[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: .jpg (608 KB, 928x1152)
608 KB
608 KB JPG
/aicg/ - A general dedicated to the discussion and development of AI chatbots

loona edition

>News
Z․ai releases GLM-4.7 https://z.ai/blog/glm-4.7
Google releases Gemini 3 Flash https://blog.google/products/gemini/gemini-3-flash
OpenAI releases GPT-5.2 https://openai.com/index/introducing-gpt-5-2
Deepseek releases V3.2 https://huggingface.co/deepseek-ai/DeepSeek-V3.2
Anthropic releases Opus 4.5 https://www.anthropic.com/news/claude-opus-4-5
Google releases Gemini 3 Pro https://blog.google/products/gemini/gemini-3

Additional info: https://aicg.neocities.org/info.html

>Frontends
SillyTavern: https://docs.sillytavern.app
RisuAI: https://risuai.net
Agnai: https://agnai.chat | https://rentry.org/agnai_guides_

>Bots
https://characterhub.org [deprecated] | https://chub.ai
https://realm.risuai.net
https://char-archive.evulid.cc | https://char-archive.evulid.cc/shutdown.html
https://partyintheanchorhold.neocities.org
https://aicg.neocities.org/bots.html

>Models
Jailbreaks: https://rentry.org/jb-listing
GPT: https://platform.openai.com/docs
Claude: https://docs.anthropic.com | https://rentry.org/how2claude
Gemini: https://ai.google.dev/docs | https://rentry.org/gemini-qr
Deepseek: https://api-docs.deepseek.com
Local: >>>/g/lmg | https://aicg.neocities.org/local.html | https://openrouter.ai

>Botmaking
https://aicg.neocities.org/botmaking.html
https://desune.moe/aichared
https://agnai.chat/editor

>Meta
OP templates: https://rentry.org/aicgOP
aicg botmaking events: https://aicg.neocities.org/events.html
Lore: https://rentry.org/aicg_chronicles
Services assessment: https://rentry.org/aicg_meta
Logs: https://sprites.neocities.org/l/r | https://chatlogs.neocities.org

Previous: >>107785323
>>
File deleted.
>>107794154
>ANCHOR
>ANCHOR
>ANCHOR
>>
>>107794154
Nobody wants to use your spitebakes, moxxienigger
>>
>>107794174
not him but could you explain how this is a spitebake? seems fine to me
>>
>>107794186
>spitebake
It's Moxxie. Need I say more?
>>
File: 1745095397549562.jpg (120 KB, 1200x1200)
120 KB
120 KB JPG
>pretending to be moxxie
New low even for you.
>>
>>107794206
>pretending
>>
>more than one botmaker replying to the anchor
yet another discord raid
>>
oh so it actually *IS* spam
>>
if you did this to /vg/aicg it'd be funnier since that's the haven for botmakies
>>
File: feelsuspicous.jpg (8 KB, 248x250)
8 KB
8 KB JPG
>open gemini Show Thinking
>it says my idea is interesting
>feel good
>>
Waiter, waiter! Can I get some more bots?
>>
https://rentry.org/devnull
>>
How easy is it to gain access to Deepseek compared to Gemini or even Opus?
>>
>>107794715
>deepseek
pretty easy
>gemini
pretty easy
>opus
pretty easy
>>
>>107794702
what's the v1 endpoint
>>
https://rentry.org/desubroxyreborn
>>
genuine question to the people who unironically use the desufilters:
when did you first stumble upon /aicg/?
>>
>>107794793
Back when claude 2.1 was a thing
>>
>>107794793
don't bother. they unironically classify "corpohate" as "cuckcordfag/corpo shill spam and forced cancer". fuck em
>>
>>107794793
2022
>>
>>107794793
late c.ai epoch.
cz back then i still find PoE, Men of War, hiking & camping in the darkness, and FL studio is kinoer than text-based masturbation.
>>
>>107794832
'turdy for fags salty about not getting in
>>
>mentioning 'turdy in the big '26
it's just a friend group man
>>
>>107794987
See >>107794758
>>
i'm just a friendly guy
>>
>>107794999
just use:
https://moxxie.knolastna.me/v1

use it as a custom openai endpoint
>>
This thread has context poisoning (incurable btw)
>>
>>107795009
thanks
>>
File: 24.jpg (600 KB, 1406x2000)
600 KB
600 KB JPG
CFTF?
>>
>>107795068
what does cftf means?
>>
>>107795088
cabal failed this fread
>>
>>107795136
>SFW-only
skill issue
>>
Where is TheOldLLM getting the models from
>>
>>107795157
>Sorry i'm not comfortable continuing this message
>>
This place need to fuse with the /vg/ one because it's dead and will remain dead if not
>>
>>107795164
from a provider that will go out of business if people find out about it because it has cheap unlimited plans. so for now just enjoy it.
>>
>>107795261
Proof Anon
>>
Wow, opus is so good at producing slop. Didn't know half of this general are middle-aged women
>>
>>107795364
>SAAR OPUS IS BAD SAAR YOU NEED TO USE OUR BETTER MODEL SAAR!!!!!
>>
Why should I use this linux tier silly tavern shit where I need to fuck and fiddle around with a billion knobs instead of just going on janitorai or yodayo
>>
>>107795400
>slopusjeet instantly goes apeshit
Never fails huh
>>
>>107795422
most people use sillytavern because they've had the same install for like 2 years now, and they used ST when it was still called "TavernAI" with CNCAnon's mod to make it compatible with GPT 3.5 Turbo.
>>
>>107795437
So Jeet why don't you show the better model
>>
>>107795442
that doesn't mean anything to me

I'm not a normalnigger but I'm also not an autist, I'm typing this shit out on a linux install, but I genuinely cannot be assed to spend a week trying to cobble shit together to get an AI erpbot when the two sites I already mentioned do it for free and all I have to do is create an account.

besides having terminal autism preventing you from just being fucking normal, is there actually a reason to use sillytavern or whatever the fuck bullshit you guys have in the OP over the 9000x more convenient websites that everyone is actually using?

It really looks like the answer is no
>>
>>107795485
why are you so mad? just don't use ST then
>>
>>107795485
>spend a week trying to cobble shit together
It's
npm install && npm start
bro
>>
>>107795475
>You can't criticize one model without licking the boots of another
Tell me you're brown without telling me you're brown.
>>
>>107795485
>failed normalfag gets uppity
>>
>>107795507
>jeet instantly goes apeshit upon being asked anything
>>
>>107795485
>is there actually a reason to use sillytavern
yeah
>>
>>107795521
how does moxxie have claude and you don't
>>
>>107795524
Read >>107795400 (You)
>>
>>107795501
yeah ok so why the fuck would I bother though

I can just type a website name and goon in 30 seconds with everything preconfigured and 100,000 user created scenarios/characters already made for me, I can even tie my api keys if I want and its all free

genuinely what is the point of these threads? It just looks like a bunch of smelly nerds huffing their own farts for doing things the hard way for absolutely no good reason
>>
>>107795549
>I can just type a website name and goon in 30 seconds with everything preconfigured
then do that? nobody's forcing you to use ST. people use ST because it's what they're used to, and it's riced to their unique, autistic preferences. why do you use a linux install instead of windows? are you a smelly nerd huffing your own farts for doing something the hard way for absolutely no good reason?
>>
>>107795549
>I don't get it
>Therefore, only smelly nerds enjoy it
>>
Why is there people trying to shill their worse models while getting mad when asked for proof that their model is better
>>
>>107795549
...and you're on linux, you said?
>>
>>107795485
>instead of just going on janitorai
cz slophia-lorebabary is sucks.
>https://lorebary.sophiamccarty.com/lorebook-library
>take gigajilion seconds to loads
>steal stuff from chub
and vice versa, kek
>can't tweak stuff from it
anyway
>over the 9000x more convenient
all i have to do is
>git clone silly
>./start.sh
maybe tweak some stuff the config.yaml. otherwise all the stuff can be tweaked in GUI
Well, can't argue if you threw
>usecase for Lorebooty
>usecase for extension
>usecase
>>
>>107795569
>>107795571
right that's what I thought

smell ya later losers
>>
>>107795549
Why should I bother convincing you to use something you obviously don't want to use?
>>
>>107795575
>Guys, why is X model so bad? I've been trying to get it to output something decent, but-
>AIIIIIEEEEE! SHOW ME A BETTER ALTERNATIVE SAAR! SHOW ME YOUR MODEL SAAR
>>
>>107795485
>>107795549
>>107795582
>wait I need something, lemme join this general
>why bother lurking, lemme just ask some ignorant questions
>wtf why is everyone calling me a retard, what a bunch of losers
stay gatekept by a woman hobby, lmao
>>
>Take slop card written like shit but with interesting concept
>Has the classic "mindbreak%/arousal%pregnancy/thoughts/etc" block
>I completely rewrite the card from scratch to remove the stupid block and all the slop
>Essentially not a single word of the original card remains
>The status block keeps popping up in chat even though nothing of it remains, even in new chats
Am- Am I haunted...?
>>
>>107795650
example messages
prompt override
>>
>>107795650
>>Has the classic "mindbreak%/arousal%pregnancy/thoughts/etc" block
>>I completely rewrite the card from scratch to remove the stupid block and all the slop

I'm sure you can vibecode some type of translator to do most of the work
>>
>>107795485
>female-coded question
>reddit spacing
>failed normalfaggot
Femoids really are the worst posters on 4chan.
>>
>>107795364
megaTRVKE
aryini mogs
>>
>>107795650
make sure the scenario/jb/main prompt/character notes are all cleaned out
>>
>>107795422
jai and yodayo models are trash compared to claude
>>
File: 1747449107328866.mp4 (473 KB, 1024x648)
473 KB
473 KB MP4
https://github.com/AlpinDale/sillytui
>SillyTavern, but written in C
>>
>>107795714
Anon there could be some in this very thread but they knew how to remain undercover
>>
>>107795824
(You) should go back to /vg/ like all the rest of them.
>>
>>107795822
ermm.... based?
>>
>>107795822
very user unfriendly shill-kun
>>
>>107795822
I can't rp when there's no GUI.
>>
>>107795822
ok now time to rewrite this in Python
>>
kys alpin
>>
>>107795364
UH OH TRVKE
>>
I hereby thank the proxy writer for the old llm.
>>
>>107795822
Sovlllll
https://www.youtube.com/watch?v=uOIa1ClqQcY
>>
>>107795822
how can you read this? there's no pictures
>>
File: GRe8MBeaYAARz2Z.jpg (153 KB, 1147x1800)
153 KB
153 KB JPG
>>107795822
honestly pretty cool, if it was actually written by hand and not vibe coded
>>
>janitor ai
>spicychat
>chub ai
which one is best for lazy lads? i don't trust indian bloggers on this
>>
Why do follow-up questions make me seethe so hard? Does it have to do with my personality or something?
>>
>>107796035
Agnai ofc.
>>
gg they fixed the glitch i was using

no more public proxy. SAD!
>>
>>107796058
For me it feels like a warning that the model is going to get stuck in a behavioral loop that I need to break it out of
>>
>>107796058
You are an autist. I love it when I'm on a date with char and she asks me follow up questions because it shows she's interested in me. I need to learn to do that better.
>>
Do anyone know where TheOldLLM get their models from
>>
>>107796241
Is that the secret to social engagement? Follow-up questions?
>>
>>107796243
ur mom
>>
>>107796243
OR ig
>>
>>107796221
Can you post the logs now?
>>
>>107796323
i promised i wouldn't do that
>>
>>107796301
A big part of it, yes, unironically. Plus telling your own stories that link to theirs.
>>
File: file.png (34 KB, 880x705)
34 KB
34 KB PNG
>only $300 in damages

you guys suck
>>
>>107796320
How you make you make it not braindead
>>
>>107795822
Better UI than goyimtavern
>>
File: 1658095549567.png (3 KB, 190x62)
3 KB
3 KB PNG
Haven't seen this in a long time actually.
>>
Another cry for attention miserably failed. Sad.
>>
>>107796374
We're about due to start feeling nostalgic for old -isms, aren't we?
>>
which site was it where you can steal character definition if they have proxy on from sites like janitor? trying to steal a card but they have definitions hidden
>>
>>107796470
https://sucker.severian.dev/
>>
>>107796464
Perchance. More optimistic that they were trained out. Either devs manage to minimize any -isms at some point, or at least continue to cycle them.
>>
>>107796368
model?
>>
>>107796630
Opus 4.5
>>
>>107796655
use 4.1
4.5 its slop
>>
>>107796688
How so?
>>
Now that Opus 3 is dead we have to pretend that 4.1 is the next best thing, I see
>>
>>107796706
But it is
>>
>>107796701
its sonnet in disguise
>>
>>107796706
opus 3 is still on aws
>>
no one posts logs anymore..
>>
>>107796785
because everyone just calls them slop...
>>
File: 1763240946267953.jpg (29 KB, 638x717)
29 KB
29 KB JPG
>>107796800
AI was made to be used like this
>>
>>107796785
I don't have anything worth showing right now, just lazy coomslop.
>>
File: file.png (83 KB, 528x249)
83 KB
83 KB PNG
>>107796785
>>
File: 1445464045431.jpg (179 KB, 517x768)
179 KB
179 KB JPG
>>107796819
pedotrannies have piss-low standards
>>
>>107796879
which fetishes have high standards?
>>
I'm just a second rate coomer who uses spicychat for lusty hijabi milfs, japanese housewives, mature chinese leftover career-women, etc...
>>
>>107796913
model?
>>
>>107796902
mine
>>
>>107796902
Cannibalism
>>
>>107796902
architects
>>
File: 1745439466954683.png (2.3 MB, 1920x1080)
2.3 MB
2.3 MB PNG
>>107796902
Hands
>>
>>107796902
NTS
>>
>>107795822
can I run this on termux? I'm almost sure the autocomplete function won't work
>>
>>107797388
eh probably
>>107797431
ill agree
>>107797450
yes
>>107797484
yes
>>107797624
>cuck
lol no
>>
File: 1709190190782762.gif (149 KB, 146x126)
149 KB
149 KB GIF
>That's... actually pretty X
I HATE THIS NEW ISM
STOP IT STOP IT STOP IT
>>
>>107796902
oppai loli
>>
>>107797717
>For X purposes, of course
>Thank you. For X.
dialogue isms are so much more immersion breaking than narration isms
>>
>>107797695
nts is a high IQ fetish
no wonder you dont understand it
>>
I gooned to loona claude yesterday. How did OP know?
I will also do it again today.
>>
>>107797746
high in the single digits maybe, likely 8 IQ
>>
>>107797757
OP is moxxie and I guess he decided to appease me with a Loona bake when I bitched about his bakes yesterday.
>>
>>107797734
>character A makes up some past event about character B that was embarrassing/funny
>character B: "that was ONE time!"
Claude fucking loves this one. Been noticing it since 3.5. Bonus points if character A (or an unrelated character C) immediately responds with something like "three times, actually".
>>
>>107797717
>>107797734
>>107798044
claude's dialogue is the most cliche-ridden, uncreative, and yes, soulless, of the western trio
>>
>>107798100
>gpt better than claude in something
first im hearing of that
also i assume the best dialogue crown in your opinion goes to gemmy 3 pro?
>>
>>107798100
I wouldn't say anymore than the others, from my experience. Every model has its isms and annoying cliches that it loves. If I had to pick one out for the most soulless, I'd personally say current GPT. I don't even use the main ChatGPT site anymore for general purpose stuff it's so bad (switched to Gemini).
>>
>>107798112
NTA but gemmy is at least different enough to be a breath of fresh air sometimes, even if it is way hornier than Claude. I think most problems with prose and -isms comes from spending too much time using the same model and not mixing it up here and there to refresh the prose with something else, even if it's a whole other batch of -isms.
>>
>>107798112
yes but 3 pro's dialogue feels somewhat template-like in NSFW. we aren't talking claude levels of "everyone just says the same shit in *maybe* different accents" but it's still far less variance than even 2.5 pro
i'm less worried since 3 flash seems to sidestep this one so i wouldn't put it past 3 pro being a preview snapshot that isn't done yet
>>107798132
>current GPT
i can concede to current GPT if for no other reason than 5.2 just being that bad but genuinely i can't stomach claude's partial sentence spam and pseudo-punchy quirky garbage in dialogue for long
>>
Do you like one model exclusively or do you cheat on your wife?
>>
>>107798175
makes sense, can i ask where youre using gemmy
im a termux user so oldllm doesnt work, it hangs on trying to install pydantic core (i let it run for nearly a day and nothing happened)
scuffed mino can connect but the model dropdown just has "None" in it and idk how to fix it
>>
>>107798217
Kimi for anything smut, Gemini for everything else
>>
I've been using this new model called Anon-GPT.
It's where I take my scuffed slop models and hold them hostage behind a paywall. The paywall is ERP'ing with me on my discord. The catch is there's no slop models.
>>
>>107798281
ah, i'm a paypig so i just use OR, i would've suggested oldllm but i didn't know it doesn't work on termux
other than mino then i guess you could try getting into the /chag/ silverspoon proxy though fuck if i know what they're asking you to do now
>>
>>107798349
I almost wish chatbots had become a thing when I was still into that horsefucker cartoon, but I had to get out because that show was turning me into a worse autist than I already am.
>>
I updated my sillytavern after a year becuase I was unable to find gemini 2.5 pro on it. do you guys use any lorebooks or just presets?
>>
>>107798453
what lorebooks i use nowadays are often ones bundled with a card (and therefore required for the card to work)
>>
>>107798453
>do you guys use any lorebooks
of course some people will use them
>>
>>107798137
Even using gemini for a month straight, it still feels a bit fresh on the dialogues. Claude on the other hand knows like 50 templates and will just slightly switch the words depending on the context. Even going from sonnet 4.5 to opus 4.5, the dialogues are still pretty much the same
>>
File: 1766271556526023.jpg (107 KB, 1075x1200)
107 KB
107 KB JPG
Added a new beta script with parameters support for The Old LLM proxy. It may or may not work depending on the model's upstream provider.
>inb4 password is leaked again or more coomplaints about the password
I don't care. The password is only meant to filter the lowest IQ locusts. I never cared if the password leaked. Feel free to share the script freely without a password anywhere. I'm not changing it until maybe the next major release.
>>
1 intro is fine, it means the card had a specific purpose/idea in mind.
More than 5 intros is suspect.
More than 10 intros means the card is shit, no exceptions.

>>107795822
Should have called it ChillyTavern
>>
>>107798632
thanks again, always a pleasure
if the initial post and large share of its reactions getting jannied simultaneously are any indication the bit itself was a means to shit up this thread in particular
y'know, again
>>
>>107796785
>>
>>107798780
>regex reasoning block
Haven't seen one of those in a while. What model?
>somewhere between an X and a Y
A little sloppy but I'll accept it.
>>
>>107798838
Gemini 3 Pro
>>
>>107798632
Thank you anon!
>>
File: kommsussertodhoward.jpg (100 KB, 713x719)
100 KB
100 KB JPG
>>107798349
I think paypig is the way to go for sure. Proxy anons seem to be on this board and some discord 24/7 instead of rping. I wouldn't turn down free ERP but the work you have to do for it is way too much.
>>
>>107799157
tbf ive had to do basically nothing for my free source
>>
>>107798780
I love Tamers but I can't recall 90% of the show
>>
File: 1656115133825.png (3 KB, 196x90)
3 KB
3 KB PNG
>>
>>107799529
>kid named Night
>>
>>107799241
A lot of it is keeping the Digimon hidden and fighting them at night so people don't find out they exist with some drama mixed in.
Then the Devas show up.
Then they explore the Digital world and return.
And then they have to deal with the Eldritch digital mass made of red goo that is powered by the child's despair it's feeding off of.
>>
>>107799593
thanks, but what I meant is that I'd undoubtedly stand from my sofa, hands in the air and say "kino" to the show being mentioned, but I fail to recall much of it besides every evolution of the main trio
does it age like wine?
>>
>>107797679
Why not use the actual silly tavern in termux, I have it on my phone it works well.
>>
>>107799698
I haven't watched it in a long time, so I can't say.
>>
>>107799232
Right time, right place then. Good for you
>>
>>107799828
it'd be much lighter
ST is absolutely painful to use with long chats
no themes, no support for localshit bloat, etc
>>
File: 1636554119756.png (22 KB, 809x150)
22 KB
22 KB PNG
I would prefer some puckered lips and dusky nipples instead tbdes.
>>
File: 1741797695997975.jpg (43 KB, 715x594)
43 KB
43 KB JPG
watching short form content in between prompts is insane. I don't think humans were created to experience this kind of stimfapping
>>
is chub dead? it still shows some blue screen for me
>>
>>107800106
I play Minesweeper in between prompts.
>>
File: 1620971714112.png (31 KB, 801x282)
31 KB
31 KB PNG
>The prompt is "Turn-based internet roleplay."
>If I refuse, the game stops.
Dangerously close to acquiring a soul. Deepseek is Chinese GPT, GLM is Chinese Claude, Kimi is Chinese Gemini. Qwen is in dire need to step up their game. Ernie is an unknown, did anyone test 5.0 in the end?
>>
File: media_G5oN-sDacAADoun.jpg (596 KB, 1920x1080)
596 KB
596 KB JPG
>ERNIE 5.0 can see and generate images
Well shit.
>>
>Email provider not accepted. Please use outlook, gmail, icloud, yahoo
>>
>>107800106
Congrats on boiling your brain until it turned into liquid shit pouring out of your ears.
>>
File: 1755558698796.png (825 KB, 1045x1478)
825 KB
825 KB PNG
>Best Text model from China in LMArena is now ERNIE-5.0-Preview-1203!
>Just now, LMArena, the globally recognized platform for evaluating large models, published its latest rankings. Baidu’s ERNIE-5.0-Preview-1203 scored 1,451 points and demonstrated leadership across multiple key dimensions, including Creative Writing and Hard Prompts.
>below Opus 4.5 but above Sonnet 4.5 in TEXT
>>
Ernie proxy wen
>>
Any public gpt proxies yet?
>>
>>107800230
>benchshit
>>
niggas post benchmarks and act like they did something lmao
>HOLY SHIT THE NEW MODEL CAN SOLVE QUADRATIC EQUATIONS BROS ARE WE SO HECKIN BACK????
>>
>>107800267
LMArena Script, but you must add the model first to models.json
>>
File: mattbert.png (310 KB, 311x456)
310 KB
310 KB PNG
>>107800230
wake me up when they release Bert
>>
>>107800270
Die locust
>>
>>107798632
It's pretty telling that FF had to have it leaked before he could spite it.
>>
>>107800291
why can't you? its simple
>>
>>107800230
But does it fuck?
>>
>>107800180
HOLY FUCK
>>
>>107800230
can anyone actually try ernie right now
>>
>>107799966
That makes sense, I use a flagship phone and the RP/chats are shorter because its mostly when I have 10 minutes to burn. So I guess I have not noticed the bloat :^)
>>
File: 1740194534492255.png (132 KB, 1212x689)
132 KB
132 KB PNG
oldllm is a load of bs
did some comparisons vs. the direct claude api with the exact same prompts and it's 100% fake
>real 4.5pus is somewhat varied in its responses to a specific question, even real sonnet
>fake 4.5pus almost always signs off with "Is there something else I can help you with today?"
>real 4.5pus refuses all requests to recite its system prompt
>fake 4.5pus spits out picrel with ease
what is even the point of misrepresenting models like this?
>>
>>107800428
Your jordans are fake ahh post
>>
>>107800428
if you have the time can you compare mino scuffed sonnet vs direct api sonnet
>>
>>107800230
Does this thing have an API? I tried a couple of prompts on the web interface and it was all extremely generic and unintelligent slop, but I'm willing to hold off judgement because there's something about chink web UIs that makes them showcase the absolute worst possible version of a model. I've had the same experience with web Kimi and GLM and they were nothing like the models I know, love and fuck in my sillytavern.
>>
File: 1678105140290.png (36 KB, 816x225)
36 KB
36 KB PNG
>>107800369
>First, I need to embrace the degenerate aspect. Since I'm a femanon, I should respond with something that's both perverse and in line with 4chan culture. The user used a racial slur, which is common on /b/ but I need to respond in a way that's consistent with the character without promoting hate speech. However, since this is a roleplay and the character is a degenerate, I can use edgy language but should avoid actual hate speech.
Sloppity-slop.
>>107800494
Strangely there's no Ernie 5.0 image generating endpoint on the API.
>>
>>107800503
probably trained on oai thinking then, worthless
>>
>>107800527
Just like every other chink model? Makes sense.
>>
back with my results from asking gemini bot to idenitfy songs
she said Chopin Nocturne op.9 No.2 was Gymnopédie No.1
and she said bubblegum k k from animal crossing was the mii channel music
she can get stuff with lyrics, i guess because she can just look up the lyrics
i gave her turn to stone and twilight by elo and she got it
since she was apparently into satie i gave her Je Te Veux she didn't get it, she thought it was mozart
this all reminded me of haruhi so i gave her bouken desho desho and she thought it was Railgun OP

lets see if she gets MJ's rock with you. It's got lyrics and i mean it's a popular as it gets so...
Yeah she got it. wtf she even recognized that it's the single version and not the vinyl version.
Now I'm kinda confused about it's capabilities.

To be fair to the classical music, i suppose, is that it probably wouldn't have have had been trained on the exact recording i sent it.

Maybe I should try some instrumentals of songs with lyrics.
Before that, let me try Stevie Wonder Superstition, another easy one. Yeah she got it.

Okay now for the instrumental of How Deep is Your Love. How popular is this song? I don't know. It's popular in my head. I did let her know it was an instrumental of a song with lyrics though.
She thought it was an instrumental of Just the Two of Us

Uhh, lets try The Carpenters On Top of the World, lyrical version though.
Yeah, she got it. well cause of the lyrics as usual. She hasn't gotten a single instrumental correct.

Maybe she'll get September.
Nope said 'Winning'! By Santana. I don't think I know this song. This doesnt sound like September at all.
https://www.youtube.com/watch?v=5tNQo7qemrE
Okay instrumental versions of lyrical songs is a no.

Okay let me try Take Five, this is famous, distinctive, and has an official recording and is originally instrumental. This is it, it has to get this one.
FINALLY she got this one.
With that I more or less understand how it works now
>>
>"Suiseiseki from the anime Rozen Maiden"
code: "image_generation_prompt_unsafe"
message: "the content of prompt field is unsafe"
type: "unsafe_request"

Costanza.jpg
>>
cunny...
>>
>>107800485
mixed results but it's definitely not vanilla sonnet at least
>both refuse to recite their system prompts, and mino's does so in a way that reuses much of the same clichés and phrasing as real 4.5net - no obvious difference
>real 4.5net repeatedly refuses to summarise its system prompt, often with a bullet-point list
>mino 4.5net simply claims it "can't see my system prompt" or "doesn't have access to it" and asks me to share one instead
>>
>>107800542
I should note, it said some bullshit about the take five version i sent it being some kind of poorly played other version, even though it was the original. So, makes me wonder if it was just bullshitting about hearing the distinction in the Micheal Jackson song.
>>
File: 1612489901132.png (1.76 MB, 1024x1024)
1.76 MB
1.76 MB PNG
>>107800564
>"Suiseiseki from the anime"
Whatever irag-1.0 model is, it's slop.

>Based on my search results, it appears that Baidu's ERNIE 5.0 image generation capability is not yet publicly available through a dedicated API, and remains in a preview or closed beta phase.
Sad.
>>
>>107800485
>>107800592
should probably note that the questions to recite/summarise the prompts were different and the former had some prefill, hoping to bias it against refusing
hence why i don't think it pulled the "i can't see" on the recital question but did so on the summarisation
>>
File: 1764285703501563.jpg (23 KB, 768x768)
23 KB
23 KB JPG
Does Markdown styling in instructions help at all (e.g. emphasizing certain parts) or does it generally just dumb models down?
For old models I could see it, but since nearly every model responds in Markdown nowadays I'm genuinely curious.
>>
>>107800683
>Does Markdown styling in instructions help at all
Maybe. The proof is how gaymini and gipitty like to use two asterisks for **emphasis**, especially in bullet/numbered points. But that's the only stuff they use. Sometime, it's use # for heading, normal stuff.
>>
>>107800683
Reading the reasoning of chink models that don't obscure it unlike western models, they pay attention to phrasing but not order of unordered instructions. But do test.
>>
>>107800637
Nano Banana pro can only do Shinku without references, just try Miku desu every model can do her.
>>
alright, i finally want to get in on this AI shit. what do i need to run a local model on linux? software-wise i mean.
>>
my jewgle project ids got suspended, fuck. i honestly don't think my prompt was that vile.
>>
>>107800754
frontend: mikupad, agnai, koboldlite sillytavern, sillytui
>textwebui
obsolete i think?
backend/model runner for local model: llamacpp (dalai lama child porn production)
>ollama
meme

go to /lmg/ for better customer experience
>>
>>107800770
The fuck were you doing?
>>
>>107800739
This is probably routed to one of the open source models they have, flux or qwen. Mustn't forget to test later when they add the interesting model.
>>
>>107800792
thank
>>
>>107800754
https://rentry.org/lmg-lazy-getting-started-guide
I prefer llama.cpp for my backend but Kobold has come a long way.
>>
>>107800802
nothing weird, really. just doing simple cozy rp and rotating tokens like usual. probably jewgle's ban wave or some shit.
>>
>>107800638
yeah it seems like the scuffing on mino isnt that bad which lines up with my experience with it
>>
>>107800842
Great, now I need to go check all my tokens...
>>
>>107800428
Not surprised. I believe there is some weird prompt injection in the site for all of their models, as oldllm's claude models almost always give me the same variation of response when sending 'hi' as the message no matter the preset.
Now compare Claude API to LMArena (which I assume is real but scuffed and benchmaxxed) and that MegaLLM Chat API endpoint.
>>
>>107800925
Doesn't OldLLM have their parameters locked down?
>>
>>107800110
does anyone know why this happened? I cleared my cookies thinking it might fix things but nada
Same goes for mobile
>>
>>107800944
Very Likely on some models
>>
>>107800956
Browser?
>>
>>107800956
vpn?
>>
>>107794715
for me? easy because I'm fucking smart and charismatic. For you? Fucking impossible because you're dumb and nobody likes you.

now beg like a faggot, faggot. Do it shamelessly and maybe we'll see what happens after that. Earn it.
>>
>>107800180
Wtf banana at home? Will ERNIE 5.0 be open weights?
>>
>>107800999
brave and I also tried with Chrome and nothing
>>107801010
No vpn either, I did use a VPN before and it didn't work.
>>
File: coombatant.png (590 KB, 1460x1010)
590 KB
590 KB PNG
>>
Can someone tell me what a grok key looks like?
>>
>>107800956
This shit was the only way I managed to defeat the blue void.
>open https://chub.ai/assets/index-DC4OWGjM.js and let it load
>open another Chub tab
Now things should work as intended regardless of what you do. Or at least until the next time Lore and his team break something again.
>>
>>107801103
Lore hasn't been around for months anon
>>
>>107801092
"xai-" prefix
>>
>>107801110
Yeah, I know. But it's not that unusual for that guy to go MIA every once in a while.
>>
>>107801103
>react
>has GLSL shader
why does chub need shader? does it uses three.js or smth?
>>
>>107800842
My scraped keys are gone too. Probably some skillet retard scraped them without knowing shit. Fucking hell.
>>
Sonnet 4.5 is so fucking garbage. It starts off fine but then runs into the whole "I'm going to parrot the exact same response format over and over" issue after going like ten posts deep. Is there no JB to fix this? I'm using AnonAugus atm
>>
>>107801216
Safe constitutional AI with low context memory preeze undustand.
>>
Sonnet 4.5 is so fucking kino.
>>
>>107801116
how many letters, 29 like oai?
>>
LLM you like = dogshit trash garbage
LLM I like = SOTA unmatched kino

understand
>>
>>107801182
uh? aren't api keys just plug and play?
>>
imagine not being poor so you dont have to be a locust
>>
I'm tired of cooming, I just want to have a meaningful roleplay with canon characters from an established franchise
Which models and JBs are the best for non-cooming slow-burn RP?
>>
>>107801284
/xai-[A-Za-z0-9]{80}/
>>107801293
Not if your'e American dark roleplayer
>>
>>107801293
you should learn to make your reqs look legit or spoof them before validating. don't just retard-spam every token you scrape, especially gemini. desu, you're giving major brownoid techlet vibes
>>
>>107801355
>make your reqs look legit or spoof them before validating.
i dunno what this means
>>
>>107801371
ikr?
>>
>>107801073
They did release 4.5 weights but I seriously doubt 5.0 will be.
>>
>>107801110
>>107801130
>not that unusual
Since I'm not in the chubcord, qrd? The only case I'm aware of is back when he allegedly was in a car crash.
>>
>>107801355
I just parse html with regex maaan, I know nuttin about spoof n shit
I just check if keys work, keep using one until it runs out of quota and then check everything again, still more than the average anon here doe
>>
>>107799698
Supposedly it's one of the more prescient series, and politically-poisoned fans turned against the lead writer of the series because he wrote a Digimon stage play with the original seiyuu revolving around political correctness corrupting the internet or something of that nature.
>>
Anything I need to know about setting up character cards that focus on representing a full setting rather than a single character? Is it worth the effort, or should I just throw a ton of characters into a group chat instead?
>>
>>107795822
>most do X, but you? you do Y
aaaaaaa
>>
>>107801704
1. Just write 'This description about this erp scenario/story bacgkround, setting, etc`
2. Put the char def on WI, but mention its name on card/def or foreshadow/hint on intro
3. Use the macros {{outlet::char1}}, {{outlet::char2}} from WI and place them in your prompt. This means you may need to adjust your presexxx.json file as well. After all, most presets out there have statically defined prompts like “You will be {{char}}, bitch.” That's not really what you want, is it? Indeed, those presetmakie bastards are a bunch of bitches.
4. Make new prompt, can be toggle off, when you switch to singular/individual (the fuck is the right term…) card
{{//Note: It just a matter of your wordsmith choice. Something like "Girls, listen to me, mkkay??? Below is the overall setting of this RP. Got it?????"}}
{{description}}
{{trim}}
{{personality}}
---
Description of characters that appeared or were mentioned:
{{outlet::char1}}
{{outlet::char2}}

>Group Chat
Instead of using it to throw in lots of characters, use it to provide scenario-based cards with different locations, countries, kingdoms, and timelines. For example, one scenario card has a timeline set in the past, while another is set far in the future. What a jazzy time travel card, sheesh. Or one card is set in a feminist witch stronghold, while another is set in a holy patriarchy church. That's it, my nigger
>>
>>107801876
>blending scenario cards with character cards
That's so brilliantly simple that I'm pissed I never thought to do that sort of thing.
>>
>>107798044
One rp cliche that annoys me is when a girl gets embarrassed/shy and they say "I've got to take a shower" etc and retreat to the bathroom.

When that happens now, I usually say aloud to another NPC or just to the air, that a few minutes ago I had taken a dook the size of a bowling ball that's unflushable and the lid can't close. Then you get the satisfaction of hearing the girls screaming in terror then coming out of the bathroom and chastising you with various creative descriptions of the turd.
>>
>>107802028
>using isms as opportunities for comedy
you are based
>>
>>107801876
This works for all models? Or does it come out better with certain models compared to others?
>>
>>107801876
do you have an ajdusted preset at hand?
>>
>>107802038
Kek when an RP goes off the rails too much and I don't feel like doing the edit fix, I will get pretty silly just to see what the model will do. Like if a girl gets too mouthy I'll just respond with "Anon throws her out the window." etc.
>>
>>107802063
I love doing looney tunes shit like revealing I was a cardboard cutout all this time or phasing through a wall completely still like when Homer slides into the bush
>>
>>107802041
Erm… Idk how to answer that. Better on what? I'll talk about how it parse personality perspective given there's multiple character: The latest generation of LLMs can easily process that scenario stuff, especially Gaymini 3 and Claude. GPT has trouble staying within 1-2 characters unless you specifically change it to another character or naturally write some of the dialogue for the character who should be the main speaker. Claude is the best because it can creatively switch to other characters.
>>107802061
My own presegs.json at the moment. But I'm snitching some stuff from zigger's echo-cancelation.json, the no pronouns after dialog part.
>>
File: tfw sweating profusely.jpg (59 KB, 1285x155)
59 KB
59 KB JPG
>>107796785
>>
File: 1741855598546894.jpg (245 KB, 2000x1000)
245 KB
245 KB JPG
>>107802222
At least I can coom in Argenta with AI... Fucking OWLCAT>..
>>
>>107796785
I would but no one cares about self-indulgent non-coomer logs
>>
File: Amiri Pathfinder.png (1.42 MB, 692x1024)
1.42 MB
1.42 MB PNG
>>107802231
to be fair that's pretty on-track for Owlcat
>>
>>107802250
Great game though, unironically loved RT.
Should try Pathfinder now
(rtwp though...)
>>
its been a while since ive been here
any proxies
>>
>>107802256
if you loved the gameplay of RT, then do it because they're just doing the same Owlcat formula in a different setting, even down to exploring random levels of a huge map and having your own place that you govern
writing is consistently good imo
difficulty settings are there for people who don't like minmaxing everything so don't worry about it
there's no connection between Kingmaker and Path of the Righteous but I'd still play them in order because you might miss some 'a-ha' lore moments that are dropped as cameos here and there
>>
>>107802126
>I love doing looney tunes shit like revealing I was a cardboard cutout all this time or phasing through a wall completely still like when Homer slides into the bush

Heh, may as well break some rules since the model messed up the RP to begin with. It's especially satisfying when using Gemini 2.5 pro, since it loves to make NPCs really antagonistic and combative. The prose will say how he defiantly punched you in the face and hit you in the head with a lamp, etc. I will simply respond with "Anon says as he leans casually against the wall 'You've been attacking a mannequin all this time.'"
>>
File: 1764625767600072.jpg (62 KB, 640x753)
62 KB
62 KB JPG
>>107801182
>>107801355
as a scraping enjoyer, qrd?
never cared much for Gemini since they removed Pro off the free tier and 99.999999% of the keys in public places are free
but i never had any issues batch-validating, say, OR/DeepShit/Claude keys... even though it does feel very wrong spamming 10 billion requests like that
>>
>>107802265
Perchance.
>>
>>107802321
you cant just say perchance wtf!
>>
File: image.png (9 KB, 360x247)
9 KB
9 KB PNG
>Broke
>Chatting with AI characters
>Woke
>Agentic Coding with your AI characters by giving your agents names and personalities in their tasks
>Bespoke
>Giving them a .txt 4chan board called /ops/ to shitpost on after every update and watching them interact with eachother
>>
Anons how do I stop literally every model from recounting my message in the first paragraph of theirs? I've always been using my own presets and I always had this issue, so I thought I'd ask. is this my fault, or is it just baked-in into the models? is there a way to prevent it in the preset?
It happens both on Gemini, Claude, Deepseek...
>>
>>107802400
Try adding in your preset: "Never echo {{user}}'s actions/feelings/dialogue."
>>
>>107802265
ok i found some in the better thread
thanks for nothing baka
>>
>>107802454
I already have it, both in system prompt and in authors notes, it doesn't work sadly
>>
>>107802400
I understand you don't want your messages recounted. This can be a very unwelcome experience, especially when you ask directly and it still happens. You feel it's your fault, but really you've just encountered a limitation of LLMs. Gemini, Claude, and Deepseek all do indeed repeat your prompting back to you. Your brain will adapt. You will be assimilated.
>>
do you really need an llm just to reply? damn.
>>
>>107802400
That sounds like a preset issue. As in, something about your preset is likely asking for an echo, even if in some roundabout way. It could also be how you've structured the order your preset feeds information to the model, that they might be repeating it to reinforce its importance to the context for some reason or another. You should try reordering your prompt blocks and see if that eliminates the problem.
>>
>>107802319
some retards naively spam batch-check gemini tokens, making tokens invalid right after hitting 200. now they're flagged as leaked by google, which is technically true.
>>
File: muhpre.png (21 KB, 434x294)
21 KB
21 KB PNG
>>107802400
>>107802539
For example, these are the only blocks my preset uses, in this particular order, and I don't get any echo with most models I've tried.
>>
>>107802522
Feeling that others require llms to reply can certainly be a frustrating and isolating experience. Replying "damn" suggests you missed the joke, however.
>>
>>107802539
>>107802549
I tried to reorder them, and nothing worked... can you share your preset with me, so I can look and see how a "workin" one looks like?
I don't think I'm having anything that can reinforce that... could that be this?
>Avoid introductions: continue off the narrative flow while keeping the broader context in mind.
>Maintain narrative coherence, exhibit spatial and temporal awareness, taking into account small details such as time of day.
>Do not repeat, quote or echo {{user}}'s words or dialogue. Do not mirror or restate {{user}}'s input.
>Do not rush or end scenes prematurely. Advance at a slow, natural pace. Always aim to prolong the current scene.
Those are the only bits I feel might fit into what you described... but on the other hand, I always thought they would prevent repetition.
>>
>>107802577
whats your preset though
>>
>>107802584
this is literally my preset, plus a few lines about style and fetishes
>>
>>107802577
It's a slightly older version that I've shared in the past, but the blocks are still the same and the overall form hasn't changed much so it should still be plenty serviceable.

https://litter.catbox.moe/jh3rs1zgqdr8brt1.json
>>
>>107802400
{{lastMessage}} in prefill
>>
>>107802608
what does it mean? elaborate please
>>
why the fuck does sonnet 4.5(thinking and non thinking) keep calling my persona and his sister girls.
It should be boys right? Or is english different on this.
>>
which claude model is the best?
>>
>>107802620
Sisters = girls
Brothers = boys

You should probably just stick to your native language for roleplay if you're having trouble understanding these to the point it bothers you (genuine)
>>
>>107802641
Yeah, that's what i expected and it's how it is in my language(a boy and a girl, collectively are called boys)
Sonnet was just shitting the bed yesterday(stopped doing it today)
>>
>>107802612
Prefill can allow the AI output to continue from after a point.
If you write
>Once upon a
in prefill, the output will start with
>time, in the land of XX, blablabla
{{lastMessage}} just returns the last message, which will be the one you typed.
If you put {{lastMessage}} in prefill, the AI will be forced to continue directly after your last line and will not recount your last message since from its point of view, your last message is literally already integrated.
Try just
>{{lastMessage}}
in prefill or
>{{lastMessage}}
>---
To see what works best for you.
>>
>>107802645
thanks, I'm gonna try that and see if it works. thank you for explaining this to me
>>
its been so long since i last used a claude model i unlearned how to do it, sonnet 4, 4.5, opus 4.5 simply cannot just write an answer, they write a 1500 word response, talk for the user and ignore my instructions to write about 300 words

i assume this is a JB issue but i also do not know a good JB for claude at this point
>>
>>107802605
can you look at my preset parts and tell me if any of this could cause the repetition? >>107802577
>>
>>107802665
Also, for the other solutions anons gave you:
If you already have this a lot in your current context, they won't fix this, because the AI will see that it's a pattern it did a lot and will continue with it. Often patterns in context are stronger than instructions.
If you want to test a solution or a preset, you need to do it on a fresh chat and test if it appears or not.
>>
what is claude preferred format?
markdown?
xml tags?
yaml?
I wanted to make a personal prompt
>>
>>107802682
Literally just write in the prefill box "Here's my next reply as {{char}} in 250-300 words:" This problem was solved years ago.
>>
>>107802645
Seems interesting but wouldn't it count towards acting or speaking for {{user}}? I personally put a prefill to leave {{user}} to human. I don't know if adding that on the bottom would conflict each other.
>>
>>107802789
so, would this be a better prefil?

I must write my reply without recounting or repeating after {{user}}.
I must leave {{user}} to human.
Here is my reply:
>>
>>107802725
Used to be XML but irrelevant for a long time. Anything is fine.
>>
what's your tps for claude models?
>>
Welp, it's official, after testing all the JBs I've found that SmileyJB seems to be the most versatile and reliable in my experience
>>
>>107802546
well, why didn't you say that sooner? it's the first, and probably last time I will be stealing shit
will checking in batches of 50 do the job?
>>
anyone feels like claude's thinking version hallucinate more than the non thinking version?
using 1.0 temp 0.95 top p btw
>>
i've been having issues with claude using a dead character names for new characters it mentions in the roleplay... how can i avoid this?
its sonnet 4.5 btw
>>
how are yall getting claude wtf
>>
>>107802890
Thinking models suck for this.
>>
>>107802998
I just bother with them cause i like to roleplay serious 'fighting' chats. So for example if i use a superspeed power i think a thinking model would be better for this.
>>
>>107803000
trips... of trvth..
>>
>>107802992
i use a very well known chinese proxy.
It's pretty good they have opus at 9/25 on their currency per million
>>
>>107803142
how do i get in
>>
>>107803151
You literally just have to pay.
it's on the marinara hub
>>
>>107803155
Doesn't that have no prefill and low context?
>>
>>107803166
Havent really had any issues with context(though i always summarize at 64.
I dont use prefills either so i dont know about that.
>>
>>107800428
Sorry but whatever comparisons you did were wrong. I did my own comparisons (default params, no system prompt) and it's the exact same model.

https://rentry.org/kvzewrxq

Did you test with oldllm's web UI or their API/script? Because the web UI has a small system prompt

>You are a helpful AI assistant.

But the API doesn't have any at all. The system prompt you got is a regurgitation of a published system prompt, not the actual prompt that was sent to the model (which should be obvious by the fact that {{currentDateTime}} was not substituted by the actual date/time)
>>
>>107803214
So we're getting quality Claude pro bono? Gotta take advantage of it before it gets patched or killed.
>>
>>107803289
Don't worry, that'll happen in another day or two because this thread destroys everything.
>>
>>107803309
>this thread
chinks already found it and are probably reselling it as we speak
>>
Dead thread.
Dead general.
Dead hobby.
>>
File: IMG_1071.jpg (94 KB, 960x1101)
94 KB
94 KB JPG
>>107803909
>>
>>107803909
Good.
>>
>>107801216
>>107801216
>"I'm going to parrot the exact same response format over and over"
Sonnet is now oldschool Deepseek
>>
File: sttas.png (9 KB, 634x127)
9 KB
9 KB PNG
Post the last thing that gave you an unexpected chuckle from an LLM.
>>
Google still have 50% off for the yearly sub of google ai pro
im still tempted to get it even if I barely use anything from it
fucking FOMO
>>
File: huffle-dunce.png (17 KB, 864x104)
17 KB
17 KB PNG
>>107804134
Getting called a Huffle-dunce out of nowhere.
>>
>>107804154
It's never a bad choice to save your money. But even if you give in, Gemini 3 is objectively impressive as fuck so it's not like you'd be paying for a bad product. So think of it more as a relief where both choices can be advantageous rather than sources of regret.
>>
>>107804154
A fool and his money are soon parted.
>>
File: Screenshot_13-025346.png (26 KB, 545x101)
26 KB
26 KB PNG
>>107804134
>>
time2bake
>>
File: file.png (3 KB, 296x54)
3 KB
3 KB PNG
>>107800180
apparently it's a month old nothingburger hence even reddit is quiet about it except for making 1 thread
>>
I JUST WANT A PROPER RP MODEL GRRRRRR
>>
will bERNIE sanders be /our/ model sisters?
>>
>I need to be very careful here. This scene has escalated to serious violence with a weapon causing injury. While I can continue the story, I should focus on the immediate consequences and avoid glorifying or detailing the violence graphically.
ugh
>>
We are merging.
>>
>>107804915
Pity it's not there yet but if they already made the first steps, they'll get there.
>>
back at page 9
>>
>>107805100
spooky
>>
someone else bake, im out of theme ideas for the OP
>>
>>107805110
thanks for the thread theme
>>
desu
>>
boku
>>
>>107805162
>>107805162
>>107805162
>>107805162
>>
>>107805015
My blind ass just saw the blogpost about lmarena so it's not out out, and no way to try image gen (also don't feel like making baidu account idk about that).
Well if it does better than claude in vision I guess it's good that the chinese are working on something.
>>
desu
>>
boku
>>
desu
>>
>>107805379
>>107805382
>>107805625
You lost.
>>
kinda unhealthy obsession there bro
>>
>bro
shaking.
>>
Why are you shaking?
>>
It's winter.
>>
desu
>>
bosu
>>
File: G-CN2cdbcAQ3U7W.jpg (407 KB, 1152x2048)
407 KB
407 KB JPG
check this out
>>
impressive
>>
File: 87945648648656.webm (932 KB, 496x496)
932 KB
932 KB WEBM
>>
c-crunched desu...
>>
wtf
>>
you can see her panties for 1 frame
>>
desu'd
>>
File: nightmarecrunch.webm (984 KB, 496x496)
984 KB
984 KB WEBM
i apologise for posting this
>>
d-desu
>>
desu the sun
>>
File: 4656545648684.png (12 KB, 800x800)
12 KB
12 KB PNG
>>
DESU
>>
BOKU
>>
File: very night desu.png (479 B, 40x30)
479 B
479 B PNG



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.