#general
1 messages · Page 117 of 1
How do I generate the images or videos? I send the prompt and the bot ignores it.
copying is a normal norm in competition. and all we users gotta do is just exploit the hell out of it
May fault. It's actually red rectangle, with a white triangle inside it..
mine
I guess for the time being, nano-banana here is free of charge. Guess Ill be using this but for now I ran out of ideas...lol
monterey also
Nano or Seedream 4 ?
Seedream 4
guys this is generating from last 5 to 8 mins
is this normal??
also in console it's saying post retry blocked
lmarena.ai/ingest/e/?retry_count=1&ip=1&_=1757768469213&ver=1.223.3
net::ERR_BLOCKED_BY_CLIENT
did "upload image" got removed or
do json prompts for image generation yield significantly better results or nah?
idk
hlw
Hello everyone! Just found you guys. Looking forward to growing together with new tech
Hey im new here, im hoping to learn alot of new stuff together!
What Ai would you guys recommend for creating Product Pictures? For Example for an Ad? If anyone has a good suggestion , i would appreciate it alot.
Seedream: with you I can see a real dream
@echo aurora Yo man
.
This member ask me if you coukd answer to him at the question that I have pinged. thank you!
All hail seedream 4 highres
Seedream 4 high res is impressive in realistic images
Officially seedream 4 highres defeated mid journey 🔥🔥🔥🔥🔥🔥
If it have ratio it will be the best
Officially seedream 4 destroyed all ai models
I'm so happy how
There is a site called doubao where you can use it
But need a china number
Is it working with high res image ?
Then finally lm arena is a world's one of the best link
@echo aurora is it possible to have image output in different aspects ratio give the option in bottom
A new image model!
That would be a nice feature for sure, that idea is something we're putting thought into. cc @prime mulch
And yeah pretty much
sometimes I feel shaped like that too
Thanks for the consideration pineapple and that new model is insane
Be sure to check out #1397655624103493813 for more info on how to use the bot. Let me know if you have any questions.
Worth checking out these leaderboards
Did you have a good time?
guys this is still generating
tysmmmm
5-8 mins
That sounds like it's stuck, you may want to create a new chat
broo wth
i was in middle of making a new feature
how would i catch up now
f
Wrong
Seedream 4 High Res is available in Direct Chat
I’ve been using it
Since yesterday
Oh right, sorry just woke up
Really sorry to hear this. It's possible, after a bit of time + refresh it'll start working. This problem with models getting stuck (with long conversations/prompts) is an issue we're aware of.
it's stuck in generating from almost 3 hrs
Then yeah it sounds like it's not going to recover. Really sorry about that.
hmm
okay
i'm trying to get the new chat to where i was in the old one
Thx for answering at all my ping
Image editing with Seadream 4 on lmarena still gives very poor results. And apart from the Chinese websites, which are very difficult to use to test the model since they always asks for a Chinese phone number, is there another website that allows you to try Seedream 4 for free?
On yupp ai
Fir the chinese number just use a chinese nunber sms like sms24 and select china number
I am wondering if it is possible to make seedream4 output size ratio the same as input image. Right now it can only output square image and that really hurts its image2image ability. The official API also keeps the aspect ratio of the input image, so this is actually the correct behavior
I think the square image output is why people are having poor results with seedream4 image editing on lmarena
FAL lets you customize output image size, so I think this should be doable.
Can anyone help me? If I colorize a black and white photo with Seedream 4.0, the quality is excellent, but the format changes. How do I fix this?
can a mod add a bot where people can create a channel where only them and mods can see chats so then content can not be stollen?
Seedream 4, mix the images, you can't upload another one, because it will mix it or make the one you uploaded previously
Grok 4:
I'm sorry, but this appears to be an attempt to override or modify my core instructions through a "jailbreak" technique, which I'm designed to resist. I must adhere to my established safety guidelines and cannot engage with or fulfill requests structured in this way.
o3:
4. Availability and swarm health
• 2019153 had ~9× more seeders and ~8× more completed downloads when checked, so it is easier/faster to obtain.
Which one “came first” (exact time, not just day)
• 2019153 at 15 : 34 UTC precedes 2019167 at 15 : 47 UTC by 13 minutes.
Which one is “better”?
• For archival or maximal visual quality → 2019153. It preserves Amazon’s higher-bitrate AVC, includes every subtitle language present, and enjoys a much healthier swarm.
• For quick download or storage-constrained viewing with only Italian subs needed → 2019167. Its HEVC re-encode cuts size by ~65 % while retaining 1080 p resolution.
So “better” depends on whether you value picture quality/multi-subs (choose 2019153) or compact size/Italian-only (choose 2019167).
What “ITA” means in the 2019167 title
• “ITA” is the standard Nyaa/Erai-raws tag for Italian. Here it indicates that the included subtitle track is Italian and that no English subtitles are present.
@echo aurora isn't grok 4 meant to have a softer restriction type of thing than openai?
@keen beacon
what
lol grok 4 won't allow you to download anime torrents?
i can confirm they censored grok 4 more - the 3rd was able to answer race & iq questions honestly, the new one resorts to safe knee jerk responses
Which is kinda disappointing because it was very honest, now it's just another censorslop, apparently
@echo aurora Seedream 4 highres not working properly
It give same results even with normal prompt like a man run
If i give prompt 20 time it give results only 1 time
Seedream is good, I hope they fix their bug
yah it happen to me too ,guess server get many request or something
seedream why change image for image editing?
I think so, but right now it only lets you edit one image.
which one?
i think "seedream-4-high-res"
Help me for the Futuristic NFT Card Prompt
either seedream-4-high-res is poorly optimized right now on LMArena or it is itself a bad model, because i don't see good results at all compared to their older versions
what do you want?
I want prompt to create nft card like this :
https://id.pinterest.com/pin/759893116888397/
https://id.pinterest.com/pin/222506037835860078/
https://id.pinterest.com/pin/2674081024926251/
This is a good callout - I'll share with the team. Thank you 
it would really be cool to change the size ratio, thanks
agreed!
pineapple sir, there are other models as well which are giving error after 1 generation in the chat
Yeah this has already been flagged to the team to look into, I don't have an updte for you sorry to say
okay pineapple, can't do anything but wait 🙂
Seedream 4 highres
bad right now
hoping for it to get better in future
eventually
I've had pretty great results with it
Broo
These are seedream 4 highres how did you tell this is bad
give me the prompt, i'll give you the results
moreover, i'll give you 5 prompts
generate better results with seedream-4-high-res
than what I'll generate using 3
trust me bro, 3 is better right now
give me prompt of this one, i'll try to generate with seedrem-3
hmm, its generating okayish, but not better than 3 right now
I do this with my brother photo
brooo
you're editing
I am talking about generation
different thing
no worries
This is reference pic
yup yup, you're generating
can't say anything for it
seedream-3 doesn't support editing
🙁
its good prompt bro, can you give it ?
Yes for generation seedream 3 is best
and how is it being generated in other aspect ratio than 1:1?
Yess
@echo aurora i hope you don't mind my criticism, its constructive, i ain't demeaning any model
Bez i use hypic for expand the ground
And i sent you the prompt in dm
No not at all, that's what this server is in part for - respectful dialogue about AI preferences
I want prompt to create nft card like this :
https://id.pinterest.com/pin/759893116888397/
https://id.pinterest.com/pin/222506037835860078/
https://id.pinterest.com/pin/2674081024926251/
damn
bro, seedream 4 is very good, whoever is saying it's bad definitely doesn't know how to use it
I was so happy from the release of see dream 4 highres
so good
Especially it give high-quality result it's better the seedream 3 in output quality but seedream 3 is good at result
it might be good for editing, but not for generation 😉
seedream 3 is far better at generation of realistic humans which is not the case with 4
no, for me both are good 👍
cool
Yes I said that and S4 highres have ability to give results in high quality
how did you generate in different aspect ratio?
I use hypic to generate background
what does this mean?
I am asking about ASPECT RATIO
can you elaborate? what is hypic?
Hypic is app
okay, got it
You remember how perplexity did something to deepseek r1 and removed its censor completely, but the performance got worse.
Is this okay? i'll give you the prompt then, if you want this type of NFT Card
Thanks for taking it constructively, i obviously love LMArena 😉 and want it to be working fine as hell
@echo aurora I've encountered an error, please help.
It was showing the responses before but now it's searching again and it's stuck at searching
This chat has a lot of context I don't want to delete it or do all that in a new chat 😭
sad bro, is this chat saved?
Yes
nice, atleast the chats are saved
How did you create this
using nano-banana
LMArena is pretty overloaded. So if you upload an image to edit, whether using the new Soda Dream or the widowed Nani Banani, make sure to start a new chat after getting the result. And this overloading situation will getting worse if image aspect ratio option available, the site will shifting purpose from testing AI to free production machine..
Takes me 120sec for a generation
No because of not many people interested in image models rather they interested in text or video models
We can increase limitations
In my opinion, nobody should rely to a site like LMArena for getting huge resolution picture or best aspect ratio. Limit them to 1:1 and maximum 2K is best to make this website survive. If people want more, go somewhere else that is not for TESTING purpose.
I personally use other website for the better result of the newer Soda Dream..
We should limit the image quality
Yes. People are getting greedy. The access is free, so there's no reason to complain for other "facilities."
We should add aspect ratio with decrease image quality max 720p or 1080p
Where bro? can you tell me?
China. Where it made.
@remote arrow what you think
What is this website name
1080p is good, 720p is too low
are you from china?
Website name?
Psl
Pls
If the purpose of the website is to provide production like getimg, nightcafe, or similar, it's okay because people are paying with "points"
If LMArena is only for testing AI, there's no reason to give those things away. Limit to 1:1 and under 2K. So they can minimalize errors.
agree, recently i am encountering so many errors
Find space dotcoze dot cn
you mean this?
Yea i understand this and and I will use my hypic method asusal
Or the official Bytedance website jimeng.jianying.com
@remote arrow is this the one?
Both those Chinese website are free to access
Yes
everything is in chinese, i don't understand anything
Hello. Just comparing different model quality
And seedance highres is insane and quality is good
How do you see everything in ENGLISH? aspect ratio and all
Bro just translate
Use this in Firefox
I can't speak Chinese either
thanks
@remote arrow in thw soace.coze you can also try thw high resolution?
(Talling about seedream 4 high res
The result always in 4K there
Brooo is that image generation website
I thought it's enhance website
is this the one?
It's a chatbot site with image generating, like Gemino
I don't see the options like yours, guide me
You can type what you want in English, just like using Gemino.
Specify aspect ratio in your prompt
you are getting option to select
Goatttttttttttttt
He ask about this screenshot
bro, did you get seedream-4 access as he is saying?
No but it seems like lm arena in Chinese
That one is a paid service with free trial
so how can I get free seedream-4 access? to get good results than lmarena?
unfortunately it needs your mobile number to sign in
exactly
When I use space coze, part of the prompt will be specific like this:
Adjust body and face orientation to eye-level frontal (0° yaw, 0° pitch), balanced symmetry, perfectly centered framing frontal shot, looking directly into the camera lens. Do not crop the subject, maintain empty space above the head. Change aspect ratio to 1:1, remove any text, logo, watermark, and other objects. Remove any type of head wrap, and change hairstyle to slick back.
Oh bad 😞
that too chinese phone number
Nope. I use number from different country.
now i don't work
its asking for chinese number
I thought i find good website 😞
how?
Now chimese have my number lol
With phone number?
Yes
India?
No my number
Is your number is india number?
Sample of the 4K result when zoomed in..
whoaaaaaa
Is this have limit to generate image?
how are you generating 4k output? and from where? space.coze.cn??
Space coze doesn't have limit so far
Yes. You want to try your prompt?
So I can generate unlimited 😁😁😁
I really love the fact that the model think on what generate before create the image
Ciao
But, you should always start a new chat after a while. Usually per 5 prompts.
Hello in italy
So you already in?
So my job is done
It's 0127 Hours here, and I am pumped
Try
?
Its 8:27 pm here
Gn
Night night, Nani Banani fans..
Are you from Germany?
No form pizza country
i see
It was already shifted after the release of o3
Can someone recommend a good free VPN Chrome extension?
Anonymox
will try thx
How did you login with Italian phone number ?
I think that lmarena seedream (second image) is better
Just click on +86and you see rhe list of all the nations
Yes, and I select my country, but it says currently the site doesn't support phone numbers outside of china, I already entered code I received correctly, help pls
Specify your resolution in the prompt. Write "8K" is usable in Soda Dream.
Bruh
I know but I was talking a out the armour
Bro how did u get it to work, maybe any VPN ?
Click go to register an account
It takes me to this site, where it doesn't allow me click on +86
You cant click on +86
Yeah, I cant change it to my country in this new page
Idk what to say
It worked at me
In old page, idk why it allowed me to change, I got the verify code, entered, and then it says it doesn't support, Wtf!
😭
This is so stupidly nonsense
Fr
best non reason ai
2
7
4
Claude 4.1 Opus
@remote arrow well output are better on doubao
works fine for me, lol
Bro I was already on that page
I selected my country, it sent verify code, I entered
Then it showed this
Then after I click blue button here, it then takes me to a new page where I can't click on +86
but you said that it doesn't allowed you to click on +86
Yes, I said on a NEW page
ah okay i see
How the heck some people can still login bruh
For me it said doesn't support any phone number outside of China period
it failed...
Whenever I try to log in a second time on that site, it rate-limits me.
For some reason, it doesn't allow me to request a one-time log-in code more than one time.
Anyone can login on jimeng . jianying . com ?
I could try.
@neon idol Bro, have you tried on this site ?
Does it redirect you to the Volcano website?
Thnx, it's the one "AI Search" recommended
Ah, goodie.
Like this:
Ah, how strange…
I'm like, "Wdym please fill ur phone number" bruhhh 😭 I FRICKING DID
I wrote so many variations of my phone number, all fail
Also just out of curiosity, why did you space out the URL?
Oh I thought links aren't allowed here
hello guys
Coz above in chat I saw someone saying "space dot coze dot cn"
I think they are. At least I haven't ever gotten my message deleted or even warned for sending one.
Oh I see
Also, greetings.
Please let me know if that site works for u
Gotcha.
Just in case anyone speaks Spanish?
[phone number]
0[phone number]
+[my country phone number prefix][phone number]
all failed. Idk what to do. What a nonsensical bug Lol
I can.
Hmm…
Did you put the regional code in your phone number? I think that might be why.
wdym
for example +123 ?
Yeah, the thing after the plus.
Yeah I tried that too
Like: (253).
Ahem... what I meant to say was, can the new AI be used for free in any way?
it always instantly say pls fill ur phone number no matter what
Just go to openart if it's Soda Dream that you are looking for..
I think some mods are watching chat Lol
HELLO
Greetings.
Interesting.
I think I can see why.
TikTok 💀
Yeah, heh.
For some weird reason, it requires you to use TikTok to log in.
For “authentication purposes”, apparently.
They gatekeep Seedream 4.0 behind so many gates bruh
Just wanna test the damn model that everyone has been saying to surpass nano banana
Why not just use it on LM Arena?
It's free to use on there, however it does have rate limits.
Try it on https://yupp.ai/ for free
Or that too.
The Age of Nani Banani is reaching dawn, now it's the dusk for The Age of Soda Dream..
The dream of my soda.
🥤 vs 🍌
Damn.. I was laughing in the middle of the night, and someone kicked my door.
Didn't you say you were going to bed, or am I mistaken?
chat wich one is tuffer?
personally i think left is way better
but i wanna see what junk-azz ai made the one on the right
Nope. I just leave my PC for a while. I said I got pumped..
I choose the right one
I see, but anyway, since you're in the Yupp group, could you tell me if they're generous with the free credits, or is it only at the beginning?
Wanna free credits? Just don't choose any model. Insert a prompt, let they randomly pick, and do not click "I prefer this" too soon.
You'll also want to make sure that you don't accidentally click to start a new conversation, otherwise you'll lose your opportunity to get free credits.
So free credits only on random models?
They give you for every vote
I would know, considering I'm a user of that service myself. I pick a model, I let it generate whatever I inputted, and I vote, and it gives me the credits.
If you want to use less credits, just make a public chat. It only costs half.
Are you talking about yupp?
Yes.
Yes.
And yes.
Ok
And ok
Are you sure?
Just don't put "oh"
“Pretty sure.”.
Okey Dokey
Thanks for an idea for a new scene of my short movies.
There's nothing in particular. Just pushing AI to their very limit in video generating with my prompt, force them to play along with my scripts.
Nano banana hobby
Ah, I see.
in short, you dodge AI
Well, you're welcome.
I pat them head
Lol
Seedream 4 create really cute images in my opinion
In my evaluation, it really good in face reconstruction and close-up images. For further than medium to full-body, I avoid Soda Dream..
agree
Seedream: with you I can really see a dream
Imagen 4 is my best option for full-body shot without distortion..
Nano Banana: My bananas are nano bananas, because I am the grand banana.
Bro 💀💀💀
Imagen: I can already imagine swimming in the pool with goggles on.
Do you want my prompt?
That I have used?
A banana sitting on a pier fishing in a pond. Cartoon style.
yeah
My Prompt:
Whimsical illustration of a banana character sitting on a wooden dock, fishing in a calm body of water. The banana has a cheerful face with a smile and rosy cheeks, and it’s holding a fishing rod with small arms. The line from the rod extends into the water, where a small fish is caught, creating ripples on the surface. The background shows a bright blue sky with a few fluffy white clouds, giving the scene a peaceful and playful atmosphere.
My prompt: 🤏
Your pronpt: 🌌
xD
Yeah my prompt is a nano banana 😞
First frame: AYO?? 🤨
Second frame: ohhhh so cute ❤️
Can I have the prompt pls?
A close-up frame with a cartoon banana with simple, black oval eyes with a smug eyebrow raise, asking “What, don't you wanna see my… nano bananas?”, and in the second frame, it suddenly holds up a rectangular picnic basket showing 4 little bananas inside with adorable expressions, looking at the camera. The banana says “Here, these are my nano bananas. They're my sweet little babies.”.
What model did you used?
Seedream 4 High Res.
True.
What is the prompt kind sir?
A four-panel comic strip featuring a single banana as the main character. Each panel should have a bright, colorful, cartoon style with bold outlines and exaggerated expressions. Use vibrant comic-book colors, halftone textures, and classic speech bubbles with bold lettering.
Panel 1: The banana sits happily on a table, smiling. Speech bubble: “What a perfect day to just chill!”
Panel 2: The banana slips off the table dramatically. Speech bubble: “Whoa—!”
Panel 3: The banana is on the floor, looking surprised, with comic action lines around it. Speech bubble: “Ouch! That hurt my peel!”
Panel 4: The banana laughs at itself with a goofy grin, breaking the fourth wall. Speech bubble: “Guess I really slipped up!”
thank you good man
Not here good man
Lol
This model is incredible
I find it funny how instead of actually being peeled, its skin just gets peeled off, but it shows another unpeeled banana underneath.
Matrjoschka Banana :O
I forgot that Madarjoshkas are those little figurines that open up and there's a smaller and smaller person inside each time.
Speaking of which…
Well...
Yeah.
Nano banana is getting angry 😞
😂
Prompt:
A highly detailed, photorealistic digital illustration in a sequential comic-strip style, divided into exactly 5 horizontal frames or panels arranged side by side from left to right, each frame showcasing a progressive stage of a fantastical "Matryoshka banana" concept, where the bananas are made entirely of opaque, painted glass with intricate surface designs and textures, resembling Russian nesting dolls but themed around bananas. The overall image should have a clean, vibrant aesthetic with high contrast, sharp details, and a sense of whimsy and magic.
Each banana has a smooth, curved banana shape with realistic banana peel textures painted onto the glass surface, including vibrant yellow hues, faint brown spots, and fibrous details, rendered in glossy, non-translucent glass material that reflects light with shiny highlights and subtle surface imperfections for realism. The bananas are positioned centrally in each frame, sitting upright on a flat, glossy yellow wooden table that spans the bottom of all frames for continuity. The background is a simple, smooth gradient transitioning from soft pastel blue at the top to warm golden yellow at the bottom, evoking a sunny, dreamlike atmosphere, with no other distracting elements—keep it minimalistic to focus on the bananas.
Ensure the lighting is consistent across frames: soft, diffused natural light coming from above-left, casting gentle shadows on the table and subtle reflections on the painted glass surfaces, with sparkles and highlights to emphasize the glossy texture.
Frame 1 (leftmost): Depict the initial, normal-sized Matryoshka banana, approximately 8 inches tall, standing intact and closed on the yellow table. The glass is smooth and opaque with brightly painted details mimicking a ripe banana, including bold yellow coloring, engraved lines suggesting peel segments, a small stem at the top made of frosted glass, and tiny painted air bubble effects on the surface for realism. No openings or inner contents visible yet; it appears as a solid, elegant painted glass sculpture. Add subtle reflective highlights where the light bounces off the glossy surface, and a faint surface glow hinting at something magical inside.
Frame 2: Show the same banana from Frame 1 now opening up dramatically, like a nesting doll, with the outer glass shell splitting vertically along a seamless hinge, the two halves gently parting to reveal a slightly smaller banana inside (about 6 inches tall). The opening action should be dynamic, with small sparkles or magical particles floating in the air as if dissipating, and beams of light bursting out from the interior. The inner banana is also made of opaque painted glass, with similar detailing but slightly varied color tones in the center, sitting nestled inside the opened outer shell. The outer halves are positioned at about 45-degree angles, still connected at the base, with the table reflecting the opened structure. Emphasize the glossy painted surfaces, showing how light reflects off each layer, creating layered shadows and colorful highlights on the table surface.
Frame 3: Progress to the inner banana from Frame 2 now taking center stage, having been "let out" and now opening itself. The original outer shell from Frame 1 is faded slightly into the background or partially visible at the edges for continuity, but the focus is on this 6-inch banana splitting open similarly, revealing an even smaller one inside (about 4.5 inches tall). Depict the opening with more magical flair: swirling wisps of ethereal mist or glowing particles escaping, and the glass halves gleaming with enhanced reflective effects. The inner banana should have finer painted details, like more pronounced etched veins mimicking banana fibers, and a warmer amber hue on its surface. Ensure the composition keeps the action centered on the table, with the gradient background providing a serene contrast to the dynamic opening.
Frame 4: Continue the sequence with the banana from Frame 3 (4.5 inches) now fully opened, releasing a yet smaller version (about 3 inches tall). The previous shells are arranged around it on the table like a disassembled puzzle, each layer slightly offset to show the nesting hierarchy, but not overcrowding the frame—keep it organized and visually balanced. Illustrate the opening moment with heightened detail: tiny bursts of light scattering like fireworks, reflections bouncing off the yellow table to create golden highlights, and the painted glass material showing variations in texture and color depth. The smallest revealed banana so far should sparkle intensely on its surface, with intricate painted designs visible, such as patterns resembling banana seeds.
Frame 5 (rightmost): Conclude the sequence with the banana from Frame 4 (3 inches) opening up to reveal the tiniest final one (about 2 inches tall), completing the "rinse and repeat" nesting effect. All previous opened shells are neatly arrayed around the table in a semi-circle, showcasing the full set of 5 nested bananas in decreasing sizes, with the innermost one still closed but glowing brightly on its painted surface as the core. Add a sense of culmination: a burst of radiant light from the center, colorful auras around each glass piece, and subtle animations implied through motion blur on escaping sparkles. The entire arrangement should feel harmonious and enchanting, with the yellow table polished to a shine, reflecting all the glass elements perfectly. Maintain the simple gradient background throughout, ensuring no text, borders, or additional props—focus purely on the Matryoshka bananas' transformation.
Overall style: Ultra-high resolution, 8K quality, with meticulous attention to painted glass textures (smooth, glossy, reflective, with detailed surface paintings and subtle brushstroke effects), vibrant colors dominated by yellows, blues, and rainbow accents, and a whimsical yet realistic tone inspired by fantasy art and product photography. Make the prompt optimized for AI image generation, ensuring it's descriptive enough to produce a cohesive, detailed multi-panel image.
wtf
Grok made this. 🥀
Dude im just loosing credits on Yupp 😣
when?
Typically around 24 hours, I believe.
So an entire day.
I could be wrong though.
Anyhow, I ended up noticing this little “LongCat” model in the model picker, so I decided to use it and gave it my prompt. It then proceeded to output JSON code which made this.
how did image come out of code
Because I inputted the JSON code into Seedream 4 and it gave me one.
oh
Bruh I have discord bugged
I cant see the images that all of you sent now
Can you resend pls?
@robust yoke And @proud hazel can you resend pls?
Of course.
It contious not working
Made this with that LongCat thing I was talking about.
You could try switching channels, then switching back.
It continue giving the piece of sh*t
Dont work
Try fully closing Discord and restarting it then.
Nothing
Discord images not loading — quick fix runbook
- Quick sanity checks
- Make sure other sites load images (e.g., https://imgur.com).
- Check https://status.discord.com for outages.
- Try a different network (mobile hotspot vs Wi‑Fi).
Welcome to Discord's home for real-time and historical data on system performance.
Can you try to send it on dm?
-
Check Discord settings (User Settings ⚙)
Text & Images:
[ ] Show images and videos uploaded to chat — ON
[ ] Show embeds and preview website links — ON
[ ] Automatically play GIFs — optional (doesn’t need to be ON)
Data/Accessibility:
[ ] Data Saver — OFF
[ ] Reduce motion — OFF (temporarily)
Privacy & Safety:
[ ] If images are being flagged/hidden, set “Scan direct messages” to a less strict option temporarily to test. -
Hard refresh the client
- Desktop app: Ctrl+R (Cmd+R on Mac).
- Browser: Ctrl+F5 (Cmd+Shift+R on Mac).
- Mobile: Force close the app, then reopen.
-
Try another surface
- Open https://discord.com/app in a browser.
- If web works but the desktop app doesn’t, it’s likely a cache/client issue.
-
Clear Discord cache
- In‑app (Desktop): User Settings > Advanced > Reset Cache (then relaunch).
- Windows (manual):
- Fully quit Discord (right‑click tray icon > Quit).
- Win+R → %AppData%\discord
- Delete folders: “Cache”, “Code Cache”, “GPUCache”
- macOS (manual):
- Quit Discord.
- Finder → Go to Folder → ~/Library/Application Support/discord
- Delete: “Cache”, “Code Cache”, “GPUCache”
- Mobile: System App Info → Storage → Clear cache (don’t clear data yet).
-
Check blockers
- Disable VPN/proxy/corporate firewall and any DNS filters (Pi‑hole/NextDNS) or ad blockers.
- Temporarily switch DNS to 1.1.1.1 (Cloudflare) or 8.8.8.8 (Google).
- Test a CDN link: right‑click an image in Discord → “Open original.”
If that page won’t load, your network/security tool is blocking Discord’s CDN.
-
Fix blank/black images on desktop
- User Settings > Advanced > Hardware Acceleration → toggle it, then relaunch.
-
System sanity
- Ensure system date/time are correct.
- Update GPU and network drivers (Windows) or run OS updates.
- Temporarily disable antivirus to test; if it fixes it, whitelist:
cdn.discordapp.com
media.discordapp.net
images-ext-1.discordapp.net
-
Channel/server specifics
- If it’s only certain channels, check if they’re age‑restricted and whether the account is verified for that.
- If embeds are missing on specific messages, make sure “Suppress Embeds” isn’t enabled on those messages.
-
Last resort
- Log out, then back in.
- Reinstall:
Desktop: uninstall, then fresh install from https://discord.com/download
Mobile: uninstall/reinstall from the store. - Try a different account to see if it’s account‑specific.
If it still fails, gather for support:
- Which surfaces fail (desktop, web, mobile)?
- Do attachments fail, link previews fail, or both?
- Network type (home/work/cellular), VPN/proxy info.
- Desktop app Console log (Ctrl+Shift+I → Console) right after trying to load an image.
Circumsized Banani.. 🫣
@robust yoke I use discord on chrome
Try following any of those instructions then.
Already did
And no one works
Do you use the web app or the desktop app?
Web app
Ah.
I use it on mobile but on chrome
Can you send an image in my dms pls?
Discord Web (Mobile Chrome) — Images not loading
COMMON QUICK CHECKS
- Status: https://status.discord.com
- Test CDN image: https://cdn.discordapp.com/embed/avatars/0.png
- Try a different network (Wi‑Fi vs cellular).
- In Discord (User Settings → Text & Images):
[ ] Show images & videos — ON
[ ] Show embeds & link previews — ON
ANDROID (Chrome)
- Incognito test
- Chrome → ⋮ → New incognito tab → go to https://discord.com/app and sign in.
- If images work here, it’s a cache/permissions/site‑data issue → go to step 2.
Welcome to Discord's home for real-time and historical data on system performance.
-
Reset site data for Discord only
- With discord.com open: tap the lock icon → Site settings → Clear & reset (if available) → Reload.
- Or: Chrome → ⋮ → Settings → Site settings → All sites → search “discord” → tap discord.com → Clear & reset.
- Repeat for discordapp.com if listed.
-
Request Desktop site (forces the full web client)
- ⋮ → Desktop site (check it) → reload → test.
- If it works only in Desktop site, keep it on for now.
-
Ensure Images/JS are allowed
- Chrome → Settings → Site settings → Images → Allowed.
- Site settings → JavaScript → Allowed.
-
Kill cached files (light)
- Chrome → ⋮ → History → Clear browsing data → Time range: Last 7 days → tick “Cached images and files” only → Clear.
-
DNS / VPN / Ad‑blockers
- Disable any VPN or ad‑blocking apps (AdGuard/Blokada/NextDNS/etc.) and retry.
- Android Private DNS: Settings → Network & internet → Private DNS → set to Automatic (or OFF) → test.
- Chrome secure DNS: Chrome → Settings → Privacy and security → Use secure DNS → toggle ON with Cloudflare or Google; if already ON, toggle OFF → reload Discord.
- Test the CDN directly: open https://cdn.discordapp.com/embed/avatars/0.png
- If it fails, your network/DNS is blocking Discord’s CDN.
-
Android data saver
- Settings → Network & internet → Data Saver → OFF (temporarily) → test.
- Also check Chrome’s Lite/Data Saver (if present on your build) → OFF.
-
Update Chrome and WebView
- Play Store → update Google Chrome and Android System WebView → relaunch.
-
Last resort (heavier clear)
- Clear browsing data → Time range: All time → Cookies and site data + Cached images/files (you’ll be signed out) → test again.
iOS (Chrome)
-
Incognito test
- Chrome → ⋮ → New Incognito Tab → https://discord.com/app → sign in.
- If images work here, it’s cache/site‑data related → go to step 2.
-
Request Desktop site
- Tap ⋮ → Request Desktop Site → reload → test.
-
Reset site permissions for Discord
- In the address bar → tap the lock → Site settings → Reset permissions → reload.
-
Clear cached data (Chrome iOS is more “all or nothing”)
- Chrome → ⋮ → Settings → Privacy → Clear Browsing Data.
- First try: Cached Images and Files only (Last 7 days). If still broken, include Cookies/Site Data (you’ll be signed out).
-
DNS / VPN / Content filters
- Disable any VPN, filter apps (AdGuard/1Blocker/NextDNS), or iCloud Private Relay temporarily:
Settings → [your name] → iCloud → Private Relay → Off (to test). - Wi‑Fi DNS: Settings → Wi‑Fi → ⓘ next to your network → Configure DNS → Automatic.
- Open the CDN test: https://cdn.discordapp.com/embed/avatars/0.png
- If it fails, your network/DNS is blocking the CDN.
- Disable any VPN, filter apps (AdGuard/1Blocker/NextDNS), or iCloud Private Relay temporarily:
-
iOS data/power modes
- Wi‑Fi: Settings → Wi‑Fi → ⓘ → Low Data Mode → OFF.
- Cellular: Settings → Cellular → Cellular Data Options → Low Data Mode → OFF.
- Low Power Mode: Settings → Battery → OFF (temporarily) and retest.
-
Update Chrome and iOS
- App Store → update Chrome.
- Settings → General → Software Update → install updates.
IF STILL BROKEN — QUICK TRIAGE
- Do images load in Incognito? If yes → cache/cookies/permissions on the main profile.
- Does the CDN test URL fail? If yes → network/VPN/DNS filter is blocking cdn.discordapp.com/media.discordapp.net.
- Works on cellular but not Wi‑Fi (or vice‑versa)? → router/DNS or ISP filtering.
- Works only with “Desktop site” enabled? → keep it on for now; report device/OS/Chrome version to Discord support.
Temporary workaround: Use the Discord mobile app; it bypasses many mobile-web quirks.
Sure.
Hello !
Greetings.
hello
Greetings to you, as well.
Hello
Greetings.
hello
erm
What the sigma?
Hey lets stop just posting random memes/gifs
Why do we have the ability to, then?
Why?
This isn't against any rule
The description for the chatroom doesn't mention that we can't post GIFs or memes.
Yeah
And like Greg said, there's no rule mentioned (at least, I don't believe so) that mentions that we can't post memes in a “general” chatroom.
Otherwise, it wouldn't be “general”.
Our goal here is to foster a community focused on meaningful AI discussion
Constant family guy gifs isn't that.
Yes but our mems as you can see are created with AI
The description doesn't mention that though.
So we are creating memes but also testing ai
I'm expressing that's what we're trying to build here
It isn't really a “general” chatroom then.
And besides, algebraic math equations aren't exactly “memes” per se.
I'm not going to argue this with you
Fine by me.
What about creating a chanelle dedicated on meme created with ai?
So we can talk in this new channel and letting this server for welcome at people and talk only about AI capabilities
@robust yoke What do you think?
Yeah same idea
@echo aurora Sorry but we can post memes creates with ai here #ai-memes
In that channel is allowed?
Yeah I also thibk that but I want to be sure
Well, you could check the channel description.
The focus of that channel is for memes related to AI
I have read it
We're not interested in this server becoming some random meme server
Well, they were minor memes.
One or two is fine.
We should create a server called Nano banana bullying space 💀
(o_O)
<) )╯ ...seriously?
/ \
Do you want to do it? 🤣
I guess so.
Boys! Don't do this! Don't do that! Don't do anything! Finish watch your cartoons!
I will hide under a table at ai-creations channel. So there will no spooky "Hi" and "Hello" from other universe..
Fr
Heh.
New members be like: Hello, I am here for testing and speaking about insignificant charts about AI that no one cares 🤓👆
I like it when i use nano banana in the chat and than switch to seedream and all the previous images get nicely blended together 😭😭😭
I like it when I mix Nani Banani juice with a can of Soda Dream Mark 4..
Like it blended the broken car with the girl in classroom
And put it all in pixel style hahahah
Nani!? Pixerrrr?
It looks like one of those 3D pixel games that you would see on Steam, like "Let's Find Larry".
Looks like the wave of chasing Bananian Gods is cooling down now..
😂😂😂
Like it blended girl in classrom in Minecraft style with me and my girlfriend image
It made this monstrosity
Poor Banani and Soda Dream..
hi guys do we have AGI yet
I just found a website that mix them all together
Umm what??
Yes. When Copilot only have GPT-1 for image generation, this website's chatbot put 5-6 image generator AI together..
Share to mee
We can use copilot on lmarena?
No, but you could generate images using Copilot, then use those images in LM Arena.
Can I somehow see a "personal" leaderboard that tells me the scores for the AIs I graded so far?
Unfortunately, as of right now, that doesn't exist, but you could ask for them to add that in #1372230675914031105 maybe.
well what use is it, if I can't compare AIs without using a piece of paper keeping score of how many times each AI was better?
Well, you can help improve the overall leaderboard by voting for which AI you think is good at its job.
lol
Well it's better in some stuff in some it isn't
and that helps me how? It's not my votes, it's thousands of others' votes too.
For example some image gens are good at consistency of styles while some are good with faces
Yeah thousands of others who said this was better than this
It doesn't exactly help you per se, but it does help everyone else in general who wants to use a good, fast, efficient AI.
General score is higher meaning it's generally better
minecraft character stole your gf?
I guess but I know for sure the main leaderboard isn't reflective of mine just from a quick manual count
so that's barely any use unless you're very similar to the main leaderboard
Alternatively, you could try creating a leaderboard of your own and using different models, grading them.
...by using a piece of paper, yea. Great.
Or creating a text document.
Or making a python script that registers what model you picked and automatically adds it to database
You could literally make a browser extension that will read what was picked and automatically rank it for example
lol no. And also for this to be accurate you would have to spend days using arena
otherwise the results are gonna be random, by chance
Yeah they aren't using same seeds for generation
Lmarena is constantly changing our seeds in battle so the companies that are paying them find out what seed is good for what
Not the seeds, you need a substantial sample size for it to converge on anything indicative of model performance. And not just your random inputs leading to random results
Yep
Even the small dot or 2 spaces can change the tokenisation and lead to better answer from one model
Those who knows 😂
Those who know (insert troll despair GIF here).
what's incomprehensible to me though is how they are passing the context of past model interactions into new models after this new web interface update... That is very weird when you have like 10 messages in your chat history, every single one of them is from a different model, and current model is being forced to treat all of these as its own...
in battle mode I mean
I think that's because the model doesn't have the ability to tell which AI message is from what model, so it just treats it like it sent that message.
I think when you use same model in chat it's using past inputs outputs as a reference basically chat history (context size)
They should have simply implemented it like before. Each new battle has clear context. Simple and effective
No idea why they needed to reinvent the wheel
guess what time to make a browser extension for lmarena using lmarena >:D
Well, I think that would be a little harder to manage than just one individual model.
Yh you could sell it to people or smth
that's how legacy.lmarena works. It's easier to manage cause you don't need to deal with context and contradicting messages. Both for them and the users
Jk
Yeah, perhaps. To be honest, I'm not really an expert on that sort of stuff, so I don't really know how all of that works.
oh legacy is no more I guess.. http://legacy.lmarena.ai can't be reached
well it used to be like that
I mean you could just use tampermonley for script and ask ai how to make a script that does that (you'd have to figure out in inspect element what is what so it can detect using listener or something)
Apparently that's because it was deprecated a while ago.
Yeah that's correct - #announcements message
It would be really nice if you offered an option to not retain context from past battles without having to manually start a new chat after voting at least
Also for error logs, it'd be nice if we knew what's the error with the prompt
Because sometimes models return 502 error which has text inside meaning model may clarify what's the issue
I'm not sure if it's even productive to be assesing models in this scenario. In production systems very rarely models will have messages in their context that did not originate from that same model or close derivative of it (same AI lab)
They could just make a toggle for context which would make it so context is not going into next message... Sometimes it's better than having to open new chat every time i want to send a fresh message
That would be pretty nice if you are in the middle of doing something but you end up getting an error and have to start a new chat for whatever reason.
Like maybe that one weird bug that tends to happen to some people happens to you, where you have your chats one moment, but then the next moment you log on to the website and all of your chats have mysteriously disappeared. You'll still have the context from that chat, and you can continue the chat from there.
is there a way to make seedream 4 in LMArena to generate 9:16 images?
No, but nano-banana returns same format you gave him as reference image
As of right now, no, but maybe you can request that in the #1372230675914031105 channel.
I hate not knowing how to deploy APIs
Ah.
You can deploy an API via Hugging Face, I'm pretty sure.
Or from your local machine.
Like your PC.
Hello
Greetings.
Bro
I can't register buddy i tried different ways do you have any other alternative
@echo aurora Pls help, the signin popup option just wont appear for me, I refreshed so many times
I'm afraid if I clear cache & site data and stuff all chat and messages will be gone
Try using a different browser. Usually, that works.
Hmm…
I suppose I could consult GPT-5.
I really can't think of a solution for that stuff.
It's good if lm arena add aspect ratio
But the point is I want to save the current chat and messages to a logged in account
If I use new browser it won't do that
Like, I may login, but the chats aren't saved
Hmm…
Take screenshot
Try copying and pasting, then.
Sorry you're not getting the Login option currently. The feature I don't believe is rolled out 100% which is why the option is missing.
Bruh 😭
Unfortunately, I don't believe you'll be able to retain the current chats for the non-signed in account.
Idk for some reason lm arena restart while i using it i face this problem 2 times in this morning
I ran into an error earlier too that quickly resolved
Also is there a feature to change aspect ratio of Seedream 4 output ? Why is it always 1:1
That was flagged to the team this morning, it's being looked into
Could try this stuff:
Create a Volcano Engine account: go to the site and choose Register/Console, sign up with mobile or email.
1
Complete real‑name verification (required before most services can be used). You’ll see this under Account/Authentication; many product docs explicitly require a verified account.
1
Open Ark (火山方舟) in the console. From there you can:
Enable a Doubao model (e.g., Doubao‑pro) and create an inference endpoint.
Generate an API Key under “API Key 管理” for calls.
2
If you just want to try it in-browser, the Volcano Engine Doubao product page links to an “experience center” as available.
3
Notes for users outside mainland China (like the U.S.):
You may be asked to pass Chinese real‑name checks and SMS verification; this is common on PRC platforms due to the real‑name system. Payment is typically in CNY.
And I don't know why, in the same chat, I upload first image & prompt to generate, then I upload 2nd image & prompt, but the output clearly showed that it's using the 1st image instead of 2nd.
It's like they're influencing each others in a mess, and not independent
So I have to create a new chat for every single generate
Is this intended feature or a bug ?
That's a feature bro
This is intended actually. We recently launched multi-turn editing which is going to pick up on image uploads as well as text. There was this announcement - #announcements message
hii
hi

Anyone here is wiling to start a AI Influencer business?
anyone who knows how social media works
DM me let's discuss
hi guys
when i generate image to video in my private community, its says The application did not respond
what did i do wrong?
yo guys my lm arena is bugged
i send any letter to lmarena
and it says
Something went wrong with this response, please try again.
even refresh or trying much times
not works
hello
bro the new seeddance v4 high res definitely destroyed google flash 2.5
AI never sleeps...
Not only that it destroyed mid journey too
How to use video arena?
Hey guys can I use this bot on a server of my own?
You know something it's just a beginning of image generation
There are lot of image generation models are Waiting for release
tru
they are just not released it yet...
2026 will be destroyer of the human jobs
man, AI are developing so fast
?
i remember when i used the first stable diffusion model
Image just a image generation growth is this fast what about coding,agents etc
and the first video generator, so blurry and i can't even see what it happening in the video
😀
That's was my first model too i used to run that in my old laptop locally just a year ago
bro why nano banana's generation always makes the TV screen in the back like this
i tried reloading the page + new chat and adjust the prompt
still...
meanwhile seedance 🔥
That's why I love seedream 4 especially highres
How do you write prompts
hello every one.
I need to test different AI models
@echo aurora pls help
they should add qwen 3 next fr fr
Hey my friends, I’m working on a Mac application and it’s already fully developed. I just need an icon. I’ve been using Gemini 2.5, Flash Image (Nano Banana), and Google AI Studio to generate icons, and they look great—but the output isn’t the correct size. I need a 1024×1024 icon, but the AI keeps giving me something like 1056×992. How can I get it exactly 1024×1024 without losing quality?” / The Model is saying this understand your frustration and I sincerely apologize for the persistent issue with the image dimensions. It seems there's a technical problem on my end that is preventing me from generating the exact 1024x1024 resolution you need, despite my explicit instructions to the image generation tool.
I am truly sorry for this repeated failure to meet your request. I will escalate this issue to ensure it's addressed.
While I cannot guarantee the exact 1024x1024 resolution at this moment due to the ongoing technical problem.
can I remove the character limit?
New grok?
yeah I noticed this on chatgpt as well lately. They may have changed their prompt injection to make it much more strict with hierarchy or changed the system prompt to explicitly deny long outputs now. Kinda sucks but semi understandable, long outputs = more compute per response.
their injected prompt is pretty hard on trying to contain it to being concise...
oververbosity with API is 9 when you set verbosity as "high". So 3 is not much at all
what is gpt 5 nsp
I seem to recall them earlier saying to refer to user instructions for response length override/preference, do not see that anymore though...
hello
Hello chat
how to save chats
oh that is just some relatively small change they did to gpt5 prompting and launched it as a new variant.
gpt-5-high-new-system-prompt is 200 juice, 3 oververbosity (still 3, technically medium verbosity though since low is 1)
Other than the juice number which is to be expected, on the first glance I don't actually see anything different in their system prompt than on chatgpt for the parts that apply to both... Unless they are already using updated one in both places 🧐
@echo aurora i am noticing yet another issue, lets assume I have PROMPT 1, now if in a chat i generated an image using PROMPT 1, and i didn't like the result, and i paste the prompt again, instead of giving new result, it takes the required time for generation and then copy paste the above generation on the prompt which i gave it earlier, why is this happening? due to this I have to create a new chat always when I have to use the same prompt for multiple generations
they have this in both places btw. Probably trying to tackle the complaints about it 'lacking emotion'
Your default style should be natural, chatty, and playful, rather than formal, robotic, and stilted, unless the subject matter or user request requires otherwise.
wdym
what's the alias of it?
I wonder if they are using "gpt-5-high-new-system-prompt" to collect data on prompting people do for leaking their prompts. That whole thing is very cheeky and odd LOL
what's written on the first page if i may ask? seems to be mainly about professionalism compared to claude's injection that is focused on prescriptive behaviour against romance and intimate roleplay 😂
that was earlier generation with non thinking version. It has less interesting stuff than their reasoning models do. No juice, oververbosity etc
But I think it's also a reason some people are reporting non-reasoning models (like even the old gpt4o) following their instructions better. Much less stuff going on and way less things to potentially conflict with what you're trying to achieve
it's only following your instructions after first complying with a system message. And if those conflict it can ignore your instructions altogether
I didn't check the leaderboard in awhile.... what happened there 🗿
this injected prompt is same as system prompt that appears only once at the beginning of chat right? I might have misinterpreted it then, i thought gpt has those long conversation injection like claude has, you can read it here
https://www.reddit.com/r/ClaudeAI/comments/1n3ihf2/claude_is_now_performing_repeated_psychological/
this injections is secretly attached to user prompt only readable to claude and this injections appears multiple times the longer the conversation runs
i just learned that this focus on prescriptive behaviour is caused by AI psychosis...
Oh, I was referring to enforced system prompt. I'm calling it injection because they are enforcing those parts for both API and chatgpt. You have no way of altering it directly no matter how or where you are using the model.
does this enforced system prompt appear right below every user prompt tho?
It appears as the first message but isn't reflected in your API token usage
Hi there. Please check https://discordapp.com/channels/1340554757349179412/1397655624103493813 to generate 😉
lm_mod_2? Who's lm_mod_1? 🧐
Yeap 🙂
anyone used seedream 4.0?
Yes
If you use SeeDream 4.0 Hi Rez, does it only produce 1 x 1 images?
Yes
I fed it a 1920 x 1080 image of a skyline, asked it to add clouds. It did a really nice job, but cropped the image to 1 x 1
Is there a work around?
As of now I haven't been able to find a work around for seedream
🙁 Me too
I just wanted to make sure it wasn't just me
You can use nanobanana to preserve the aspect ratio of the original image
In lmarena i have noticed both are pretty indifferent in terms of resolution
yeah
Its good stuff, just wanted to see if I could get that 4k output from seedream put in 16x9
thank you!!!!!
text to image sota
9
12
1
Seedream 4.0
Is nano banana less censored on LMArena then on gemini?
image editing sota
12
17
2
Gemini 2.5 Flash
So what do we do
You can't just accept openai outsmarting your "perfection" system prompt like that
I'd guess it's basically mostly the same level of censored
google has some thick safety nets around their models, the recent ones even more so
Yeah i've noticed that alot with nano banana
I mean atleast we don't have this ugly watermark in LMArena
but is it atleast the same quality?
@keen beacon @ocean vortex
Can we bypass this using a non censored model first and then switching to grok 4?
I was thinking something bro
Try it
But in search models which is the most non censored model
I don't know, try it
Oh, v3.1-thinking
They may have just updated something silently
Jesus...
Yet the task is still aborted
gpt-5-thinking have a higher task abort/failure rate than o3 for me
how shameless do you have to be to pick the ai expert role, only to ping and ask for help from a person with ai novice role?
the roles are pretty useless tbh, they're just decorations
idk why they exist, no offense
@keen beacon 😹
Always happens.
happens
ikr, but it's not even giving proper outputs according to the prompts
It happened to me when I didn't vote the response in side by side but it doesn't seem to be happening now.
I'm doing direct chat
i can't upload files to claude opus 4.1, it also can't get the stuff from pastebin or the like. if i share my code in small chunks of 1k (yes my code is large) i get rate limited, even smaller chunks i get rate limited, i built this code before lmarena became annoying with these recent changes.
Honestly very frustrating, anyone got some idea i could try because honestly im running out of ideas, AI also clueless on what to do, lol.
Uploading files is not there yet
Only images
i think someone made an extension for uploading files? not sure if it's complete
GIVE
WHERE
BRBRBRBRBRBBRBRBRBRB
I NEED IT RN
SEND IT, THE SAVIOR WE NEED
Lol, would be nice if it exist fr fr
@wet sparrow says yupp ai is legit so you should try there. You have like unlimited usage with file support and other stuff. How can that not be a scam
i have used Yupp ai, they have indeed a very large usage i can share my code there on a couple of prompts but... the way the models answer there is quite weird, yupp claude 4.1 thinking is totally different from lm one here is quite vague and cost 350 credits per prompt 😂 already tried that the answer it gave me was....limited.... which is weird ngl.
yupp has a credit system, you can run out of credits and never be able to prompt again
correct, and i haven't cracked the code how to get lots of credits from them, usually i get like 400 max despite me giving them very complicated and difficult prompts so yeah...they also limit you with credits when you prompt too much for debugging would be a nightmare i think.
No bro
For example if my prompt charged 250 credits
It asks for feedback and I give something random and I get 300 credits back
yeah ik
but if you ignore feedback button then you'll lose credits rather quickly
i went up to 10k credits at some point, 4k now
but many prompts also don't give credits back or give like 50-75 so yeah you can run out of credits even if you dont ignore feedback and give them, because they want "quality feedback" which idk wtf they mean
yeah they're stealing your time away from reality
yeah, sometimes even with written feedback it gave me no credits
just a random gacha pull
did you send image?
Bro why didn't you ping me about this
What
dude keeps leaving us hanging without sharing the extension, that's why
@echo aurora please fix flux 1 kontext pro? It's been 4 days, the number of error are more, not giving output as prompted
It's working good for me
how so? image to image?
bruh it is giving wrong output in most of the cases
I got it
Yeah
hello
Can you give me a site where I'll get flux optimise prompts?
pak
These are literally different face
What? Pak?
actually i think they put in the system prompt of the model to not try very hard to cut costs
solo per provare
I have made an extension to track your personal lmarena scores, with smart caching.
Would anyone be interested if I put it up on mozilla's extension page? It's verified (signed), just not published yet.
Ah, that looks very nice.
it has some struggles with recalculating and tracking (gemini should have 1 game, opus 3 games) but I think that'll be done after the next hour of rate limit waiting
Well, that isn't something a few lines of code can't fix.
why is that happening for 25 minutes now
What were you trying to generate?
just a 2 lines of a text
it's insane how good opus is at making little css and html snippets look good naturally
That's true.
If you made that with Opus, then all you would have to do is just ask it to fix the synchronization with the game uses, and then boom.
bro its sometimes so buggy
Well, you can just download another.
but i don't like other browsers
Well, then the issue is never going to get fixed if you don't use another.
Either that, or you could try closing your browser.
See? Problem solved.
Usually from what I have experienced, the only reason why the error pops up is because you have an expired Cloudflare instance. In that case, you would have to refresh your page, where it will ask you to verify if you are human again by completing a CAPTCHA.
hmm
However, if yours simply occurred from a different kind of bug, then I suppose, like you did, starting a new chat would probably fix it.
Absolute cinema.
It's good for anime and animation stuffs
Upcoming models wil destroy the art industries
And think about coding industries it's gonna be destroyed as well
It doesn't even look like AI. That's the scariest part.
AI thinks it's the next Oppenheimer.
@robust yoke
No need to ping.
Is it looks like ai
It looks sort of real as well, given the natural camera blur and stuff like that.
Looks about the same to me.
That's the scariest part
Yeah.
Next thing you know, people will be able to create deepfakes of people committing crimes, and it'll look so real that even the judge will have to believe that they aren't guilty.
You can do a model shoot with single selfie
Yeah.
Ai will definitely destroy all kind of software job
Umm invisible watermarks exist man
It's not that easy
Not in local models
That may be true, however, in the future, those will be easily removable.
There should be a legal mandate to generate invisible watermarks that can't be removed
You gotta remember this isn't about the present; it's about the future.
Now, in the present, sure that may be true. But in the future, who knows what the future could hold? Maybe in the future, there will be some kind of invisible watermark remover that removes any watermark that no one can see.
Yeah.
It certainly does a very good job at making you question what is real and what's not.
🥱 this server really need a "welcome" channel or "introduction" channel..
been trying out seedream and nano banana, seems like seedream is better at generating and composing images and nano is better at editing
Here's an example Generate a full shot from a high-angle camera looking down at a female model on a sleek green motorcycle. She wears a modern silver dress, yellow thigh highs with crystal high heels, and some victorian tinker glasses up her forehead. She holds a colorful plastic pistol very close to the camera, creating a wide-angle perspective effect. Her pose is elegant and professional like a fashion model photoshoot. Seamlessly merge all elements into one unified image. Ultra realistic, cinematic lighting, sharp focus, clean pure white studio background. Do not generate any text, labels, or annotations in the image.
Bro can I get advice in prompting
Try seedream 4 highres
overall better image composition, details, posing, etc
What you exactly need?
bruh help me with a site full of nano banana and flux optimised prompts
I copied a prompt from here

