change from October)
#26 Gemini (+4)
#91 DeepSeek (+1)
#128 Grok (+27)
#167 Perplexity (-26)
#176 Claude (-11)
#174 Character AI (+3)
#318 Microsoft Copilot (+7)

X · Dec 8, 2025, 11:27 AM

[torn mantle]

yea

mostly cursor usage

its a coding model

not good for general use

[flint sandal]

nah the nano banana is not diffusion

[fleet lintel]

we dont know that?

[flint sandal]

we know it because its gemini 2.5 flash and pro is gemini 3 pro. Both these are not a diffusion model

[swift oyster]

We are giving away a free PRO membership to a LmArena member who can make the best SVG using Tensor 1.5 winner will be picked here live in 1 hour. share you results here. (MOD's are welcome too)

Let us kick start it off.

[torn mantle]

diffusion architecture is there, its just whats on top of it

[stuck orchid]

Maybe nano-banana-pro-flash coming soon

After gemini-3-flash

[fleet lintel]

https://www.reddit.com/r/singularity/comments/1phhzxc/new_nano_banana_powered_by_gemini_3_flash_is/

[stuck orchid]

My guess was right. I hope Nano Banana Pro Flash will be free on AiStudio, just like the regular Nano Banana

[sterile tartan]

Where the hell is GPT 5.2

[stuck orchid]

TikTok * nano?

Nano Banana Pro Flash?

[zealous sparrow]

Anyone know who the hazel image models are

They are on imagearena battle

[opaque dock]

Hello 👋

[neon idol]

Haii

[neon idol]

1 week or tomorrow

[zealous sparrow]

There is a new "google" model called holo scope

Cant verify or deny it being google rn

[bleak scarab]

hello every one

[daring rock]

@ruby juniper Please head to #1397655624103493813 for a detailed guide on how to use the bot

[sterile tartan]

💀

[fading rover]

"something went wrong" error keeps flashing after few seconds on nano banana pro 2k .. it's been consistent whenever try fresh attempt it just keep throwing this error

[austere sundial]

This site is horrible now it's detecting repeated prompts and it's blocking 😡😡😡

Don't these DM's see this?

This a.i. site is getting worse, a literal s**t.
Soon there will be limits...

[zealous sparrow]

They had a reason for this

People would ignore other benching of the model

And just test one thing with ig

It. Therefore they would cheat benches

Also ratelimits exist for high cost models to keep LmArena free

[austere sundial]

So their reason is daily limit.
Because everyone else had it.
Except this one...

If that's it, this site stops working.

[zealous sparrow]

Their reason is reducing people who test one thing and discredit a model because it cant do it

You cant understand that they want scores to be fair

[austere sundial]

The problem with this is that if you want to generate the same thing, others don't do it and hit limits.
So you will have to wait almost 2 hours to try more often.

[austere sundial]

Fair was
Now is price.
Like other i.as
That's what made LMarena different, now it's just one more.

[zealous sparrow]

I know it sucks but it helps keep LMArena fairer

[zealous sparrow]

AI prices are too up to let you use models in direct very often

They pay for a Lot of Models

Imagearena,codearena,textarena,videoarena

[austere sundial]

I made it
With old LMarena

Now i can't do it enymore

[zealous sparrow]

Video models have changed

It is not their fault

Models degrade and improve.

It is the reality.

[austere sundial]

No
I'm not talking about the videos
I'm talking about the character, there were 10 prompts to leave him like this.

[zealous sparrow]

Videomodels are not in chat form

They make a entirely new one

Unless you mean importing an image to gen a video.

Video models arent perfect.

[austere sundial]

I'm not talking in video mode
I'm talking about the character, it doesn't always come out perfect
LMarena let us try until it made them perfect
To use them in video models.
But now it doesn't work anymore, they block repeated prompts.

[zealous sparrow]

I know you are mad but. If you dont wish to work with a service for free. Use your money and use AI yourself.

[austere sundial]

So instead of trying until you get it right, you try until the system blocks your prompt.
This is sad for an a.i. that was in my concept the best of all.

[austere sundial]

Exactly what I said, now it's just another i.a like the others
Since the others do this, but it is paid and this one is becoming paid too.
After obtaining our data, it is no longer free to use.

[golden ocean]

why not save the image and then create new chat and paste and type new prompt to edit that image

[golden ocean]

lmarena collected data since the beginning of the universe

[zealous sparrow]

Name one thing LMArena has paid

[golden ocean]

he means you pay with ur data

[austere sundial]

I already do this, the thing is that it is not always generated correctly and we need to remove flaws such as piercings out of nowhere or details placed by mistake.

[austere sundial]

Ik that

[zealous sparrow]

This is the models fault and not LMArenas fault

[golden ocean]

they recently added the feature where u only had to upload the image once in a conversation and new messages would remember the conversation/image history

but it was reverted again week ago

[austere sundial]

The site provides the template but the LMarena site block the template.
Because I've done this before and I didn't like it.

I hate "pay to use" really

[zealous sparrow]

LMArena is free?

[steady grove]

is there any promo for free llm ?

[echo aurora]

Hmm what do you mean?

[steady grove]

probably i mean only one, i think, are there any popular trials for claudes 4.5

or account rotation does thing? 😄

[proud bobcat]

New image model

Riverflow V2

[queen veldt]

Yooo

"guys it beaten the 3 pro"

[zenith scarab]

👋

[half pelican]

👋

[queen veldt]

https://x.com/ArtificialAnlys/status/1978891167795417092

Artificial Analysis (@ArtificialAnlys)

Riverflow 1 from Sourceful debuts at #1 in the Artificial Analysis Image Editing Leaderboard for All Listings, using a vision language model trained by Sourceful to allow chain of thought reasoning in combination with a third-party open weights diffusion model

Riverflow 1

X · Oct 16, 2025, 6:29 PM

Funny how this model doesn't exist

It's just a scam

[echo aurora]

👋 hello @trim wave @half pelican @zenith scarab

[ruby hollow]

can someone help me by generating a simple 2 seconds video using veo 3 or any other ai model ? ill give u the starting adn ending frame ( it would mean alot its for my youtube project )

[austere sundial]

It supost to be a joke ?

[austere sundial]

Technically But They use personal data as "payment"

[jade egret]

gpt 5.2 today?

[hollow ivy]

no

[zealous sparrow]

Every company sells your data

It doesn't matter which

[stuck orchid]

@jade egret
Nano Banana?

Pro of Flash?

[torn mantle]

https://x.com/MistralAI/status/1998407335308358028

Mistral AI (@MistralAI)

Our next-generation coding model family Devstral 2 is available in two sizes: Devstral 2 (123B) under a modified MIT license, and Devstral Small (24B) under Apache 2.0.
Both SOTA, open-source, free to use, and available now via our API.

X · Dec 9, 2025, 3:00 PM

[zealous sparrow]

everything is better than grok

[torn mantle]

grok is so bad

[proud bobcat]

Devstral small at 68%?

I smell benchmaxxing

[zealous sparrow]

I feel like GPT 5.1 would have had potential

if not for the HORRIBLE frontend design

[proud bobcat]

I think OpenAI’s new architecture is just cooked

It’s awful

It’s ass

I don’t even know what they did

4o was perfection

All they had to do was improve the 4 architecture

[zealous sparrow]

5.1 has nice backend

but im curious

how the hell did they mess up the frontend

to look the same on every gen

[sterile tartan]

I AnosCranel Present to All of You
The 🌟 GRAND OMNI MASTER SYSTEM PROMPT 🌟 Version: 7.7

This is A Sophisticated Prompt Engineering Orchestration System of Current 8 SOTA Models

Anyone can use it for META Engineering System-Persona and Such Expert Meta Prompt Engineers

[open birch]

nice one

[sterile tartan]

Use it on LMarena or Yupp Ai

Since these platforms have all the needed models in one place

[zealous sparrow]

if you asked an AI for a 3d escape room, would you expect this? https://019b03dc-e8f7-7f50-b4cd-2297dd6c727e.arena.site
[seahawk]

[whole sundial]

<@&1349916362595635286> (crypto scam btw)

[zealous sparrow]

Very useful update to LMArena [codearena]
It shows you how many lines it added or removed

[placid flame]

Thank you!

[zealous sparrow]

Haven't seen lots of mods, are you a new mod?

[sterile tartan]

@whole sundial can u review this please?

[echo aurora]

Nope, Catluv has been with us for a bit now

[austere sundial]

But it's a form of exchange / payment, my man.

[whole sundial]

honestly doesn't look like something for me, looks like a lot of effort for not much benefit tbh

[zealous sparrow]

Ah. I guess im just not on, or they aren't on when i am on. Been loving the skyhawk and seahawk models on codearena. Impressed with the quality if it is really a google flash model.

[sterile tartan]

Ok!

[stuck orchid]

gpt-image-5.1 coming soon?

[zealous sparrow]

Here's a cool windows 10 recreation either skyhawk or seahawk made.
https://019b03e4-025e-79a7-8fcc-777b3eec893f.arena.site
Warning: Shaking screen

[zealous sparrow]

It fetches your date and time which is sick

[whole sundial]

yeah probably

[zealous sparrow]

Which model made this?

[stuck orchid]

gpt-image-1.5?

[whole sundial]

hazel edit 6 (left) and hazel edit 2 (right), both OpenAI models

[zealous sparrow]

How can you know they are OPENAI?

I know hazel has been added to imgarena.

Not sure who made them..

[whole sundial]

looking at c2pa metadata

[zealous sparrow]

Ah..

[sterile tartan]

Probably GPT Image 5.1 and GPT Image 5.1 Mini

[stuck orchid]

Error generating image: Error code: 400 - {'error': {'message': "Invalid value: 'gpt-5'. Supported values are: 'gpt-image-1', 'gpt-image-1-mini', 'gpt-image-1.5', 'gpt-image-1.5-mini', 'hazel-gen-2', 'hazel-gen-3', 'hazel-gen-4', 'hazel-edit-2', 'hazel-edit-4', 'hazel-edit-5', 'hazel-edit-6', 'hazel-small-2', 'dall-e-2', and 'dall-e-3'.", 'type': 'invalid_request_error', 'param': 'model', 'code': 'invalid_value'}}. Retrying...\

[zealous sparrow]

hm so this is a list of OpenAI img models?

[stuck orchid]

https://www.reddit.com/r/OpenAI/comments/1phyva1/are_we_going_to_have_gptimage15/

[neon idol]

Prompt

?

[zealous sparrow]

Ah, alright then

There is 2 textarena models that identity remains unknown

[sterile tartan]

Something with Studio Ghibli Style

[zealous sparrow]

silentnova and holoscope

holo-scope said its google although idk

[whole sundial]

i know that a few of these are on lmarena but i've never seen hazel-edit-4, hazel-edit-5, hazel-small-2, or hazel-gen-3 before, must be more internet checkpoint

[neon idol]

https://tenor.com/lmOyhqgu9dc.gif

[zealous sparrow]

According to a list i have of models, these aren't on LMArena

[whole sundial]

this new model completely changed up the style, might spend a bunch of higgsfield credits now before the new model comes (although I kinda like the new style? must be a copyright thing, ik studio ghibli has gotten a bit angry at openai after sora 2'a launch)

[zealous sparrow]

Unless the list is outdated, because the creator of the bot said his bot bricked due to a new captcha update to LMArena

[zealous sparrow]

Yeah, might just be them complying with 3rd party laws, to not earn themselves a lawsuit especially in a code red.

[whole sundial]

don't worry, it can still reproduce copyrighted album covers though...

[balmy mist]

do we have a new model? been mia for a few weeks?

[zealous sparrow]

They might patch that then..

[zealous sparrow]

Depends on what company

Because we got plenty new

[balmy mist]

whats current SOTA?

[zealous sparrow]

Gemini 3 pro, Opus 4.5,

Both can be listed as SOTA

[balmy mist]

ahh same old never mind lol

i heard we got gemini 3.5 pro

or gemini 3 flash

[zealous sparrow]

It's 2 models on codearena that are prob just flash

[balmy mist]

smh

[sterile tartan]

Yeah u should and true

[zealous sparrow]

People got impressed by the UI capabilities

[balmy mist]

thats what im talkning about

flash is going to be go to model

cant wait until its live

[stuck orchid]

Maybe in March 2026

[whole sundial]

idk i have all these models lol

[zealous sparrow]

It will be the same as gemini 3 pro on AIstudio in terms of prompts for free.

[sterile tartan]

Unlimited?

💀

[whole sundial]

yes, i can use any model that say that an unlimited number of times for a year (as long as i still have the sub)

[sterile tartan]

Daym

[stuck orchid]

Does anyone know of a provider with free access to grok 4.1 via API?

[sterile tartan]

Wait they gave u 1 year subscription?

[stuck orchid]

for free?

[balmy mist]

ahh okay, is flash open router?

[balmy mist]

how?

[stuck orchid]

@echo aurora What were you doing in 2017-2020, when there were no such AI models?

[balmy mist]

so you got opus 4.5 for free unlimited?

[whole sundial]

huh? no i paid for 1 month, i'll still have it when i renew until a year from now

[zealous sparrow]

wdym is flash on openrouter?

[sterile tartan]

As long you renew it i see

[whole sundial]

one of their crazy promotions that they do to get people to sign up, no way they are making money they way they spend it

[zealous sparrow]

@echo aurora is it normal for seahawk or skyhawk to be failing to deploy apps on codearena rn?

[balmy mist]

gemini 3 flash?

also it seems like gpt5.2 coming out this week? openai really panicing lol

[stuck orchid]

According to Elon, grok-5 should also be released by the end of this year

[whole sundial]

probably also the reason why they made gpt image 1.5, they don't like the fact that google has the best image gen

[zealous sparrow]

no its only on LMArena codearena or textarena

[whole sundial]

honestly gpt image 1.5's ghibli outputs look more like nano banana 1 than gpt image 1

[zealous sparrow]

3 flash is really fast tho

400 lines in less than 2 minutes

Actually 1 minute

[stuck orchid]

Is the quality good?

Is it similar to gemini-2.5-pro, or better?

[echo aurora]

I can flag, thanks!

[zealous sparrow]

50/50 i think so.

I'll do one test and show you.

[zealous sparrow]

https://019b03fa-6495-733a-b1dd-f92c8285b55b.arena.site
prompt: macOS recreation, with everything functional

not everything was functional but ill take it

change the background or you wont see all the UI

[neon idol]

ia nbp unlimited on mixboard?

[stuck orchid]

Not bad

[zealous sparrow]

did you change the background

to see all the UI

[stuck orchid]

Yes

[zealous sparrow]

ok

btw

it took it like 1 min or around that

to make this

[balmy mist]

anyone tried that amazon model(amazon/nova-2-lite-v1:free) or mistralai/devstral-2512:free?

[sterile tartan]

Do any of u use perplexity?

[unreal steeple]

Hello everyone I am new here!

[neon idol]

i use comet

[zealous sparrow]

amazon doesnt make a lot of good models

[neon idol]

always

[sterile tartan]

Well that's also by perplexity

[zealous sparrow]

their beluga models and phantom models take up so much of battle models

lots of those got purged yesterday

[neon idol]

[sterile tartan]

Use Perplexity too i guess

🤔

[balmy mist]

yeah thats what im used to, sad they still do not got i together

[neon idol]

[fleet lintel]

Good news!

[zealous sparrow]

indeed, it was annoying getting amazon every few gens

they did add a new phantom and beluga unfortuanently

[fleet lintel]

Idk why Amazon is even trying

Their CEO just wants to do layoffs. I doubt you can get any creativity out of that organization

[whole sundial]

amazon's nova website has a lot of models, also i'm pretty sure nova 2 lite is a phantom

[torn mantle]

looks good

im telling you flash is really something

[zealous sparrow]

all these novas are either phantom or beluga

[torn mantle]

but also what i noticed is that you need to be specific

like give it the extra details

or else it wont change much

lemme see if i can get something good with it

[whole sundial]

i think raptor is the unreleased nova 2 pro, completely new base model with decent world knowledge

[sterile tartan]

Is it free?

[zealous sparrow]

there's a lot of raptors

some are old prob

[whole sundial]

yes

[sterile tartan]

Link

[whole sundial]

https://nova.amazon.com/chat

[sterile tartan]

Not accessible in my location

💀

[zealous sparrow]

Probably US only

[whole sundial]

yeah makes sense, i'm from the US so of course it works for me

[sterile tartan]

Well i have a workaround

[whole sundial]

a vpn?

[torn mantle]

seahawk is meh

bad at instruct following

[zealous sparrow]

what bout skyhawk

[sterile tartan]

GG Opera Worked

[torn mantle]

got it in an endless loop twice

[sterile tartan]

💀

[torn mantle]

it has this issue