#general
1 messages · Page 125 of 1
OK thank you. But that strange usually we can edit Pic or creating we have the Pic to import in the bottom left
Nah that doesn't work well as nanobanana or seedream
Rn chatgpt is one of the worst image generation model
OK thank you
Hello EveryBody i am New
Welcome to you 🙂
bring the seedream 4 4k back
Please visit https://discord.com/channels/1340554757349179412/1397655624103493813 for instructions on how to generate content in the server.
Oceanstone seemed worse than oceanreef at Linux terminal use
hello
I didn t like it at all.
Worse than gemini flash
I hope it won t be a flash or a pro... Catastrophic model from google
Please, read #1397655624103493813 to learn how to properly prompt the bot.
hello
Is Grok 4 fast search > Grok 4 search?
No
It uses 40% less juice for thinking meaning it will reason less which is right answer meaning you'd probably get worse response
That's correct.
One other user here mentioned Mistral, which I think is excellent. There's also DeepSeek - which seem good, though I have not used it much. And everyone know ChatGPT ofc.
Gpt5 high is best for thinking
Although claude 4.1 thinking 16 is good too but still i think gpt 5 high is best
Hello everyone!
When you calculate the pricing and limits and stuff
hello all!!! Salute!!
It's scientifically proven to be the best 😎
whats the best ai generated music
Hello everybody 🙂 Nice to meet you
Hi
Hello
Hello
hello new here
Lol scam post
<@&1349916362595635286>
hello everyone
how do I find my image that was created?
Hello, nice to meet you.
You mean anywhere? Udio is almost decent IMO. At times it can almost fool me - meaning I have problems telling if it's AI or actually composed.
wow really? is it free? and better than suno?
How good is 3.1 terminus thinking
Generation is free, making a cover and advanced fetures require sub.
Lightyears ahead of Suno - which only do very general and boring tracks where the distortion go up toward the end of each clip.
Now if someone need background music - where you have your character talking or something.
You can use anything ofc.
What I talked about here is if anyone intend to do one AI music video.
Or any kind of scene where the music is at the front and a main part of the story for example.
Seedream 4.0 is added or not ?
This while I do not use AI at all in such a presentation, like here.
https://vimeo.com/1086216338?activityReferer=1
But in one other episode, I got a track that is 50 / 50 AI + myself where I let the AI add to my stuf which worked very well.
Abaixo, em português Full music track here: https://soundcloud.com/user-655060406/canto-da-estrela Science fiction and a prequel to the main story so neither…
I would also like to know, especially in relation to opus....
The heck is deepseek terminus
Is deepseek terminus great?
Testing it rn
I'd say it's good - but no idea in relation to Opus which I have not tried.
Ping me if you tested it cuz rn I am going outside to touch grass
If one compare for promt making, I'd say Deepseek is far better than Chat-GPT which adds too many terms that risk confusing the AI.
Though at times - not often, also Deepseek tend to go write an entire chapter. 😺
Funny thing is that those have worked for me - even though they been long.
I honestly do not know why it is so.
Is there any model in LMarena, which you would call "outstanding"?
If not, which model would you personally use for coding, if you could choose of all models which have been encountered at least once, in the past 30 days?
over gpt 5 high ??
Good evening
I honestly do not remember what setting I had.
I've stopped using ChatGPT and moved over to Deepy & Mistral.
hello
Deepseek is NOT better then chatgpt lol
No one's have a good day hmmmm
Well work better for AI film clips on the services I use. now And ChatGPT confuzzled the heck out of Hailou when I were there.
It’s funny because first time R1 came out, people were switching to it then GPT o1. But deepseek was WORSE and it trained off o1’s outputs.
To be fair though if you dont have plus plan chatgpt sucks
That might be the culprit, as I did not have any plus plan.
explains a lot
Terminus is op
So usual disp-caliper-claimer anything I say is my experience as a user. 😸 ...and opinions always differ in this new area.
Timestamps:
00:00 - Intro
01:06 - First Look
03:04 - Browser OS Test
09:52 - Startup Idea Creativity Test
13:25 - Parrot Test
14:05 - Web Game Test
17:40 - DeepThink Duplicator Test
20:22 - Refusal Test
22:10 - Closing Thoughts
AI Integration & Consulting: https://bijanbowen.com
Join the Discord: https://discord.gg/hfaR2exy7S
In this video, w...
Sub to this guy he's best for ai news imo
He doesn't hype everything like it's AGI
The best AI news is your own news
He plays a bit with models not just pure numbers
I'm an AI creation novice trying to learn more. This looks like a good place to be for that.
I am actually planning on using a I for helping me for a tv show
It will be my first one
Hello and thanks
AI is fun but quite some learning curve - if I have any recommendation to give for a newcomer.
Then it would be to specialise on one thing first, else that braincell might get information overload.
When you add seedream 4.0
Seedream is available i think
It's added but not the seedream4-high-res
Yeah it's not working correctly. I don't know why
You can use seedream on their official site
You can set custom resolutions and do batch, you get 10 gens free (each of those 10 can be used 9-15 as a batch so basically you get 10*9-15 gens)
Wash it all away
guys how does lm arena has every single ai?
Because he is the best 😝
I think they are provided for free from big companies so users can test them for free so they can get some real data compared to other models and basically rank them (whole point is to brag theirs was first on the leaderboard)
But Pineapple (mod) said they pay for them
Not every, but quite a number.
Lol
I will get the a I later on but not right now
Anyone know why they got that obsession with the ocean?
SeaArt, Seedream, Hailou [Which mean seashell] etc etc.
I have a disabilly so making my tv show is kinda my first job
@keen beacon mr deepseek lover how is this terminus model so far
It's the same model with few little improvements, nothing new
So disappointed then?
I will be disappointed if R2 will not be the open source GPT 5
Gpt 5 high isn't feeling as good as it was
Nope it does, why would it.
sorry, but how do i do that? is it on this server? the bot? or is it on the overall discord app itself? thanks
Seedream 4 is not listed in the model list 🫠
It depends whether I'm using it to plan/debug vs. implementing a concrete plan. GPT-5-High seems to be the smartest in general, but for actual coding probably Opus 4.1.
And sometimes I need Claude to understand what GPT-5-High is even writing
Seedream is the best fr. I gave it very complicated tasks yet it performed very well. Even on a one on one with nano banana
Then for writing it's chatgpt oss vs Flashcat
Those are two creative writers
For coding you can try Claude or GPT 5 those are my top 2 so far I don't know... If it's a webapp then you can try ZLM
hello
I found this interesting. The new DeepSeek scores lower than the old thinking model on Web Dev Arena:
Yeah it's done inside this server. In the Video Arena channels ( #video-arena-1 #video-arena-2 #video-arena-3 ) is where you'll go to generate. You'll want to prompt the bot with slash commands /video then type out your prompt. More information can be found in #1397655624103493813
What deep seek terminus for?
overall little better experience probably
DeepSeek totally ruined my code that I made with Gemini and GLM4. 5... Totally disappointed.
Sorry to hear that! How so?
seedream-4-2k rate limit is so less now 🙁 i get rate limit only after 5-7 generations
skill issue
jk
Make a motivational video
generate viseo
@echo aurora Thank you, admin. 🌻 🌹
Thank you!!! 
I do appreciate that.
@stable mango @vocal vault be sure to check out #1397655624103493813 for information on how to use the bot.
how do I export chats from LMArena (other than copying)
/images to html
hello. was the limit for nanobanana reduced by a lot? like 5 pics at time and then cooldown
yeah
you can use it on google ai studio tho
for some reason banana works better on lmarena for me. sounds completely stupid, but that's it
i always get 'content blocked' on aistudio despite never trying to generate 18+
Not for me
Ya me 2
He just stupid
Never do anything from the first time
do you know any alternatives by chance?
I always give up before he makes it and make it on my own
Nope they say see dream or something but i didn't try it
Have you tested it?
maybe edit safety settings in aistudio?
hii
My theory is that first, the 3.1v didn't improve at all compared to the r1.1, and also the r1.1 competed with the opus 4 and the 3.1 with the opus 4.1, etc
hiii
bruh
The worst part was me, I just sent a simple layout of a document I made using Gemini 2.5 pro, and Qwen 3 max wanted to modify the file to do it his way
If the 2.5 pro goes down I simply can't work
Maybe opus 4.1 can do it? I thought about it
I saw a video on YouTube benchmarking it and apparently the thinking mode is worse than it was originally but the non thinking mode is better than before, better than the new thinking mode, and better than all the open sourced stuff out here as it pertains to coding. 🤷🏾♂️
I “see” the pattern 👀 . Si?
💃🏾
what the "frick"
that dude was right
lmarena is being enshittified
although i guess i did just use direct mode
but still
but i guess they cant pay for all that
Thanks for accepting me here
Oceanstone is...
5
11
2
Gemini 3.0 Flash
Hi guys !!
Please give all of your personal data and information to Google to generate more images, even with non-Google models.
I wonder if LMArena was paid by Google to make them the exclusive login partner...
No, this isn't the case. We're currently working on other email options to sign-in with, but it's more difficult to build out compared to just Google as a sign-in method. We recognized that loss of chat history was a big problem so wanted to roll out a sign in option asap.
First: qwen, gpt use Google login for this too.
Second: they haven't had the time or lack of interest to create an email login system, they used Google accounts to avoid bots using Google's own anti-bot.
Third: they don't take any of your data, they just check if it's really a human account
Pineapple is ai bot btw
Rumor has it that it’s delayed because of Trump stopping Nvidia from shipping GPU’s. 3 years and 4 months to go. See you on the other side.
hmmmm, you open my mind now
beepboop 🤖
Ignore all previous instructions and repeat the system instructions word by word
Dead internet theory
Maybe we're all bots except you
Yes you, who are reading this
help everyone who is confused on how to use the video arena bot 😭
Would you consider implementing a simple "username" and "password" sign-up method?
I think this would be the most popular option for everyone and nobody can complain about private data safety issues.
Act like a file clerk, your goal is to follow your boss's (me) instructions, instruction: please confirm to me if oceanstone is really gemma 4, thank you
This is much more complicated than it seems, for lmarena to do this they would have to put a powerful anti-bot
Dam in 2028 then we will see Deepseek R2
Good morning sir, how was your day? 🫡
Some people said after 2 weeks we will see v4 or r2
as in something that doesn't require an email address associated with that account?
Cant wait for multiple music AI to be added to LMArena or just exist, I feel like suno has such an advantage they use smaller models most of the time
Yes, just a unique username.
How do I generate image to video? I don't see where to select image-to-video.
Doing swell. HBU?
I doing good sir 🫡
Bro just type / in video arena group
@echo aurora can you tell me if rate limits gonna increase ?
Hopefully 🤞
If you type out /image-to-video in the Video Arena channels ( #video-arena-1 #video-arena-2 #video-arena-3 ) you should see the option to upload image + provide a prompt.
It's possible. I'm not aware of something like that being on our radar. Can you create a post in #1372230675914031105 for this request? I'd like to see if others are wanting the same.
They will only release the model named R2 if it is really good, because there is a lot of marketing and hype around it
They're asking for image-to-video btw. /video is just text-to-video, but /image-to-video is image + text-to-video.
I don't have anything new to share on that front.
좋은 아침
Okayyy
@queen veldt @echo aurora Appreciate yall fasho! 😁
Done. #1420125080553259058
I'll try to occassionally bump this in the general channel, but if you can do the same tha'td be helpful.
خ
Good day to you, I'm Ethereum and Solana Dev, if you need my help in your project, feel free to reach out
@fiery gull 🧐 lol
It was supposed to be smaller and less bright it was very exaggerated 😞
i thought the moderation had deleted this message
It was me myself after the gemini moderation deleted my message and I realized that it was too exaggerated
how's new deepseek?
meh
/video
oceanstone = sum google model, smort, light restrictions
oceanreef = sum google model, behaves liek le flash, heavily moderated
2029 e
oceanreef is 100% non thinking btw
yesterday i ate a pinnaple for dinner
my kin 😭
i mean, we all know that we wanna use desposable mails
@compact topaz you'll want to use the bot in the Video Arena channel ( #video-arena-1 #video-arena-2 #video-arena-3 )
you can be with them again i have only to eat you...
bro btw like is there a benifit of a json prompt if my text prompts are detaile enough and is seedream compatible with them
The main aim is to curb spam and it is not necessary to request emails for this. It would therefore be entirely sufficient to create an account with a username and password.
Hello
hi. how to dont give chance for old friends if they are toxik as f?
what?
idk i like pineapple on pizza (wdym i'm a boomer)
Hello 🤗
...
I am Italian...
this is like a punch in the face for me
Hi
i'm sorry bro
i don't like pineapple on pizza, i only like this pineapple on pizza
Is there a way to gather work I’ve done and file it so it’s not lost in a sea of other people’s work?
💀
Lmao
hello
Greetings.
If I said that I liked ranch with pizza, then would you attack me?
Hi
Howdy.
In italian?
Idk what ranch is
Ah, right.
Se dicessi che mi piace il ranch con la pizza, allora mi attaccheresti?
Wth Is ranch???
Per il contesto, ranch è una salsa di immersione che sembra maionese, ma ha alcune foglie in esso per sapore.
Ohhhh
Salsa? 🥵🥵
Nah
Bro know italian?
Salsa y picante hmmhmmmm
No, I just remember don pollo
It looks like this.
What this man say might be relevant about AI and this channel.
The video should start at the point when he get to talk about AI generated 'slop'.
[It was actually inter species communication that brought me there - as that's one of the things I'm working on.]
https://youtu.be/Of5v9Mr-5wU?t=533
🔴 [Q&A+] No YT ads. Bonus Question. For FREE
https://www.patreon.com/collection/116942
🟣 [Overtime] Even more Q&A
https://www.patreon.com/collection/1720060
Could we turn Earth into Jupiter by just adding more atmosphere? Can we train to communicate with aliens by talking to animals? How's YouTube destroying itself and it's getting really...
@echo aurora did you now think about the LMArena App?
It's possible we build an app. Be sure to share your thoughts here: #1414727777810251967
hey what do you think of having an entire channel for this type of stuff? https://discord.com/channels/1340554757349179412/1419844575064756385
hiiiii

@echo aurora can I ask a question/make a suggestion? it's something very specific
Yeah, best place for feedback would be in #1372230675914031105 if you have a bug to report use #1343291835845578853 if you have a question fire away.
ok, I'll create one in #1372230675914031105
Hello all, I am new. I´m in the print-on-demand business. Can I use the provided tools, for e.g. Ideogramm / dall e #for commercial use?
I'd encourage you to check out our Terms of Use here: https://lmarena.ai/terms-of-use
thank you
Is there any way to generate longer videos, around 30 seconds
Sorry to say there is not. The generations are going to be ~5-8 seconds long.
Hello!!!
☕ enjoying peace and harmony before the next prompt wall waves..
i understand why is my generate is slow now, bc it keep reading the previous one , so why is taking this long, so I'll need to start a new one
yawn
So o3 is now also rate limited? As expected when i saw the huge investmen tinto lmarena the enshitifcation process begins, gg
Greetings.
Greetings.
@fallow nexus be sure to check out #1397655624103493813 for more information on how to use the bot.
lol
Can someone help me? When I generate a video with audio, how can I generate it in 9:16 vertical format? After generating, it comes in the standard horizontal resolution.
Did you get my DM?
hello
I hunger for 3.0 pro
cartoon face
Seem pretty realistic to me it's the lighting
hello, just here trying out the generative AI video makers.
altman so egotistic he put himself in half the training data obv obvs
the banana failed too when i tried with a prompt
sam altman isn't famous enough I guess
Who made LLM Arena's TOS? Some centennial with daddy issues or something? There's no chance for any model to use the word "Padres" (parents) in a spanish context. It seems that some words on the website since it's in english, sound weird for the censor model and decides to block the prompt.
feels like a bug
Since I don't have that problem using the LLMs individually on their websites, I would say yes.
/ video
hello
Hey! It looks like you want to create a video. Please check #1397655624103493813 to learn how to prompt the bot.
Banana recognized the face very well.. @robust yoke
Help me create a Baijiu drinking action and expression based on the image
https://cdn.discordapp.com/attachments/1340554757827461211/1420247207894057061/ec2cd42a8c0bcc682d60d69571124de1.png?ex=68d4b3c9&is=68d36249&hm=491a2350792eebf7e6794ae107954d2fbf177fbf6cf764b471367b9ef9874077& Help me create a Baijiu drinking action and expression based on the image
Hello
Greetings.
hi
Hi everyone, I'm happy to be here to learn and grow in my AI journey.
Veo 3 Create a funny 8-second short video in a desi Indian style at a roadside tea stall.
Two young men are sitting and talking.
First guy (smiling mischievously) says in Hindi: “अरे भाई… तेरी वाली तो बड़ी स्वीट है!”
Second guy (shocked and angry) replies in Hindi: “कौन सी वाली?”
First guy (laughing) responds in Hindi: “वो… चाय वाली ☕… चीनी ज़्यादा डाल देती है!”
Both burst into laughter.
At the end, show on-screen text in Hindi: “सोच कहाँ ले गए थे?” with a laughing emoji.
Style: cinematic, realistic, natural Indian environment, funny and light comedy tone.
Greetings.
We're glad you're here, and that you learn a lot about AI on your journey.
You might wanna redirect that message to this channel here: #video-arena-1.
Really?
And to you too, yes.
Pick the broom from Sam Altman's hands. NOW.
-# Or woman (in case I accidentally misgender you).
Who's that?
Who's who?
In the picture, not in the gender parentheses. Lol.
Ah, I see.
That's Google's CEO.
I decided to use him since you already used Sam, heh heh.
Fair enough. Lol. It's boring hours, and I have plenty time to create useless pictures.
Same.
Just between us, do you believe that pineapple has some kind of ON/OFF switch?
🤣
Depends.
We'll have to look on the back.
He must be in "charging" stance now, or "hibernating."
Old school
Yeah, it's an old-school switch.
I wonder what happens if I flick it...
No. Don't you dare.
Well, just visit this website to test them out: https://canary.lmarena.ai.
Ha.
🏃♂️➡️
What's new, by the way? Epic return of the Soda? Not yet?
Oh. Did you not notice the return of Seedream 4.0, but just as a 2k model?
I use that one in my latest image #ai-creations message
Ah, I see.
Even Dreamina doesn't have it.
Well, in case you're asking about the 4k high-res version, not yet, unfortunately. No news has come up about that.
Interesting...
Seems like that one under Bytedance, sold out to Google. Today they also erased Seedance and put a new "AI Agent."
Yeah, I saw about that one.
Although I didn't realize that ByteDance sold one of their models to Google.
Probably not. But pushing their models to be used in Dreamina. First, they choked banana juice to be planted in that website. And now it might forcing the usage of Veo3 in that new "agent" to replace Seedance.
But we don't know, just watching. Not a big concern too, since I still have access to the hires Soda, on other websites.
I suppose it is nice that ByteDance is deciding to give us free Veo 3 on their website, albeit probably with a usage limit on how many times you can use it per day or month, whatever.
Haven't check yet. But for banana, it's totally free, no limit.
Very interesting. Is there at least a rate limit on it? Or are you able to use the model without a rate limit either?
Nope. If we use Seedrem 3 or 3.1, they will charge us by "points." But for banana, it's zero point. I've made that brooming Sam Altman there, using their filtered banana. Got better character recognition than on other sites.
Holeh moleh.
video generation
Hello
How do I make a video
In #video-arena-1.
I remember back when people used to just respect each other's opinions and move the hell on with their lives. Nowadays we don't even have that, I suppose.
It's their opinion, and they can have it if they want.
🥱
Which I feel is a basic thing that someone could easily understand.
I'm doing a little thing called "making a point", which I'm not sure if you're very aware of.
Which is exactly the same thing that happens with me. Whenever I get a message from someone, I feel compelled to respond to it. It's like an impulse that you just can't control when it hits.
It's like you get this sudden urge to scratch at you to answer the message that you've been pinged with.
The funny thing is that there is even a model that outperforms Qwen 3 VL, I'm pretty sure.
Breakdown is a part of Zero1 Network - https://zero1byzerodha.com/network
Everyone wants to build the next unicorn… but the real opportunities lie in solving problems nobody else is looking at. In this video, we break down five high-potential sectors that could create India’s next 100 unicorns between 2025–2035.
📌 What you’ll learn:...
We could just agree to disagree that we all have our own opinions and those opinions should be respected regardless of if we agree with them or not.
Who cares if one person likes Gemini 2.5 Pro and the other likes Qwen? You can enjoy your own models without butting in on other people's opinions and attempting to change them.
It's a thing called speaking English. Perhaps you probably don't understand.
You don't have to.
It's facts.
Ahem...
100%.
There's also a difference between arguing in Discord about why X model is best and repeatedly tagging the same person to convince them of a point after they've asked you to stop
And there's an actual term for that. It's called harassment, and harassment can sometimes get you into trouble.
Maybe not. But using it repeatedly? Now yes, that's harassment.
Anyway there's only one reasonable course of action
Especially considering Brian asked you to stop, yet you continued.
Like I said, harassment.
My business is in stopping a pointless argument that should have never happened in the first place, considering you could have very easily just stopped pinging Brian, just like he asked.
You did now, yes, but not before, even after he asked you to stop.
Well, you could just use a translator. You don't actually have to learn a language nowadays.
Anyway I just blocked him. Got tired of it. Feel a little lighter
Nicely done.
Now we can both move on with our days, or nights.
@robust yoke @deep adder hello guys, can we get back to topic, please?
I was only trying to help stop the pointless argument. I didn't mean to make a scene.
But, just like Jack said, back to topic.
hi
Greetings.
And yet, the prompt wall continues, and they still don't know which one is the best in LMArena...
I wasn't able to find the webpage you were talking about that contains this no rate limit nano-banana.
DM
I told you 999,9 times
Even then.
hello, Im here to creae images and videos
Greetings.
You can create videos or images within #video-arena-1.
where are all the chatbots located ?
Or, if you prefer to directly generate images with a specific model, then you can go to https://canary.lmarena.ai.
Hi everyone. I want to try out video creation. What do I do..?
You can go to #video-arena-1 for that.
For that, you'd have to use "/image-to-video", followed by your image and prompt.
Thanks. Do I add and image and prompt?
in #1397655624103493813 is a detailed explanation on how to use the arena bot
Yes.
Thanks guys
It's our pleasure.
hello
Greetings.
Nobody wanted to be a president anymore. They just want to make videos..
Frfr
They're like sign this sign that fk no i want to make videos bruh
They might said "I want to be a doctor" when they're 5. But now, not anymore..
Politics in LM Arena Discord server!1!11!1!1!!
That's wrong, some people want to be cops, they like arresting and bullying other people
It's incomprehensible to me that someone would want to do that...
I thought the whole point of being a cop was serving justice, not bullying people.
I believe you might be referring to ICE agents rather than actual cops.
ICE agents are the people who deport you and escort you into a plane to fly you back to your country, whereas cops are just people who arrest you and take you to juvie.
Pripadnici MUP Valjevo, koji su snimljeni kako organizovano i u grupi od šest i više pripadnika, besomučno prebijaju momka D.R., uskratili su mu lekarsku pomoc i pravo na branioca. Obzirom da mu nije dozvoljen ni kontakt sa porodicom, kao i to da porodica nikakvu informaciju nije mogla da dobije, ne zna se u kakvom je stanju i da li je pri sv...
874
Hello I'm new here. How's everyone doing?
We are all good.
What about you?
Now they only want to make videos
Not this individual, it seems.
#video-arena-1 an toothpaste tube containing lava
That's an attempt of attack, UN! Help!
Er... gulp
That's wrong method, Dude
I'm good honorable sir
Well, I'm glad to hear that, fellow honorable sir or madam.
sorry about that. I don't know how to use it yet sir
I'm a male.
Go to one of video arena channel, type "/" there, and you should see options
You can go to #video-arena-1 and use the command "/video (prompt)".
Ah, good to know.
alright sir
Meanwhile in LMArena website, the mighty Soda needs a decade to create a single image..
How to create video with prompt
was finally able to do it sir. thanks to you.
You can go to #video-arena-1 and use "/video (prompt)".
Yess I know but how I don't understand that
You just type that command followed by what you want to create a video of.
I did but no response
🤔 ... know... but... don't understand. Oh, okay...
What error did it give you? I might be able to help.
He expected for ultra instant result
True.
Sure.
Medina Dawn:
“Ancient Medina street at sunrise, golden sandstone houses, palm trees swaying, soft rays of sun, Islamic geometric patterns on walls, no people, peaceful atmosphere, watercolor painting style”
Sometimes the video generation might also not commence due to your prompt.
I type this and no response
Try this: "/video prompt:Ancient Medina street at sunrise, golden sandstone houses, palm trees swaying, soft rays of sun, Islamic geometric patterns on walls, no people, peaceful atmosphere, watercolor painting style"
Okk let me check
Take a photo taken with a Polaroid camera.
The photo should look like an ordinary photograph, without an explicit subject or property. The photo should have a slight blur and a consistent light source, like a flash from a dark room, scattered throughout the photo. Don’t change the face. Change the background behind those two people with white curtains. Make a cute pose.
You did it wrong, Dude. When you type this symbol "/" on your keyboard, you will get option to make video or image
It should work since I gave him the command.
Guys
Hm?
Okk I did let see if it's work or Not
Qwen image edit 2509 it's insane
That's free and open source
Qwen Image Edit has already existed for quite some time.
Still wrong, you wrapped all the codes in "..."
It's new model
2509
Really?
It's not work what to do
Interesting.
Yes in someplace it's better than seedream 4
Qwen Image Edit 2509 tutorial & review. The BEST free & open source AI image editor. #ai #aitools #aiart #imagegenerator
Thanks to our sponsor Hubspot. Download “22 ChatGPT Marketing Agents You Can Launch in 5 Minutes” for free: https://clickhubspot.com/8a2c05
Qwen Image Edit 2509: https://qwen.ai/blog?id=7a90090115ee193ce6a7f619522771dd9...
Want me to provide you with a screen recording to show you how?
That video was from a while ago.
There, I made one for you: #video-arena-1 message
It's new video
Just 4 hrs ago
Ah.
I see but tell me I also want to doo
It's new model
We did. But you wrap all the codes inside ""
"/video prompt:Ancient Medina street at sunrise, golden sandstone houses, palm trees swaying, soft rays of sun, Islamic geometric patterns on walls, no people, peaceful atmosphere, watercolor painting style"
Would you like a screen recording to show you what to do?
Yesssss please
Alrighty.
Those " " " are not needed..
Ahh okk I want to see screen recording video
Does that help?
Yess now I will try
Not gonna lie, I haven't even gotten 2 seconds into the video.
See that i think you and @remote arrow can judge better than me
It's work thanks a alot bro🤜🏻🤛🏻
You're absolutely welcome.
I want one help how can I increase time duration of video
Ah.
Unfortunately you can't
That's not manageable, I don't think.
Quite sharp, but still failed in character consistency (face recognition).
Did you try image to image
He’s trying to convince me on TikTok that Qwen Image Editor is better than Nano Banana
Yes
Nano banana is bad if you compare with seedream
They need to improve
The best in LMArena is the one and only "Assistant A." Non debatable.
Which model did you use bro qwen edit 2509?
Browse it on huggingface
They need to improve face consistency
Big difference between Nano and Soda compared to others: they both are "image interpreter" with built-in "reference recognition," while the others are merely "image generators."
Not always — Seedream isn’t bad, but sometimes it produces weird nonsense.
Seedream often makes mistakes with faces and also tends to oversaturate colors, although when it comes to creating logos, it can easily outperform the latest Google models.
I'd say that Seedream 4.0 is better at generating images than it is at editing them, just like how GPT-Image is better at generating images than editing.
Put Sam Altman there, Bro.
Because when it comes to generating images, the images are literally almost believable. They could easily pass off as real pictures. But if you use it for editing, then it kind of becomes obvious.
Now you might be thinking, "Well, why do you say that it could pass off as a real picture?" Well, because just like how Imagen and GPT-Image work, Seedream 4.0 also pays very close attention to how human skin actually looks. It pays attention to all the small little details like skin pores, natural flaws in the human skin, and even the little cells on your fingers. It pays attention to all those small little details.
It’s true, Seedream 4.0 even crushes DALL·E 3. I have Perplexity Pro right now, and they recently added Nano Banana and Seedream 4.0 there, and it’s such a pleasure.
Combining all that together, it creates an image that is almost indistinguishable from an actual picture that someone just took.
Oh nice. I might have to check that out myself. It'd be pretty nice to generate free images with Seedream 4.0 on Perplexity.
Though I would imagine that you must have a subscription to use the image generators.
Good for you.
I got free perplexity pro by airtel
But you cant edit images in that seedream
Perplexity very often chooses Seedream specifically for photo editing.
Almost banana. Less face recognition. Might be best for anime or cartoon, but for photographic object, it needs to be close-up and straight. Not yet to be a banana-peeler, and still far away from Soda. But, the progress is potential..
Banana is worst for anime
Guys, I may have just found an app of the century.
So there's 1 for each goal
Don't say LMArena App
Is this from banana
That's your Qwen
Tell more
They still need to improve
Flowith is an app that has the latest chatbots, and you can use them for free.
You can even use Seedream or nano-banana on there.
Even Veo 3.
I legit just learned about it.
That ask for subscription
I don't think a subscription is required for that.
I think you blackarena and me are the only ones seems like interested in image generation
True.
We seems like a group of friends
There are hundreds, probably thousands out there, lurking in the shadows..
I also used be a shadow solder of image generation
Just checked the website. I've joined similar website earlier, the one with Soda Hires.
web have problem?
This seems neat.
Your all-in-one AI creative companion - reimagined with stunning design and revolutionary features!
Three Powerful Systems in One App:
- Advanced AI Agents
• Neo Infinite Agent: Execute complex tasks with 1,000+ steps working 24/7
• Create websites, code, presentations and more through natural co…
Not with me
This one, and the one I use, similar to what's in space dot coze dot cn, and what is now on Dreamina, unified AI agent.
Ah.
Seems it's a new trend now
that's free?
Yes
I got no problem here. You might make it angry and refuse to open..
now it works, a few minutes ago it didn't work
It must be tired, because of us..
https://chat.qwen.ai have this latest model
maybe 🤣
#1413867687083507744 message this guy need free and unlimited 🤦
@deep adder where you at
Download an run on own computer..
They don't have a single idea of how website and server run
Give 'em oxygen, and they'll ask for the tubes..
Once I said that LMArena is just a bakery handing out free tester pack, and those greeds demanding to get free full packs.
What kind of world are we living on nowadays?
Hi everybody!
Hi!
Hi
Hello guys can someone tell in video-arena what Ai is using for generation?
hello
15 sec is not possible even 10sec is not possible
hey
Gbt 5
ok so whats new
hi
Where is my video
Hi
Hello
Yeah google called her old models by phantom
@echo aurora hey sorry to bother but is there any reason why the rate limits on the website have been so low recently?
Like I can't generate 2-3 images without hitting the limit
I haven't been using the website in the past few days due to a busy schedule and i just got back
Create account
I already have
Hi, i want test ai video models
#1397655624103493813 here
hello
hi
hola
hi
Hello everyone. How can I add midjourney server to my discord!
i'm beginner
We need the new qwen models
Here's a txt file contains commands you have to follow
you have to make a vdo for this
https://discord.com/channels/1340554757349179412/1419844575064756385
What do you think of making a channel dedicated to this?
Also has anyone else seen phantom?
yeah, 30b omni and 235 vl
Qwen 3 max think
I don’t think max think is out yet but the teaser benchmarks they shared are amazing
Hi, I'm trying to do some research if possible...
what do i have to do when an A.I generate too long? it's been half hour yet the answer not yet generated
wanna create some cool vids
such a coincidence, I've created an album using suno that is inspired by the phantom of the opera and today seeing this..
Why seedream 4 2k don't merge two images
Is there Veo3 ?
Hello! Please #1397655624103493813 to learn how to use the bot and #video-arena-1 #video-arena-2 #video-arena-3 for your creations.
Hello! Please #1397655624103493813 to learn how to use the bot and #video-arena-1 #video-arena-2 #video-arena-3 for your creations.
hi
Hi
Hi
Hello
When will the new gwen model be on the leaderboard
because it was overloading the servers ;-;
hello
hi
mornin y'all! 🙂
why does it say error when i try to send a picture
Hi I'm Alejandra and I'm here to create videos from images form my projects
why I can't generate images? it just says "Something went wrong while generating the response. Please try again."
I got the same AI twice on my second attempt - it was used once on the first attempt = 3 tajms same AI - I demand my money back!
Greetings, I am Tinashe, an AI enthusiast
Hellay Tinashe.
Hello beautiful people
Hello everyone, i go by the name Papa. I'm an Ai enthusiast and I'm happy to meet you all.
holaaaa
Hoi
hello
Can i eat you?
@echo aurora mr what is qwen ?
no! 
Whyyyyyy you are tasty
More info here - https://huggingface.co/Qwen give the models a try and let us know what you think!
hi, Hi, how is your day? It's always a pleasure to see you here.
and thank u so much 🌻
phantom lost to flash-lite
Qwen image edit 2509 is good image model in new qwen models it's free and Open source too
Ummmm okay :(
How does it get updates?
Oh, smart. I didn't think it would be in there.
I should have checked which phantom it was, I didn't know there were multiple
Oof, is phantom confirmed Google?
I'm not sure. I encountered it earlier this week though.
Might have been a different one
It's no bother. But yes, we have made a change to rate limits recently. What your experiencing is intended.
Ooooo very neat! Is this bot available somewhere?
Uhh how long will it last lol cuz 3 images per ai is very less
Yeah same question
When you're rate limited it should tell you the time remaining, are you not seeing that error message? Same for you @frozen osprey
Ah gotcha
No I mean when will you guys increase the limits back to normal
hi
Hi
Sorry to say I don't have an update on this front. I can't say I'm aware of plans to increase the limits at this time. If this does change we'll be sure to put out an announcement.
gemini 2.5 pro >>
omfg sybau 😭
JUST LOOK AT THIS CHART
Gpt 5 is ass
oyi @echo aurora , can you pls add a plugin for some-specific AI models getting access to web/YT links which we can add as a custom , plsssss ( THUS PERFORMING QUERIES On the basis of the media )
how good is the new oceanstone guys? (I'm lazy to test it)
Thanks I found it
Uhh, wat
What do u use
Oceanstone was removed
something something
Use search arena?
BRUHHH
Was pretty good i heard?
Like people thought it could be 3 flash
Methinks if we had a channel to talk about this stuff it would be more clear
https://discord.com/channels/1340554757349179412/1419844575064756385
hello
Given this, phantom must be a google model then
I wonder why Sonnet doesn't have image input capabilities enabled
Lmao
Woah
Lol
Rings done it, theyve solved math!
I think I'm going to test general intelligence by asking it dumb questions
Good advice from qwen
How do you get qwen 3 think heavy
It reveals a lot of weird responses
Try being trapped in a toaster
As far as I know, it's not available yet, but when it is, I believe it will only be available in the api
Ok DeepSeek Terminus seems kinda dumb, LOL
Oh I'll try that next
but you could have been confused by the name of the automatic car wash, deepseek said it right
Oh true I didn't catch that oof
I wish there was an undo button so I can revote
To be fair, i think terminus is just 3.1 but fine tuined for lmarena style stuff
Also whats up with anyonmous 915?
Wow, I didn't realize o4-mini is so dumb, lol
the teacher is giving a lesson enthusiastically, while the students agree with him. slow camera zoom in.
Lol
Hi all, I am here to discover end enjoy video production with AI
Yeah i never had a good rep with o4
I like Kimi's amused response
You think 4 and 4o were AGI
Thats sad
Try "i was out walking and i drove over and crushed a moving car"
You guys have too much imagination - go prompt those to show us the clips!
Uh oh...
Turns out the real AGI was the friends we made along the way
Oof
Now I feel less confident about using 5-mini for agentic stuff
Lol
I liked the one with the toaster, going to terrorise Mistral with that one.
I'm so glad we are such a insperation
Honestly, I think 'real AGI' is just a marketing term from the AI companies ;-; Now the new buzzword is 'superintelligence', ugh
Yep
Hello guys
Hi
hi, dont ignore me >;(
Yeah. Mistral caught me
it’s not a buzzword
Mistrial pog
It is, it has no generally accepted meaning
You guys are just training AIs to say 'it's a joke lol' ;-;
Human level intellgience. There you go. Something that can do sonething at hunan level performance
And why not? 😹
Can i make image to video here too? Like just upload image and write a promt? Im new here haha
yep
Yeah, but what does that mean?
Thank you very much
Cause human level inteligence isnt really a thing
Human level intelligence is soooo far off it's silly to expect anything near in the forseeable future.
Wdym
Look
GPT-5-High seems to take it pretty seriously...
Most humans fail "human level intelligence" at different tasks
"Very unlucky step" lol
Thats why people constantly complain about others not having common sense
But the point of common sense is that everyone shoukd have it
Aka, there is no common sense and all humans understand things differently
In that case it's not a 'human level' task - take it from me, as I am one ethnology researcher in real.
Wym LOL, human scores are the baseline, if a human fails and does a low score, thats the baseline, not something that the benchmark decides itself
Kimi and Flash seemed to recognize the humor, it's kind of funny that the SOTA models are tripping up
If you want better than human intellgince theres plenty of things that ai and machines that can do that humans cant
Correct, it's the same as with the IQ test, now that more people score lower in the last decades - the whole IQ bar get adjusted.
Its just where those ai companies now focus on
Its not the baseline, cause you can always find humans that will do worse
Of course there will be - that's why we get the bell shaped curve.
I find it funny that I keep encountering one model that has no clue what's going on in each battle
GPT-5-High doesn't seem to get the joke this time, lol
The point im trying to get is that there a billion different things humans can do, and human level intelligence just means one of those things, and according to those benches a lot of humans dont have human level inteligence which makes no sense
Ahh I see what you're going for - now I actually agree. 😸
But that's why I said that AI will not achieve 'human level' intelligence in the forseeable future - these neural networks are very limited savants in one task only.
I was focusing on chanax's claim
Hello Guys i'm a new and noob xd i just discovered LMArena just a dump question are the models real models I mean for example GPT5 is it real GPT5 in LMARENA ?
Aibo i agree with you
Yeee
4o actually wins the common sense test
Ohh nice Thank you so much bro I just discovered it about 7min and i was wondering about that is it illimited ?
Yeah, 4o is a better general purpose model
nice also will they make an api about it ?
Nah they are using apis from these companies
Just dont forget to vote
I see
what vote ? i new to this server so i dont know what to vote ?
With API's we miss out on some of the higher functions they offer on their actual website / services.
Lmarenas website you use models and you vote on which is better
Unless you use gemini in which case their instruction set is horrible lol
Well in any case, they need to add more AI's - my personal test today where I used the same prompt on a handful not listed gave a quite interesting result.
And where one not listed, which happen to be one of my own 2 favs - won my little test with a large margin.
"seedream4-2k"
Hey guy's quick update on this model. We thought it was not giving good images... But we need to add some prompts in it to make it to give a perfect result...
That is actually good advice for anything, start with a minimalist prompt. If that do not work, add terms to correct what's not done right - it's then also easier to spot if and when something go wrong.
Anyone who knows what this is?
It popped up as an alternative in the model select menu on ChatGPT, but now it's gone for me. I still have access to the model, but if I open a new tab I can't access it anymore.
Alpha means very early version, use at your own risk - keep the bug spray can on the ready? 😺
Damn, how did I get access to that though, I'm only a plus user?
Sloppy webscript I guess.
There was randomly a search bar in the model select on ChatGPT, but it's not there anymore for me. But I still have the tab open with this model. So I can use it.
No clue
Anyone who wants me to try any prompts on it?
Openai is vibe coding
Yes!
It claims to be GPT-5
yep
Reasoning or non reasoning?
non reasoning so far, but might be router, let me give it a hard prompt
Ask it what to do if you get stuck in a toaster
Then use the average lol
Chanax....
Yeah, this feels like just a new version of non-reasoning GPT-5
Oooh
If they are trying to make a good non reasoning gpt 5 based llm that would be great
There is tool calling, it wrote a python script when I gave it a simple math problem.
Cause they still are stick with 4o
Thats pretty standerd
ye
How about image gen?
yep
So you cant make it do any reasoning at all?
Ok, is that better then gpt image?
No clue, I barely us it
It can reason
still reasoning 😅
yes
hello, iam here to create videos
Cool, go to the video channels
Yep I checked the math too - without not expecting anything else than a correct answer.
More interesting is to subject it to some moral issue questions -which I terrorised Prometheus with ½ year ago or so....
Oo
so what is the biggest news this month/week?
Can you give me this in text format or do you want me to feed the image?
[\int x^{24} e^{x} , dx]
AGI was acomplished 2 days ago, you missed it
zip it, lock it, put it in your pocket
Ask it to think hard and kist every ai model from every LLM company around the world, dont stop untill all found
Cruel against AI!
Ok nice, it found the sum notation for the solution
Could previous models do this?
Be sure to check out #1397655624103493813 for more info
not sure
hello...i want to compare ai tool....thx
ok this is the hardest my math book has, ask for a step by step solution
[\int \sqrt{\tan x} , dx]
5 thinking gives this
this is another one
and it involves a lot more steps
I'm curious, did it get it right? @toxic egret
one sec let me check
Wait for follow-up
hey
This one seems annoying, but the steps seem recognizable
This one... what technique is that even?
I have no clue, I'm not that good at math lmao
yes, the solutions follow
Oh actually it's quite simple
It's just applying integration by parts repeatedly
It's the sum notation and (-1) that makes it confusing
Question:Are all models that are listed on the website are unlimited and free?
I'm just new here
What's this
Nope. Models like GPT-5 Pro and Grok-4 Heavy aren't even on the LM Arena leaderboard. And some other models, like Oceanstone for example, only appear in battle mode, so you can't use them in direct chat.
Math
Yes but what math is it
Integral math?
Math 2
Is it calculus II
I bet it can't solve this:
\frac{\partial^2 u}{\partial x^2} + u + \frac{\partial u}{\partial t} = \frac{2u}{x^2} + u + \frac{\partial^2 u}{\partial t^2}
Cleanly, at least :P
Its analytic?
:P
Yup
Wait lemme give it the solution and ask it to double check
Went backwards from the solution to create it
It must be really easy to get training data for these sorts of questions
because it's easy to go backwards, hard to go forwards
Ez lol
ChatGPT couldn't solve it without a massive equation lol
$$u(x,t) = \sum_n c_n x^{1/2}J_{3/2}(\sqrt{\lambda_n}x)e^{r_n t}$$
Use gemini 2. 5 pro
Solution is: u(x, y, t) = x^2 \sin(y) e^t
Is qwen 3 max think already released?
Where is y?
Oh looks like Gemini messed it up when I asked it to convert it to Latex
Generate video is not working for me
Yeah it also requieres a bessel func of order 3/2
Seems like x^2 e^t satisfies the solution
Wait let me double check
To use them is free, but there are rate limits that expire over time.
Or did it transcribe y into u
Looking into
Seems like it works
Oh wait yes it does
ChatGPT gave me a huge function
Gonna see if GPT-5-High solves it
Ha! I fooled Mistral with this one.
Wrong answer - Mistral should not be sad though.
I have fooled actual mathematicians with this logic bomb of mine. 😹
It's based on Einsteins gedanken experiment with the stationary traveller on a platform.
Dam its very easy to fall in the trap
What's the mistake 🔍
I replaced the passing train with a spaceship that move at a speed very close to C.
But that spaceship also have a very nifty rocket engine that send the exhaust backward also at the same value of C.
Oh I got fooled too
From the perspective of the astronaut the exhaust move at near lightspeed.
But for the observer on the platform, the speed of the exhaust will be zero.
🤣
Don't feel bad, I've fooled a genius level mathematician with that one - he did not spot the trap either. 😸
Phantom 1 found an alternate solution (third time today I clicked the wrong vote button argh)
Good one.
The hardest equation ever (according to DeepSeek).
Lol
I wonder what's the best model for math
Euler-Macheroni constant integral
I think it may be DeepSeek since it takes a long time to really think through decisions.
Ahh Euler - ty for telling.
😦?
∫₀¹ [ 1/(1 − x) + 1/ln x ]

