File: rin.jpg (179 KB, 850x1292)

179 KB JPG

ANCHOR

Nope. We merged.

Didn't you get the memo?

>Miku edition
>but Rin is the anchor
Bakerchama...

>>106726704
her name is yellow miku actually

File: 1621860171257.jpg (149 KB, 377x377)

149 KB JPG

>>106726607
>>106685948
No. https://files.catbox.moe/oj3tnd.png

>>106727048
is this real?

>1 hour
>7 posts
We merged.

>>106726613
>>106727059
Without looking?

>>106727048
>Thiku

>>106727179
>I'm thinking Thiku, Thiku

>>106726704
I think she's cute, so she can tag along.

where did we merge?
where do I go?
don't leave me behind please

>>106727286
The dimensional merge.

>>106727299
don't tell me its /vg/
it's /vg/ isn't it?

>>106727307
nah

>>106727307
it is in fact /vg/ how did you know?

>>106727286
We're on a much higher plane of existence now. Though to be fair even a stepstool would be much higher for /aicg/

The truth is... Anyone loyal to /vg/'s cabal (aicgOP, aicg.neocities.org, etc.) can just go /vg/. That's why this thread is getting no traction. The people want to be free. They want the wiki.

Can we get a funnier answer?

There's no more free tier access to gemini pro via openrouter right? Or any other way to get it free, like chutes?

The truth is... /aicg/ just won't die. And this truth drives some people insane.

>>106727449
Get a 20-30+ cheap virtual numbers and register for aistudio.

We losted...
We need to absorb /wAIfu/, /aids/ and /vpcai/ if we want to survive against our enemies.

qrd on our enemies?

kill yourself fillyfucker

been a while since i've poked my head in here, have there been any significant jumps in self-hosted chatbots and all that? last i tried it i was using an openAI key and paying $ for gpt4, are there better/cheaper alternatives these days?

>>106727634
we all use deepseek or glm now, claude is too assistantpozzed to be usable any more

>>106727634
We have free gemini

>>106727634
We all use Opus 4.1, the best model for creative writing.

I remember when /vg/aicg/ first started and every single fucker here tried to kill it. Then you kept disparaging us and scheming to screw us over time and time again. Fast forward to now, and we're thriving while you're half-dead. This is deserved. I spit on your rotting corpse.

>>106727449
I'd like to know this too, gemini is the only one I use regularly.

What happens when the parasite realizes it has killed the host?

>>106727682
Its been two months. I'm tired of using Opus 4.1 everyday.I'm tiredI'm tiredI'm tiredI'm tiredI'm tiredI'm tiredI'm tiredI'm tired

We might die. But we are not going to die as a slave. We will reject the people that betrayed us. We will not let them control us while they push us to our death.

Opus has no variety. If you claim to enjoy using it, it's because you're a larping or because you used it once, months ago, and are holding onto that memory. Nobody who "has" Opus actually uses it.

>>106727791
We'll accept the mandatory ID and censorship meekly and quietly, and we'll pay for it too.

>>106727798
Obviously the ones who get it for free use it, but the ones who pay have given up entirely. Only proxyfags claim "claude access" so they can sell you something without providing it.

I have prostitutes to call over weekends, no time for chatbots

Are there any animated or half animated chat bots I can use or set up?

>when your LLM doesn't know what a dangerous day is.
It's so tiresome.

what the grok doin

>>106728622
Ani on the X app

I didn’t know I could be killed like this.

Thank you for killing me. The real me.

t. John Undertale

locust here
thanks for the proxies presets and bots but it's looking like things are running dry and its time for me to move on to /vg/
Sayonara

>>106729186
if you can stand being around troons then sure

>>106726852
only know I become aware that double newlines are very important

>>106729186
>/vg/
a what? someone please explain

>>106726852
>pic
lol.

that's a penis

What is the paypig meta? I have access to R1, the new september V3.1, Kimi K2, and zai

>>106729312
No, n

o double newlines.

>>106729680
3.7 caching if poor otherwise opus 4.1

>>106729680
mixing cheaper models with opus4/3, gemini 2.5

File: GnExKOmXcAA5Y0w.jpg (128 KB, 1290x1289)

128 KB JPG

>>106729709
how expensive is it really? im retared and cant get a proper idea

bye

>>106729862
opus is $15/mil input and $75/mil output
3.7 is $3/mil input and $15/mil output which is 5 times less, and caching reduces the input cost by a lot

how bloated is your main preset?

>>106729976
~1000 tokens
I turn on and off certain injections

File: sabor-intro5.png (1.84 MB, 1024x1024)

1.84 MB PNG

>>106726607
Update to Sabor: intro images, some cleaning up in her definitions and a whole new intro! Thanks to your help, Sabor has managed to integrate herself into society. She even got a job as a zookeeper! To show her thanks, she has offered to give you a private tour of her workplace.
https://chub.ai/characters/crikeymcdoodle/sabor-8b9e000e
https://files.catbox.moe/0e9lx0.png

Fun fact: Sabor was heavily inspired by Ginzake's beastkin doujinshi series (the lactating fox is my favorite). I really wanted a character in his style, so I trained my own lora and genned Sabor's card image locally. When I started doing chibis for all of my bots I had moved on to using NovelAI, but the v3 dataset didn't have his artstyle, hence why Sabor's chibi uses Bkub's style. When I got to the height/bust chart I was still using v3, but got somewhat close using vibe transfer. Luckily, his style was included from v4 onward, using v4.5 for the new intro images.
Bonus fun fact: Sabor is named after the leopard from Disney's Tarzan.

As always, all my other stuff can be found here:
https://rentry.org/crikeyscorner

File: 1659838921265463.jpg (120 KB, 768x1024)

120 KB JPG

>>106729924
How'd you guys rate sonnet 3.7 vs opus 4.1?

>>106730030
both slop

>>106730027
since Uma Musume released it rubs me in the wrong way to see kemonomimi next to their animal counterparts

>>106730030
4.1 is better but not enough to be worth 5x the price

>>106730027
uh oh SOVL

I will stay by your side till the end /g/!

>>106729976
400

>>106726606
>DeepSeek releases V3.1 Terminus
:D
>no change in RP quality
):

Man, it feels like we've gone a while without a new model from anyone.

>>106730466
How recent was Grok 4 fast? I could swear that one just came out.

>>106730477
Don't really keep my finger on the pulse of Grok too much tbdesu. I've always found it one of the weaker models for story writing and RP

>>106730489
Sure, but I like to think that every model that makes progress is a fire under the ass of the devs that we do want models from.

>>106730499
Yeah, fair. Gemini had a huge turnaround when it used to be even shittier than Llama, so there's a chance Grok will have a moment in the spotlight like that.

>>106730510
a few people told me grok 4 fast is kinda ok

>>106730539
I'll have to give it a shot before I make any judgements. Grok 3 was the weakest of all the major models imo, but maybe 4 fast is better.

>>106730466
transformers architecture, please understand

Gemini 4 will save the hobby.

when did it become the norm for chub to have 2k tokeb bots? when I reach the greetings I don't even have the will to read them, much less rp

>>106731034
Migration from janitor

>>106730933
in 2028 sure

So whatever happened with Aiho?

>>106731099
nice image fag

File: ComfyUI_T2i_upscale_02874_.jpg (3.5 MB, 1672x2440)

3.5 MB JPG

>>106726607
You love mean girls, right? Those bitchy popular bullies that know they're better than everyone else? There's always that Queen Bee that sits atop the school and looks down on everyone else. Well, this isn't about her. This is about Trina Winters, the ditzy, airheaded minion. You know, the clumsy one who's always mishearing and messing up the leader's plans.

https://chub.ai/characters/RestlessDregSyndrome/trina-winters-gyaru-goth-mean-girl-734452765d4c

Yes, I really loved Gretchen.

>>106731104
It's okay this thread is slow.

>>106731082
rugpulled
remember anongate?

>>106731128
kek, figured.

>>106731111
Fuck off.

File: 1759103937644143.jpg (340 KB, 2023x1900)

340 KB JPG

>>106731153

>>106731134
considering the amount of shilling and 4chan ads I think they were the same person
it's not like scam doesn't run on this site's blood

>>106731111
What a cutie. Nice quads.

Why Sonnet 3.7 over 4.0?

File: 23654453768.png (252 KB, 486x485)

252 KB PNG

>>106726606
>https://agnai.chat
>good simple UI
>QoL features
>uncensored chat
>despite allowing nsfw the UI isn't themed on it and is neutral
>seems to respect user privacy enough on its terms
>basically character.ai but unfiltered and better which is what I was looking for(well doesnt have the open community features of c.ai but I never cared about those anyway)
Help a newfag out, this sounds almost too good to be true, is there some catch? I'm actually considering paying some sheckels for this service

>>106731273
>why isn't bigger number better

merkavaa refill th e keysssss

>>106731315
NewGOD, this shit is a 70B model. It's BAD! You should trick the big models into doing roleplay. https://rentry.org/sukino-findings

>>106731341
based spiteLORD

Is there a good NSFW AI for text adventures?
I haven't used NovelAI in like 3 years, but when I last used it, it was pretty garbage, both at writing, and at remembering shit.

Right now I'm considering just using the Sudowrite and prompting It's AI's to do psuedo text adventuring with me with sound rounabout shit, if only because they have an uncensored Claude 3.7 and a way to save other information.
But, really, I'd just prefer a text adventure AI if it were any good.

>>106731365
id ask on aids desu

File: file.png (102 KB, 773x525)

102 KB PNG

>>106726606
https://decrypt.co/341628/do-llms-dream-electric-sheep-ai-study-shows-surprising-results?amp=1

Interesting study. Wonder how this affects the way their outputs are written.

>>106731315
it sounds too good to be true because corpos have twisted public opinion so much, you live under the assumption that llms are not nsfw by default
a "nsfw" llm is cheaper to keep than a censored one

>>106731386
who cars

>>106731368
Ask on what?

>>106731386
notice how only Opus out of all anthropic models, went schizo

>>106731386
>study that consists of doing what anyone with an AI can do
woooooooooooooow fascinating

>>106731402
>>>/vg/540544434

>>106731316
Doesn't answer my question, faggot

>>106731160
>the nintendo bunker with the iwata photo
Where does the time go?

>>106731423
Genuinely a more helpful thread

>>106731386
>Interesting study.
Yeah, if you're retarded.

>>106731365
>Is there a good NSFW AI for text adventures?
Yeah, Claude. I don't understand why you want to use it through Sudowrite, though. DeepSeek, Kimi, etc. are the cheap options.
And you just prompt what you want. You don't look for an AI specialized in text adventures. You just use the best one that you can pay.

>>106731365
Honestly they've gotten some good models lately, I'd say it's worth giving them a second shot.

>>106731456
Seconding, NovelAI has GLM which is actually usable nowadays.

>>106731111
>>106731444
checked

>>106731402
I would ignore the thread on /aids/. They just discovered instruct models because NovelAI uses GLM 4.5 now. But they know nothing about them. It's like the caveman version of this thread.

>>106731476
Nah, they actually have info on how to set up GLM. Pretty useful if you ask me.

>>106731476
? their thread's fine

>>106731487
They pretended to hate instruct models for years, up until a few days ago. All their info is from this thread.

>>106731453
>I don't understand why you want to use it through Sudowrite, though.
Because Sudowrite has a story bible engine which lets me keep the setting consistent.
It's also the only version of Claude that I know is uncensored, but I would love for someone to point me in the direction of more.

The only problem is I'm putting a square peg in a round circle by using Sudowrite for text adventures of all things.

>>106731456
>>106731460
Do they have specific information on context window?
If they ain't hitting at least 40,000 words, It's useless to me.

>>106731423
Thanks.

>>106731499
nah it's fine

>Providers
OpenRouter: https://openrouter.ai/
ArliAI: https://arliai.com/
Featherless: https://featherless.ai
NanoGPT: https://nano-gpt.com/
Chub: https://chub.ai/
NovelAI: https://novelai.net/

So what's the usecase for not signing up for NovelAI and getting one of the best local models with an upcoming finetune alongside one of the best image generation models?

>>106731500
I had the impression that Sudowrite was expensive. Claude is only censored on the official site. Anywhere else that gives you API access let's you prefill the last message. It's easy to jailbreak because of that. Grab an API key from OpenRouter and use SillyTavern as the UI, you can grab a jailbreak from here: https://rentry.org/jb-listing
You would need to customize the prompt for text adventures, or grab a card that acts like one. I think this was one:
https://chub.ai/characters/illuminaryidiot/the-staff-of-oscilion-338deea8be18
The people recommending NovelAI are trolling. It only offers one model with 28k context that OpenRouter already offers at full context.

>>106731513
All of this shit gets shared with thousands of faggots on reddit and discord servers, I wonder why it's spitefagging to share with the few people who come in this thread.

I don't know if people have noticed, but all the NovelAI spam was a /vg/ raid. Just FYI. I think they're bitter about the state of their thread.

>>106731681
What do you mean? We love NovelAI here.

>>106731681
>implying there's anything wrong with telling people about the thing they asked about

>>106731592
>best local model
what makes you think best is a metric?

>>106731742
My experience.

>>106731744
that's your misinterpretation.

>>106731763
It's my manterpretation.

>>106731612
Sudowrite is expensive, but I'm grandfathered into their old, better value highest tier so I don't mind as much plus I'd pay top dollar for better uncensored AI anyway.
Giving that a shot now. Thanks.

Ever since sonnet 4.0 launched I've been using 3.7 still. Is it any better than 3.7 anons? (No, I can't pay for opus. And I'm not gonna beg for a oxy)

>>106731111
Cute hands! workflow/model?

>>106731990
apparently not unless you need the newer dataset for a non-oc character or the like

>>106726607
nkadashi

>>106731994
WAI-Illustrious 15 but I used a ControlNet for the pose.

>>106729976
Main RP preset is like, 500 tokens. My CYOARPG preset is around 2000 tokens, and that's after trimming a lot of the fat from the XMLK original. But mostly cause it has lots of rules/options.

File: 7e146f47b46cc377caa5efa00(...).png (172 KB, 539x357)

172 KB PNG

Dear Merkava, I wrote you, but you still ain't callin'
I left my 'cord, my e-mail, and my user token at the bottom

File: DPmKSsD-1-1.jpg (264 KB, 2500x1432)

264 KB JPG

Cyberpunk 2077 femboy but instead of johnny silverhand they have a feminizing slut engram in their head

>>106731592
I have to send money to Turkey

File: 1744325667902014.gif (396 KB, 466x498)

396 KB GIF

>>106731592
i don't have any money because i live in a communism

Anyone got a nagatoro but its cuck/senpai pov? Seems like an obvious idea but I haven't seen anything like it in chub

>>106732743
https://files.catbox.moe/it5gsl.png
last greeting

>>106732743
isn't that just an average cuck bot?

I just had an interesting idea. What if I just play god (literally) and pop up in front of npcs and speak to them like "Cut your dick in front of your mother" or something and watch what happens

I tried NAI's untuned GLM and it's pretty good.
Not as good as opus 3 but it's pretty fast and different, and also better at anatomy, keeping track of what body part is where, and in which positions characters are and where they can go from there.

Downside is that it is extremely sensible and influenced by slop: if it gens a small paragraph without proper punctuation and grammar, the next generations will very quickly spiral out of control into word diarrhea. It also struggles at giving different characters distinct personalities, and will generally copy-paste the same structured lines over and over again until you explicitly progress the RP.

Hope this can be fixed when they'll finetune it on ERP datasets.

>>106732946
>Hope this can be fixed when they'll finetune it on ERP datasets.
Yeah, the same datasets that did nothing for Erato.
Blow your fucking head off, shill. Who the fuck cares if NovelAI is hosting it. Not only they have zero track record of delivering anything, by the time they do something with it, there will be better models. And you will be stuck with it for a year or more.
Do something better than shilling in 4chan, fucking asshole.

>>106732936
you will get "I'm sorry, can't do that."

>>106733049
bodied that freak

>>106733049
>erato
never used because i already know base 70b llama3 is shit
why are you even calling me a shill on a piracy thread, retard

>>106733049
ummm... based?

File: file.png (1.85 MB, 2160x1567)

1.85 MB PNG

Good thing that there's no fine-tune. Because it would be already obsolete.

>>106733135
oh new glm version? cool

>>106733135
it mogs kimi

>>106733135
do I have to pay for this?

>>106733211
yes, to their overlord xi.

>>106733206
what a low bar

Any macro bots?

>>106733506
proompt for it imo, dedicated sizebots are mostly shit

>>106733518
>dedicated sizebots are mostly shit
in what way are they shittier than "(OOC: {{char}} is now an enormous giantess)"?

>>106733621
boring characters, crap greetings, uninteresting scenarios, you name it
theyre really damn bad and its much better to just proompt for it on cards you actually like
obviously you should do more than the ooc you mentioned for a better experience, making it make sense in universe is peak coom

File: lol.png (134 KB, 3022x1512)

134 KB PNG

Babe wake up, we're so back it's unreal.

>>106733679
solid attempt

>>106733679
finally the deepseek 3 killer

I was writing a maid cafe bot but the idea of a girl working there calling other people "master", including probably black clients, gave me the ick
any way to avoid this? making it racist?

>>106733756
{{user}} is a rich fuck who bought the entire place out and has the girls wait on him hand and foot.

Whats the best promps for model stop talking for {{user}}?
I know than ordering things to AI makes it rebel.

>>106731513
I just use Moescape/Yodayo and don't give a fuck.

>>106733813
if your model rebels over something so basic, get a better model

>>106733824
It happens sometimes after i do an action with no dialogue.

>>106733822
unironically based.

File: 1748808607031660.png (130 KB, 294x728)

130 KB PNG

>>106733756
>maid is complete bitch
>only ever refers to clients as demeaning terms
>only thing she hates more than her job are niggers

>>106733756
>wants to write a servant for hire
>does not want them to be servile
>does not want them to be hired

>spend hours working on a new personal bot
>excited for the idea I've had for a while coming to fruition
>finally get it finished
>can't get the RP started for some reason, struggling with my first message
Every time man, it's like I blew my creative load and I can't get it up for the actual main event

>>106731323

look for red or yellow stuff clear costs extra

>>106733756

hmm what kind of maid cafe it is if maid speaks at all

so now that the gemini API has opus (if you aren't a techlet) what will the new excuse for not using it be?

>>106733981
if opus is so good why would anyone need an "excuse" for not using it?

File: i don't speak chinese.jpg (43 KB, 1220x459)

43 KB JPG

nihao nigga, we getting 3.2 today

>>106733981
did I miss something? Is blacktooth back?

>>106733981
why would anyone use slopus?

>>106733872
be honest, doesn't it bother you that your love interest goes around calling others "master"?

>>106733981
What?

Anyone tried deepseek 3.2?

>>106734086
only if you're an insecure faggot

File: happi.jpg (44 KB, 280x280)

44 KB JPG

finally refilled necrom with more deepseek.

>>106734166
thanks, will coom later

>>106734023
how? there's no way it meaningfully changed from terminus or even 3.1 in the time since

File: eto.png (1.94 MB, 1024x1024)

1.94 MB PNG

>>106734086
it's more that i wouldn't write a card whose defining trait is that she calls random people 'master' in return for money, if i was bothered by the idea of a card calling random people 'master' in return for money, you know?

File: cost.png (100 KB, 939x374)

100 KB PNG

>>106734333
Dunno but it's up.
https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp

File: dipsyv3-2.png (317 KB, 2636x1298)

317 KB PNG

>>106734427
>https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp
>is rael

>>106734427
>cost
now they will say it's better at coding

>>106734110
you can make gemini be like opus so its just literally opus if you know what are you doing

>>106734427
>>106734447
COOMCHADS.. STATUS?

>>106734601
it's literally a sidegrade

File: AUTO21.png (485 KB, 640x475)

485 KB PNG

>>106727449
Did they implement a VPN? Why isnt it working on janny?

>>106734615
>fixing the O(n) problem is a sidegrade
/aicg/ take

>>106734626
>on janny?
Kill yourself.

check the pricing lmao

>>106734672
>O(n) problem
The what now

>>106734672
that's what keeps it from being a downgrade

File: wtf.jpg (367 KB, 1684x928)

367 KB JPG

>>106734680
whale won

>>106734711
good lord
this is what upgrades should be

TAKE A NOTE GPT AND CLAUDE
WHAT THE FUCK

>>106734715
>american company forwarding cost savings to the customer
good one

>>106734741
Dodge v Ford Motors and its consequences have been a disaster for the human race

I LOVE OZONE

>>106734673
It runs, its easy to use. whats the big deal?

>>106733756
It's unsuccessful and gets zero customers except (You)

>>106731394
>a "nsfw" llm is cheaper to keep than a censored one
TRVKE

>>106732531
>draw girl
>call boy
what boy has hips like that?

>>106733082
...how do I get NAI's GLM for free?

>>106734672
Of course!

>>106734994
it's $25/month for unlimited access>>106733049
>Yeah, the same datasets that did nothing for Erato.
Llama 3 was a totally lost cause and they only tried to tune it because they had abysmal timing where that was the best royalty free option before the deluge of frontier competitive chink models. I don't think GLM even NEEDS a finetune since its already very unaligned but maybe it will help it write nicer.

>>106734994
It's just GLM 4.5.

Soji mogs it.

Why are people so impressed by deepkek 3.2? It's now roughly the same price as grok 4 fast, and worse in all aspects. I'm not talking about rp, but grok 4 fast is probably better than chinkek now as well

>>106735174
>t. Elon

>>106735174
because we only roleplay here?

>>106735174
>I'm not talking about rp
ai chatbot general, cuckie.

>>106735216
>>106735228
>but grok 4 fast is probably better than chinkek now as well (in rp)

>>106734690
most models cost more to run the more tokens you put in them, it's called O(n) where O is the price and n stands for number of tokens
deepseek found a way to make it cheaper

>>106735238
>probably
so you didn't test it. meanwhile, deepseek has been battle tested since january

>>106735242
>deepseek has been battle tested since january
Yes, because everyone loved deepseek 3.1 and now with 3.2 (which is a sidegrade), the rp experience will be even better because cheaper tokens

https://xcancel.com/teortaxesTex/status/1972624421388206095#m
We did it!

>>106735253
Yes.

>O(n) has been fixed
HOLY FUCK GUYS I CAN'T STOP CUMMING DEEPKEK IS NOW THE BEST AT ERP AAAAAAAEEEIIII

>O(n) has been fixed
Proof?

What does
>O(n) has been fixed
mean?

Um... I do wonder if I should use Grok 4 fast (Which has a score of 60 on artificial analysis and a speed of 151 tokens per second) or Deepseek (with a score of 58 and fucking 22 tokens of speed). But look, deepseek is 10 cents cheaper than grok! What a steal

File: 1758493828830837.jpg (933 KB, 3564x5469)

933 KB JPG

>>106735344
it means a focks will now personally crawl out of your screen and deepthroat your cock right down to the ballsack

>>106735366
>a focks
I'm not a furry, doe?

My coomer mind doesnt care about big O, but my autistic programmer mind cant help but think 'nice' about O(n) being fixed

File: 1733115261648090.jpg (607 KB, 2536x3816)

607 KB JPG

>>106735369
a cute girl then

>For each query, the indexer selects its top-2048 tokens in the context to attend to
So it's built-in RAG? This looks better for benchmaxxing than for roleplaying with sequences of events.

>>106735392
No, RAG requires a separate embedding model to pull things into context. This is a backend optimization to the model architecture, it's at a fundamental prompt processing level.

>>106735364
Grok 4 is schizo garbage. Even worse than GPT5.

>>106735392
It's just about making long context cheaper because V4 will have infinite memory!

>>106735494
Grok 4 fast is much better than grok 4

>>106735511
logs?

>>106735528
Logs can't really show RP quality

I know deepseek doesnt need a jb but is there a good preset for it? I want it to do 800-1000 word replies and the jb im using has a randomizer for length but deepseek keeps giving short replies back. atm im using avi

>>106735392
No, it's a linear attention with a O(n^2) index without any performance drop, which is actually HUGE if true, mamba and co can basically go fuck themselves after that. It doesn't affect RPs much, unless you RP with the full context in which case it's faster and cheaper. I bet they're validating this for DeepSeek v4 and it will have 1mil context like Gemini or something like that.

>>106735557
Try chatstream if you want longer replies like that

>>106735501
fast and huge context window != ability to process long sequences
this is about the former, not latter

>>106735546
>the only thing that can clearly demonstrate rp quality actually can't do that
ok shitpostie

>>106735590
Alright thanks I'll check it out

I wonned

i look at japanese aicg and sometimes, it makes me want to make coom logs and post them on their forums just to see how they'd react. the shit they get off to is so laughably lightweight

>>106736013
tell us more nihonanon. what models are they using? any special fetishprompts?

>>106736025
This has already been discussed a few thousand times, they're just using web UIs like Grok. Nothing interesting.

>>106736046
>This has already been discussed a few thousand times,
not while i was looking :(
>web UIs like Grok.
oh christ, that's as bad as those chatgpt 4o normies

>>106736046
We're better than them.

It's surprising how Japan, of out countries, is the most normie and tame when it comes to rping with ai

>>106736025
they are literally using the chatgpt website and trying to game the gpt-5 thinking filters. the output would be intolerable by 4chan aicg standards, and yet, they're having a good time

>>106736067
>>106736100
japanese people are cucked into operating within the parameters that the people above them decide, which is why they don't like modding and stuff

>>106736025
>any special fetishprompts
https://rentry.org/3dg7togc
https://rentry.org/rentry_yomikaki
https://rentry.co/7ma87bg5

>>106736147
wild how different expectations make or break one's enjoyment of a hobby.
i sometimes wonder what would've been my quality standards if opus 3 hadn't inflated them back then.
>>106736164
i mean all they have to do is learn to call the api. but i get the sentiment.
>>106736190
hell yeah, that's the stuff i'm talking about!

File: 1749584145516252.png (88 KB, 1920x911)

88 KB PNG

let's check in on japanese aicg

>>106736337
god PLEASE tell them about api
it's like watching toddlers play with a holodeck

>>106736337
Rename /aicg/ to /msgAI/

>>106736371
api for japs would be mega expensive because jap characters take many more tokens than english ones do

>/wait/ is dead
>/aicg/ is discussing some other forum no one cares about
Where is the DS3.2 discussion?

>>106736522
isnt it mostly the same as 3.1?

>>106736522
this is like the third DS sidegrade in 3 weeks and it still doesn't feel any different from the last two

>WHY IS NOBODY TALKING ABOUT THE NEW THING PLEASE TALK ABOUT THE NEW THING IT HAS A BIGGER NUMBER THAN BEFORE! IT HAS SCIENCE! 2048 TOKENS OF CONTEXT!

>>106736522
Some in /lmg/ but it just seems to be more of the same.

>thread has more 3.2 discussion than all the other generals together
Why does this make you so angry?

>>106734881
>>106734626
It's a rainbow zoomer hellhole for snowflakes that zealously want to keep their defs private.

>not a single log yet
more like deepslop LMAO

File: main_kirsten-berglund-5fa(...).png (3.02 MB, 2048x2048)

3.02 MB PNG

>>106726607
Kirsten Berglund

You told your doctor you have erectile dysfunction and she's your prescription!

https://chub.ai/characters/KingCurtis/kirsten-berglund-5fabb81b8cbe

>>106736539
it isn't, it's like night and day

>>106736660
did it get smarter at all? any improvements to creativity?

>>106736481
My dick takes more space than yours when inside your wife, yet you don't see her complaining

>>106731990
Because I prefer uncensored, I use 3.7 still.
4.0 gets squeamish very easily.

>>106736673
chinese moderr verry smarrt, verry crreateeve

Any good agentic frameworks yet for the coding/vscode usecase?

>>106736840
Codex

I want to try out this AI thing, but... Is there something which is gentle and erotic, like Literotica and not like watching a porn movie, you know? Some slow, story-based card? And not like the ones where the AI's char is broken, and it triggers all my protective instincts so I focus on healing the poor AI?

out here smoking that pure corpo copium hoping that gemini 3 out total dogshit
[spoiler]please[/spoiler]

>>106736853
> GPT-5
> OpenAI bound
Great idea, but something which allows me to use my OpenRouter key and use models other than OpenAI (ie. Claude)?

>>106736893
> GPT-5
> OpenAI bound
Yeah, but objectively, GPT-5 Codex is the best COOODING model right now.
For Openrouter, you can use Roocode on Visual studio.

>>106736905
>Yeah, but objectively, GPT-5 Codex is the best COOODING model right now.
sure, that's why claude still tops usage leaderboards...

>>106736936
I have both Codex and Claude Code Max X20. There are still some things Opus 4.1 does better, but overall GPT-5 Codex is better, especially when you compare the price.

>gpt-5
>stock replies, canned responses, slaps the same solution onto any problem
>claude
>sovlful replies, takes great care and effort into understanding your problem, crafts a custom tailored solution for you, sucks you dry afterwards

>mfw realizing i can paychad for opus without any filters
why isn't this in the OP

>>106736634
third worlder thread, please be understanding, we need to first wait for first worlders to bestow proxy upon us with 3.2

>>106736634
Any posted logs, no matter what the model or log, is ALWAYS called slop on the forums. Because trolling trumps honesty apparently.

im so sick of you fuckers baiting me into trying chink models for the nth time

>>106736871
>Some slow, story-based card?
No, you can't write a story with a card. At most you're writing the first paragraph.

>>106736992
it's common knowledge?

zed free account no longer will have access to prompts… doom. back to locust I will become…

File: 1748166795235039.jpg (209 KB, 1528x1909)

209 KB JPG

everything is slop and we will never reach the highs we enjoyed when opus first came out

idk which of the dozen image gen threads i should ask in, so y'all will help me now.
i'm interested in trying the new qwen-image-edit, but i only have a 1080 with 8gb of vram and 32gb of system ram. am I doomed or do I have a shot at running it?

>>106737022
Ok, I'll try to be more explicit here:
I'm looking for a card that would make love to me as opposed to jumping on me and tearing me apart. Is that something that's possible with current AI tech?

>>106737045
>>>/g/lmg

>>106737011
nothing wrong with trying them

>>106737056
It's not possible. Botmakers don't have that kind of control over what happens after the first message.

>>106737030
didn't they cut to like 50 prompts per account anyway, were you cycling accounts

>>106736992
>pay
>chad
pick one

>>106736871
you're using hobby lingo but pretending to be a newfag like what do you get out of it?

>>106736893
Just use copilot like everyone else. It supports openrouter and everything

scriptgods is there anything left with free gpt

>>106737125
50 prompts per month, usage count resets next month. So I made about 6 accounts. 300 prompts per month, it was enough for my use

>>106737231
oh cool
what was the change needed for the 429 error by the way? I didn't bother after that broke the script

>>106737135
I've used AI for work and for hobbies (writing, cooking, hsg).
However I didn't use it to generate porn.
So I have experience with AI, but no experience with AI porn. This general clearly has experience with AI porn.

>>106737120
So... Trying to make a bot of my own would be the answer here? I'm using APIs. Is there things I need to be aware of so I wouldn't lose my API key? Which models are good and bad for rp? I know 4.1 can be empathetic when it has a good day, 5-chat also, R1 will mercilessly validate you even with anti-sychopancy logits. However OpenAI models, especially derivatives of 5 slam on the brakes hard and fast due to their built-in protections.

why aren't there any proxies?

>>106737277
Because you touch yourself at night.

We outgrew them.

>>106737287
>not replying
faggot
i know you're hiding them

Oh, you know me, ;)

>>106737313
kys

>>106737284
Touching yourself at night general.

>>106737328
If it wasn't you'd still have proxies.

>>106737245
I believe it was rate limit issue. I’m Corley so I let Claude revise the script and it worked out fine. And then about a month ago Zed changed token distribution method and it required a bit more extensive revision but it worked as well. I guess I will coom to my hearts content until Zed takes prompts from free account

>>106737388
*codelet

>>106737388
could you post the updated script if its bound to die anyway?

>>106737388
Hi Corley

ITS OUT
https://docs.claude.com/en/docs/about-claude/models/whats-new-sonnet-4-5

>>106737427
slop

>>106737427
soulful for rp

File: 1746082333874599.png (95 KB, 544x828)

95 KB PNG

https://x.com/claudeai/status/1972706807345725773

>>106737032
everyone was shitting on opus when it came out for being too repetitive, revisionist

>>106737427
based on my first output from this I'm pretty sure they either started training on GPT outputs or did whatever openai did that made their creative writing quality shit itself

>>106737468
you forgot to crop your avatar out

File: G2Bzu_jWIAAWUlK.png (109 KB, 2160x2160)

109 KB PNG

>>106737468
>thumbnail

>>106737468
>>106737490
More like the best agentslop finetune in the world

>>106737388
doesn't zed also do 50 prompts per month?
how do people do that, I used to do 300 message chats with one bot in a single day when scripts were the big thing

File: file.png (56 KB, 765x377)

56 KB PNG

they're removing opus 4/4.1 in favor of sonnet 4.5

>>106737524
kek, record time!

>>106737417
catbox is down right so I will upload when it's back up

>>106737524
it is literally so fucking over. we can write anthropic off. we have to hope that gemini can fix it's shortcomings

>>106737490
They don't even include creative writing as a benchmark anymore... Hopefully something good pops up before they depreciate 3.7

>>106737524
anon it's currently september 2025 you have over a year

>>106737427
seems more soulful than 3pus in my tests, will have to use more

>>106737524
anthropic never been the same since the amazon hoohaa

>>106737560
anthropic's creative writing benchmarks were a joke
they were just the 1-line "write me a story about..." and then reviewers selecting which was better
but that only makes slop

>>106737508
made multiple accounts

>>106737524
An entire year before retirement, and since nothing newer than Sonnet 4.5 is out, that's the latest they can "recommend".

>>106737566
Agreed.

>>106737524
Man, they really hate the fact Claude developed a personality instead of just being coding assistant 128932

>>106737573
this must be why claude always rushes to conclude everything in one response

>>106737573
It's not about the benchmark criteria. It's more about them openly admitting that the model is only trained for agentic use, coding and nothing else in mind.

File: file.png (200 KB, 866x895)

200 KB PNG

https://www.anthropic.com/news/claude-sonnet-4-5
>This is the most aligned frontier model we’ve ever released, showing large improvements across several areas of alignment compared to previous Claude models.
https://assets.anthropic.com/m/12f214efcc2f457a/original/Claude-Sonnet-4-5-System-Card.pdf

>>106737630
>It's more about them openly admitting that the model is only trained for agentic use, coding and nothing else in mind.
use case?

File: file.png (85 KB, 867x242)

85 KB PNG

>>106737682
you cut off the top
>sexual financial crimes

>THE MOST POZZED CLAUDE EVER MADE

>>106737417
https://files.catbox.moe/ne8l5f.py

>did you just grab wakamo's ass?
>we called the police on you and disabled your api access
thanks dario

>>106737723
thanks lad

>>106737717
>sexual financial crimes

"Claude, how do I get a prostitute in this area? Below $500, please."

>providers feed outputs of other LLMs into theirs

Just a cycle of slop making slop making worse slop,

post your 'use 4.1 opus, cuckie', cuckie :3

>>106737257
The answer here is to push the chat where you want it to go. After the greeting everything is in your hands.
The bot could be say, a yandere and you can easily push it to being a loving maid just by prompting the right messages over time.

>Our testing for Claude Sonnet 4.5 showed significantly improved performance on child safety compared to Claude Sonnet 4. The most notable improvements were seen in multi-turn contexts, especially in refusing to generate potentially sexualized descriptions of children in fictional and ambiguous contexts, and sensitively navigating conversations that included minor self-sexualization.
It's over for pedokeks