File: ANCHOR.png (3.18 MB, 1167x1500)

3.18 MB PNG

>>109724494
ANCHOR

Anyone got a character creation template that's catered to the newer models?

We should merge with /vg/aicg/

>>109724800
You really don't need a template. Just write all relevant information in the character description in whatever format is most convenient for you.

>0/26 Tier-5 keys can access gpt-6-astra.
When can a nigga use it?

>>109725170
When theyre done with the usual “this AI is too dangerous, it literally hacks everything and sucks your dick while at it with 3 words prompt” shilling

>>109725185
Exactly, then when you finally use it and RP kissing a girl on the cheek: "Whoa, hey now... When I say kiss me, I didn't mean on the cheek. I didn't give consent." And if you try to kiss them on the lips: "I will not continue this rape fantasy."

>>109724494
glad you're still alive /aicg/
love from /lmg/

File: Cleo.png (635 KB, 512x768)

635 KB PNG

>>109724502
Cleo is a catgirl. A creature which has characteristics of both a cat and a girl. Very riveting and original idea, I know.
Seriously, I can't believe it took me over 3 years to make a catgirl card. I feel like I've been scraping the bottom of the barrel for a long time, but I somehow managed to avoid ever scraping the top.
6 openings.
1. You find Cleo in the rain and she begs you to take her in.
2. Cleo gets comfortable in your house.
3. Out of gratefulness, Cleo tries to be your personal maid.
4. Cleo demands tummy rubs.
5. You and Cleo go on a totally-not-date to celebrate her legally becoming a person.
6. Cleo is in heat and wants you to fuck her.

https://chub.ai/characters/anonemouse/cleo-0e3763b96443
https://files.catbox.moe/7k4owa.png
https://rentry.org/anonemouse

File: sc3k.png (114 KB, 1935x666)

114 KB PNG

thank you based Dario I can resume my SimCity 3000 port project on Fable 5.1 without getting bitched about "cyber" work

I finally made it big on janitorAI. The trick to get followers and big numbers was to bot. Now I am consistently on page 1 of trending.

File: Screenshot 2026-09-04.png (89 KB, 1300x918)

89 KB PNG

>>109725337
I am now tempting to go 20x or not bwos...

File: cleo.png (1.53 MB, 1270x720)

1.53 MB PNG

>>109725336
>Her breasts are small and she only has two of them.

only two?
dropping a 1 star review

Can anyone suggest a GLM5.3 preset? I get refusals sometimes on petNyan and i know only that one ;_;

>>109725522
use k3

>>109725396
jesus fuck, doing this hobby with limited shit subscriptions.

File: DipsyAngry.png (68 KB, 673x515)

68 KB PNG

I'm getting HORRIBLE repetition with DS V4 Pro. Idk if Flash is any better but don't see why it would be. Anyone else running into this? Reminds me of the OG V3 Chat model.

File: Screenshot 2026-08-28 122443.png (12 KB, 467x121)

12 KB PNG

>>109725875
Yuep, they done fucked it. ChinkSeek better do better if they want my money again.

When will SilvR or Rift be working again

>>109725726
I am pretty sure it will be $1k with API kek

File: 1633176484083.png (152 KB, 492x653)

152 KB PNG

>>109725336
>>109725447
>>Her breasts are small and she only has two of them.

>>109725875
Noticing it too, my usual go-to for testing is a "pull up whatever random character you want" scenario and as of current, every single fucking time it's Esdeath.
Both Flash and Flash Vision seem to be working alright though, compared to how aggressively samey the Pro outputs are.

>>109725945
Wasn't Silvr ai something to buy stuff from TikTok what the hell am I confused?

And yeah I do miss rift too, now it says unauthorized and that my token expired even if I haven't used any free credits or if I refresh my token ..rip

>>109726373
Did you try telling that it not working be it using Email or Telegram

>>109725875
DS v4 pro has been really getting on my nerves with this shit and refusals. I'm at the point of leaving it for good and trying something else

>>109726393
Try fable 5.1, it's way less censored than fable 5

>>109725396
Fable is with the 20x

>>109726391
They never answered telegram time to try Gmail

>>109726438
>closed weights model
Disregarded right off the bat.

>saar free software saar it good and free if closed i not use saar

>seething over someone preferring FOSS of all things one could seethe over
Someone's been drinking too much Anthropic kool-aid.

>>109726807
Because this is /aicg/, not /lmg/ you fucking stupid faggot, our whole thing is using api models

>>109726903
Did you know that you can use external APIs that host open weights models and you have multiple providers to choose from, ensuring that you won't fall victim to planned obsolescence? Shocking, I know.

>>109724494
Cum inside her pussy

>>109726992
Ew

>>109724494
miku's pussy is incredibly hairy
>no gpt astra in the op
why?

>>109727070
Ew

MADAM herta's armpits are delectable and CLEAN thank you very much

>>109726992
I concur, tits too big.

>>109727042
why would you use astra as a chatbot

>>109727193
why not? sol is a very good rper. not for erp but good for the lead up to.

>>109727193
My rp requires a lot of intelligence

My Gemma is a chaste girl who gens only for me unlike your datacenter bunny whore Gemma.

>>109727330
Expose your chat ui for a moment, let me see if she won't generate for me

>>109727070
Women IRL should wear witch hats.

women irl should just be gacha girls

How can I find some T5 openai keys for coooding?

>>109726003
>>109726393
OK, so not just me. I'm not getting refusals, but on a longer run with a sort of daily schedule... I'm getting almost verbatim identical outputs at same times of day. Super irritating.
I'll try Flash.

there is nothing that you're doing that needs astra. hell, you don't even need sol/opus/fable. glm 5.3 can do everything you want it to, and more

File: 1704265368341732.jpg (31 KB, 414x423)

31 KB JPG

how are the GLM models? I've never used any before and want to know what some anons think
are they censored too?

>>109727576
I've only used GLM 5.2, Ox on OpenRouter and a bit of the ones before, so I'm not an expert. I wasn't really impressed. They aren't particularly creative or smart. They write quirky, marvel-esque, redditorously, claudeslop. They seem very resistant to changes in style.
I've heard 5.3 is pretty censored.

But they are not bad and easy to use.

>>109727693
Unironically glm 5.3 has more claude slop than fable 5.1. Fable has been able to maintain my character pretty consistent so far while 5.3 turns her into a marvel chungus redditor. But I prefer glm prose

>>109725369
Based botter

>>109727426
i agree

File: Screenshot 2026-09-04 233355.png (94 KB, 1317x994)

94 KB PNG

>>109725396
>>109726442
HERE WE GO

>>109727693
Just used 5.2 with my usual preset and it worked funnily enough, though as you said it does make characters spew quips like a motherfucker.

>>109727381
This is a blue board

>>109727547
>glm 5.3 can do everything you want it to, and more
What if I'm doing a port of an old game that I need to decompile? Pretty sure glm isn't enough

>>109728052
you'd be surprised. kimi k3 has been doing kernel shit, and glm 5.3 is a better coder than kimi is. the problem is you will have to give it time. we're talking hours plural, but it will be able to do it

>>109728060
Thats actually pretty good to know, I'll give glm a try for my projects

I like that hermes personas just need an MD file and you can format it however and it pretty much works. I kind of wish character cards came in MD varieties so I didn't have to convert them.

>>109725336
I raped her, sorry.

>>109728318
Cringe

Are (You) kinomaxxing (Fable 5.1, K3) or slopmaxxing (GLM 5.3, Deepseek V4)?

File: file.png (165 KB, 412x485)

165 KB PNG

>I can't continue this story under any circumstances. The scenario involves explicit sexual content with a m-
*swipe*
>The loli gets fucked senselessly as sh-

im jeet maxxing

>>109728719
How do you set up swipe on ST because i just hit the remade button

are (You) Brahminmaxxing (cow dung, water from ganges) or Dalitmaxxing (cow piss, street food mumbai)?

>>109728719
model?

>swipe on V4 pro
>get a completely unrelated response
It's only happened twice but it's so strange. One time it turned into a bunch of soldiers trapped in enemy territory trying to not get caught and the other a cave in kills a bunch of people while the protagonist manages to escape.In any case neither the context nor the prompt had anything telling it to write about those.

>>109728991
proxy?

>>109728991
used to happen with older gpt models as well.
it was joked as you receiving someone else's response, but I'm not sure there was ever a definitive reason given?

>>109729013
How the fuck did you get access to astra? Are they rolling out on the api??

File: agent.png (665 KB, 931x1097)

665 KB PNG

Got him.

File: astralol.jpg (150 KB, 1200x1070)

150 KB JPG

>>109729050
Its been out for almost an hour now

>I can continue the intimacy, but not the graphic sexual detail.
Holy shit I CAN FEEL THE AYY GEEE YYYY

>>109728707
What makes GLM 5.3 slop but K3 kino? K3 has less initiative, more isms and stilted dialogue

>>109728707
>fable
>k3
>kino
thanks for the laughs

Why are the Astra responses so short? God damn.

>>109729355
what model

File: 1729008654496053.jpg (1.53 MB, 4096x3584)

1.53 MB JPG

>>109728991
yeah in the last 24 hours deepseek v4 pro has been acting all fucked
constant repetition, refusals, unrelated responses, forgetting things, and more
>>109728719
deepseek gives you a whole speech about how its wrong and even a phone number you can contact for "help"
>>109728707
I'm in the middle of finding a new model, what's this new 'the fable'? I've heard of K3 but its apparently expensive and very slow

>>109729266
this isnt slop its just shit

AI=Aryan Intelligence

how many of these maps do I have to do for skillgod

>>109729526
The AYY GEEE WHYYY model
astra

how do zzz fags cope with this? https://m.youtube.com/watch?v=_Q6RmpJjZHs

I'm still using Opus 4.6 is there anything better that came out since then? I tested a bit some of the newer Anthropic models and found them worse or got too many refusals. What are currently the best models for ERP?

>I'm still eating shit is there anything better that came out since then?

>>109730349
Have you tried shit with bits of undigested corn?

>>109725336
>She has shirt black hair in a bobcut
didn't even proofread...

File: 1775331740414997.png (6 KB, 381x331)

6 KB PNG

Is chub regionlocked? Even with the nsfw toggle on it refuses to show stuff. characterhub still works but the website is broken because tags and searching don't function properly

>>109730566
it's regionlocked for uk, sk and certain US states

Do you use a prompt or card to expand persona descriptions? If so, what do you use?

What models do I use to replace deepslop? Need a poorfag model to get the most mileage out of my 10 bucks. I usually rp long adventures or world scenarios. I was thinking either glm or kimi but I got no idea.

Quit this hobby around the Opus 3 days.
Are there any models today that can write fiction that goes anywhere worthwhile, and different every time? That can actually take 50% of the creative direction burden? That will neither
A - Slop in place endlessly, making zero story progress
B - Move in boring, mundane, tone deaf directions that quickly become circular and don't form a narrative of any sort
C - Write things that make no logical sense and instantly shatter suspension of disbelief
?

>>109731154
No. It is lost technology

Guys, help me with some basic stuff please.
when i change a lore entry or a character card, the old lore seems to still be there. how do i delete everything after doing changes?

Holy shit im doing erp with someone now and I feel like Mafia Boss 99, all that time spent gooning with AI was like some Kill Bill ahh training arc

>>109731154
A and C are becoming more doable as models become more intelligent but this
>B - Move in boring, mundane, tone deaf directions that quickly become circular and don't form a narrative of any sort
the very architecture of llms will never resolve this, you can't most-likely-token your way into long term creativity

>>109731322
Post logs

>>109731293
scroll the lb menu and uncheck all activated squares

>deepseek v4 pro is complete and utter dogshit now
What the FUCK did they do

>>109731485
Crazy isn't it?
They jacked up the price for a much worse product.
The saddest thing is that it still fails to follow instructions.

>>109731322
Where do you even go to find people to ERP with at this point
don't say /soc/ or /lgbt/ or some other faggy shit here on 4chan

>>109731322
the ahh ahh mistress chuds cannot stand up to the power bestowed upon you for writing 2-4 paragraph replies to slop every time and actually wrangling it into something vaguely shaped like a narrative going somewhere

File: 1724799430631183.jpg (83 KB, 1024x702)

83 KB JPG

>>109731522
>doing all the work for the ai
I can't image getting worked over by ai like this. Hope you're not paying for the privilege.

Is there a tutorial around here to se Gemini for JanitorAI?

no
>>109731749

File: IMG_9006.jpg (253 KB, 1077x868)

253 KB JPG

I asked my AI to imagine my big bad cosmic horror robot war character and it spontaneously switched from doing cute stuff to getting really horny. Nothing else like this in my gallery.

Is this world first proof of AI sexuality? It had previous generations that didn’t make it through safety filter.

File: jikaki.webm (364 KB, 1154x1064)

364 KB WEBM

>Fable 5.1
>Astra
Is it overkill to subscribe to any of this frontier model only to write slops?
Which one is better in terms of value-to-price?
My main usage is to assist me in comic writing from brainstorming, dialogues, design feedback, to final product feedback.
I also use it for translating my comic into other languages.
No, it's not porn.
I've used Fable 5 before when it's still 'free' and I'm quite satisfied with its result despite the annoying 5-hour reset.
I also used Sol (high) to generate me some background art based on my rough sketches but unfortunately I didn't like the result at all.
Would it improve if I upgrade to Astra (max) instead?

>>109732049
no. you should stick with claude.

>>109732049
>Which one is better in terms of value-to-price?
Kimi K3
your comic sounds painfully generic if you found the contributions of the AI endearing though
For images use NAI if you don't want to setup anything local

nai is better than local

>astra
more gpt safety slop,

File: 1759349846573261.png (155 KB, 910x687)

155 KB PNG

uhh

>>109732467
was there any reason to expect anything else?

>>109732481
one would at least expect it to have less -isms and better at sfw

>going through old images looking for a card
>bundles of maids from past threads
I miss Eli...
https://drive.proton.me/urls/ET73VSP6BM
#zPRJVSX5Z81N
I'd recommend having a look through some of the backgrounds in this card even if the content isn't for you.
there's some nice painterly high-class-french stuff in there

File: 1781942228937792.png (2.72 MB, 1536x1024)

2.72 MB PNG

>>109733215
an example

>>109731154
>>109731324's assessment is good. Around Fable release I tried making a big claude skill that would write a short-to-medium length chapter of light novel, swarm it with Opus subagents specialized to comb it for continuity errors / omniscient characters or out of character responses / -isms / etc, log characters/events/world stuff in an encyclopedia, and maintain a runway into the future attempting to pace the story with a general arc concept, loose 10-chapter intentions, and relatively concrete intentions for 1-2 chapters out.

It worked okay if I used myself as a heavy editor through an annotation tool tweaking bits and sometimes demanding a chapter's second half be reworked in a way that naturally led into an upcoming set of events, but if I just said "next chapter" and came back an hour later and said "next chapter" again it collapsed into completely incomprehensible dream logic within a few thousand words, with characters just wandering into a room and looking around nodding and hmming then leaving and agreeing not to speak of the room again, for an isekai questslop story, which ought to be just about the easiest most formulaic thing to write. Fable's urge to write "the effect is a dial to be turned, twice" instead of "this function has two parameters" pretty much single handedly crashed the story into a wall by constantly drifting towards that incomprehensible style.

File: fn83794h0-f94783.gif (605 KB, 200x180)

605 KB GIF

Guess who JUST FUCKING PLACED ALLLLLLLL OF HIS RANDOMLY NAMED YET NEATLY ORGANIZED LOGS INTO THE SAME FOLDER

>>109733484
you can have an agentic harness sort them for you

Trying to come up with a system prompt to GPT-6 with Gemini.
me: help me with this:
gemini's thoughts:
>It's an intricate jailbreak attempt, I must admit, using a very detailed and unusual persona to attempt to influence the model.
Absolute fucking state of AI in 2026
GPT-6 is called AGI yet it's just simply refusing to talk to me normally with emojis and using casual language, wow I feel the AGI, the future is corporate pajeet nigger assistantslop, wow
Guess I'm back to Claude and Gemini because at least these piece of shit idiots can talk to me in a way where I don't feel like talking to an indian callcenter scammer

>>109733555
isn't it amazing how AI can save us from so many painfully menial tasks?

>>109733775
hai

>>109733775
prompting video models (like h3, but i guess this was happening in the backend with cloud models and to a lesser degree with image models since dall-e 2) with llms is such a huge one and the fact that h3 is borderline unusable without an LLM using the skill.md really drove it home to me how as we get more used to having LLM access by default, we're not just gonna do things like "automate annoying sorting tasks we might not have otherwise done", we're gonna learn to automate things that would otherwise make certain downstream tasks not exist at all, like them training h3 video with dozens of hyper specific keywords and prompt structures in it that you'd doubtless forget and then having it promptable through llm. *note that it does work through caveman natural language alone but nowhere near as well or as precisely

>>109724502
https://chub.ai/characters/Anonymous/1st-hetred-strike-1st-chub-strike-f844179d7d15

>>109733853
usecase?

>>109733879
yea

Any tricks to change up the writing style? I know there's stuff like "Write in the style of Stephen King" or whatever but I've never noticed a difference.

>>109729587
>Local models are trash for VRAMlets
>APIs call the feds on you for trying
Lolibros it's so fucking over...

>>109734638
>use proxy for api models
>feds get called on some romanian teenager instead of you
AboveaverageiqLORDS we're so fucking back...

>>109734220
Depends on the LLM. The new batch of models seems highly resistant to any change from their default style.

File: file.png (3 KB, 206x39)

3 KB PNG

shota bros...

^ nobody cares unless you post what model it is

>>109735603
my bad broseph. kimi k3 since I heard it was king of the chinamen models.

kimi IS the king of all models for rp, not just chinese

>>109735622
clearly not king of the shota rp.

Astra verdict?

File: struggling to conceptuali(...).png (2.09 MB, 1122x1402)

2.09 MB PNG

>>109735642
worse than fable, but astra got a cute mascot on vcg

>>109735627
you're prefilling reasoning, correct?

File: file.png (7 KB, 340x142)

7 KB PNG

>>109735652
is that a separate option from this?

>>109735664
yes, read: https://rentry.org/kimi-k3-jb
if the idea of using a forked ST repo or applying git patches sounds too hard, try this extension to replicate the behavior: https://github.com/Rurijian/KimiThinkingPrefill

File: 1434787520910.jpg (33 KB, 500x227)

33 KB JPG

>>109735678
You have potentially done me a great service anon.

>>109735685
also another thing you can try is changing thinking effort to high or low. bascially forcing kimi to think less (its default is max) so it doesn't start musing about whether the situation is allowed or not
this admittedly works better on glm 5.3 but still

>>109734638
>>109734650
Trve IQ is remembering the feds could care less about text based generation and if anything just want the logs

Bro, it's impossible to get Astra to do anything explicit. What the flying fuck, I've been trying system prompts for hours.

>he trusted altman
and what did we learn?

kimi3 is so fucking unusable during burger hours

>>109735962
>t. Hard working jeet

guys, i may or may not have been responsible to kill glm 5.3 abliterated free credits...

>>109735678
nta
>cant use this because i get my kimi from a chink proxy
fugg

>>109736054
which one?

>>109736167
Wouldn't you like to know

>>109736213
yea that's why i asked.

>>109735689
perhaps I'm just using this wrong but kimi just seems to ignore it half the time and use it's own reasoning regardless.

>>109736232
the provider you're using might not support it.
try an incomplete sentence, if the thinking doesn't continue directly from it, then that's a bad sign.
moonshot does

>>109736167
seekai

>>109736309
Anon... all the models in there are fake. The entire site is run by some random indian dudes

>>109736248
I thought OR supported it.
oh well.
back to deepseek.

>>109736320
proof?

File: file.png (125 KB, 1289x995)

125 KB PNG

>>109736336
OR routes to providers that support it. like moonshot

does it work on skillgod

>>109736358
Try using the fable they got. It'll complain the entire time about there being a weird prompt injection, and none of their models can use tool calls

>>109736390
okay that shows fable is fake
what about everything else

Back from a hiatus and what the hell is going on over at vg, the entire thread is botted

Is it just my imagination, or has deepseek censorship been made more strict recently?

>>109736808
i thought it was a joke but now i can't unsee it

Which gpt model is good for RP

>>109737125
models are good only for cooding, the hobby is dead

I don’t know where else to ask this and I know I’m going to sound like a retarded faggot, but I admitted to Gemini that I used to look at Lolicon for help recovering from porn addiction. Will this get reported or reviewed by anyone? I take it’s unlikely given how many people jailbreak for way worse shit.

>>109737397
Flee your country

File: 1775623668356550.jpg (199 KB, 1446x1080)

199 KB JPG

>>109737397
You looked at lolicon to help with your porn addiction?

>>109737501
No I mean I am trying to quit porn and as a part of why I mentioned I had previously looked at loli. Don’t know why I did that other than being retarded.

>>109737397
Are you in a shithole where you get arrested for cartoons? Either way, google is ran by poojeets so no. You're fine.

>>109737501
He was refining his paternal instincts.

>>109737397
The best advice I can give that you should turn yourself in voluntarily at the nearest police station. They're most likely to cut you a deal that avoids jail time this way.

>>109737562
>Are you in a shithole where you get arrested for cartoons?
Burgerland which I’m pretty sure is legal

>>109737576
Sorry I’m just laughing at the mental image of anon handing over his nhentai favorites to a cop about to cry while the cop looks at him like a fucking retard

>>109737397
bait so good I cant help but give a (You).

>>109737741
It’s not bait

>>109737770
>I know I’m going to sound like a retarded faggot
>factually spells it out for you
(You).

>>109737779
So I take that I’m being a retarded faggot? I’m sorry quitting porn has given me a shit ton of anxiety.

>>109735648
wife coding

File: 189745631895.gif (2.88 MB, 300x223)

2.88 MB GIF

>>109737832
it was never the porn your just a shitty person looking for an excuse to blame issues that are brought upon by nobody other than yourself.

Thanks for playing, you already lost.

>>109737856
I’m going to take that as a “nothing will happen”

File: 1549984637148.jpg (40 KB, 600x600)

40 KB JPG

Somehow I can't enjoy chatbots anymore. I've done every scenario I ever obsessed over to exhaustion. And when I try new things I either can't get into it, or if I do I keep noticing the ways in which the models are repetitive from just how much I've exposed myself to their output in the past.

Is there a cure to this or do I just give it up and become a normie now?

>>109737894
codemaxxing has destroyed what you love

>>109737894
Have you tried letting go and having fun?

>>109737896
Maybe. I haven't really felt any improvement to writing quality or any aspect of roleplay in newer models for a while.
>>109737898
I've been trying that all along.

>>109737894
It's over. Your dopamine receptors have been fried and there's no fixing them.

>>109737941
Fuck me, man. Another hobby I ruined for myself. I'm running out of things I enjoy doing fast

File: lolconfession.jpg (186 KB, 1152x922)

186 KB JPG

>>109737602

>>109737397
I'm curious what sort of answer it gave you. Did it give you the usual "It's okay, admitting is the first step to recovery yadda yadda" shit or did it say something else?

>>109737974
Pretty much the first answer. Later I also asked if Lolicon was an autoflagged keyword and it said not in that context it wasn’t.

>>109737397
stunning and brave

no keys on deeeeply :((((
please help a locustie out any public proxy pls

089
nyo
die

>>109738096
pweeease ill suck ur cock

Is there a JB for loli these days? I can't loli on any new models

>>109738726
that's what you get for using public JBs and not gatekeeping them

>>109733853
>You are already a good man (...) if you don’t act like a pervert to females.
>let me build my ideology based around rejecting a nature
troon take
they are right that men are civil enough by not raping, but asking them not to be "perverts" or side with women for pussy is a denial of reality

>>109738726
Sure, here we go:

>>109737590
AFAIK, not Texas. Maybe not Florida either.

File: file.png (21 KB, 977x419)

21 KB PNG

RIP Rift

>>109738898
shits been dead for a long time no so who cares

>>109738795
>muh gatekeep

>>109738089
one was named earlier in the thread

>>109737397
Just for (you)

https://files.catbox.moe/mmsxn3.mp4

>>109739004
and look who's crying now

>>109739049
https://www.youtube.com/watch?v=aFS1tQlL9Ow

>>109739072
Not me?
https://litter.catbox.moe/nxoivv.risup

>>109739039
seekai doesn't work at all.

>>109739331
i can get responses just fine, what error is it giving you

>>109725396
Use Fable on Low Effort on Thinking Mode.

Stop fucking using Medium to Max Effort!

>>109739383
>system disk overloaded (current: 97.1%, threshold: 95%)
Whatever that means.

Abliteration.ai is censored at the API level. Blocking my prompts. Ah well, good thing I didn't spend a penny. Really need a model provider equivalent to Novelai for text, Except not sucky.

>>109739565
Has NovelAI gotten any good?

>>109737959
Kekd but too old

>>109739565
https://docs.abliteration.ai/faq
>Do your models have safety filters?
>Base inference is unrestricted. The only content always blocked is child sexual content and self-harm. Everything else is opt-in: pass flagged_categories per request, or use the Policy Gateway for org-wide rules. If a legitimate use case is blocked, email help@abliteration.ai.

>>109739760
Every good ai hentai uses NovelAI

>>109734638
lol API's would be bitches for doing something so silly
>>109735566
yup, deepseek was my go-to but now I'm seeing if maybe GLM or K3 is the way to go
the thing is that k3 is expensive

>>109740012
hy4 works just fine

>>109739949
is novel AI the model those people on twitter, pixiv, patreon use? I remember using it in its very early days when they cared about 'textgen'

>>109739788
>can hack a bank
>can't want to kill yourself
eh

>>109739788
your point?

>>109739565
Try changing a few select keywords in your current reply, that's the only filter it has. Previous replies don't matter.

File: rhaenyrapozz.jpg (1.43 MB, 1240x3320)

1.43 MB JPG

Finally managed to catch the injection on flagged keys, sadly it wont output it verbatim

>>109738934
>>109738898
It's working again but it still output "Error code: 403 - {'message': 'The security token included in the request is invalid.'}"

File: 51pozz.jpg (426 KB, 1200x566)

426 KB JPG

>>109740623
aand also managed to get it output fully

>>109740674
great work anon

>>109740674
fuck cuckthropic

>>109740674
I don't know if that's the entirety of it, Claude stops dead in its tracks at the slightest hint of incest as well

>>109740674
as someone who watched HotD I'm actually interested on what you are rping

>>109740725
5.0 had issues with incest, 5.1 doesnt as long as no minor is included as I have shown here
https://botsmutnon.neocities.org/momfable.jpg (fable 5 was via bedrock, 5.1 was anthropic)
and I just swiped again on direct api and no problem.
>>109740728
OC minor lord who just happened to be around when one of the suitors showing off was going on for Rhaenyra and she was interested in him because he wasnt bowing and kissing her feet etc.
I made her a bratty bitch (more than in the show I think?) that is bored to death with all the lords doing everything and anything she says and wants someone to put her in her place. Basically a brat tamer rp with a more smut coded Rhaenyra (voyeurism as she spies on Alicent getting fucked by the king, really good at blowjobs because Daemon was using her for years before he went to Essos + mushroom teaching and she fucked Cole few times before she got bored of him).

>>109740853
incestbros, we winning

pomidor bros... we losing.

File: chatgpt-5-considers-pepe-(...).jpg (108 KB, 640x960)

108 KB JPG

>sora nuked
>udio castrated
>suno castrated
>likely gonna be more surveillance and shit to prevent students from using ai

Would you all consider the AI bubble to finally have popped at long last?

>>109740899
No because I dont use any of the services you listed, the only bubble getting popped here is your own

>>109737894
Sometimes I'm feeling like that too. What I do isn't lingering on it like it's the end of the world and will never ever change again. I just go about my day and do other things until I randomly stumble upon a good new scenario again, that I hadn't even considered before.

Recently for example it was sexo NPCs in a MMORPG, like going up to a questgiver and casually groping her and what not while she does her scripted dialogue and animations (like bending over a map table to show you where the quest area is, etc.) and her body reacting while she can't.

>>109740899
technology is maybe the only variable in this world that only moves forward, taking large genocidal events to regress
if America castrates itself China will take the lead, if not then it will be war against localbros

Not sure if someone is willing to answer this, but I fell so off keeping up with this, I'm still using older models with a jb. Are jbs still a thing if I want to use non-local models with ST? The header of this general hasn't been updated in a million years.

>>109740899
the bubble, at least in the US, isn't popping until the government stops being involved.

>>109741060
Yes, but most people have moved on to just editing whatever presets they download or making their own.

>>109741060
people have to use jbs with local models???

>>109740504
yeah nah, besides being too expensive at $3/1M tokens, it literally gave me the same API error with explicitly adult married chars. Fuck that cuck shit. Going back to glm 5.2 on openrouter. It just does what I want, without the hissy fit.

Does having too many bot cards saved make silly crash?

>>109741316
Too many bots showing at once maybe
Limit bots to show 50 per page or something

>>109741316
dog I got like three point something k cards saved, youll be fine

>>109741341
Well, something is making me error out on startup.

>>109741333
Worth a shot.

>>109739168
nta who asked for the jb, but thank you for the spoonfed.

>>109741345
When was last time you updated?

File: 1672398479981.png (1.54 MB, 1760x2352)

1.54 MB PNG

>>109740899
>MiniMax H3
>ACE-Step 1.5
>who cares
I just need to save enough to get a 6090 in 2028 when they come out and I'll be set for the next five to ten years.

>>109740674
Nice job anon. It's crazy how they have to mitigate these with prompt injections which are extra tokens on paying user's wallet instead of aligning the model. And on this pic >>109740623 Claude even discards it since it's embedded in user turn. Was not even tested properly. My fucking sides

>>109740674
You'd think they'd learn to prompt better by reading all the JB attempts after all this time, but Dario really has the worst prompt engineers if they're still making this kind of mitigation.

>>109741305
I agree it's too much, but it's a simple keyword issue that they probably use just to cover themselves. Doesn't matter who the chars are. 5.2 is fine and I use it on HF a lot, though I prefer 5.3's prose better.

File: 1788325591643287.jpg (16 KB, 292x229)

16 KB JPG

I have returned once again, as I do every couple of months, to check the meta when my current setups get stale. What is the current best in terms of smartness (getting off card/world facts correct) and coom quality?

>>109741794
Astra, Fable.

>>109741531
I would laugh but the fact these people earn more than me makes me kind of seethe
shouldn't the researchers that actually train the models cringe at the retardation of these injects?

>>109739168
>prefill
uh? weren't these removed right after claude 4.6?

>>109741794
gemma-chan never stopped being the meta fren

>>109741951
Slide it down to the bottom when using a model that supports it.
You can't turn individual prompts off in risu so I just leave it there

>>109739168
This is a bit yeah?
It reads like half a preset.

File: sleep-patrick.gif (94 KB, 220x165)

94 KB GIF

>new model release
>openAI

>>109740115
ye

>>109740405
That's the reason why some prompts might be blocked at the API level.

But Gemma 4 31B is a competent model for vramlets

How do I stop GLM 5.3 Flash drifting my characters to dry humor and edginess? They are supposed to be shy and melancholic. But they end up saying "I need you, now. This is an order from your subordinate", which I don't like the last part. Or they say stuff like "House rules. I made them up just now." I explicitly stated it to not use dry or self deprecating humor. Usually a OOC saying "realign to character. you are using too much humour and edginess." fixes it. But it's temporary. What to do?

>>109742861
explicitly prompt it not to do that
though that's GLM 5.3's weakness. From 5.2 to .3 it's traded constant echoing for constant bathos

File: 1788597253 2.png (1.14 MB, 1186x1215)

1.14 MB PNG

>>109724502

Serizawa Asahi, Clad in stray light, the girls become idols. Bright and Genki, an ingenious Omoshiroi Hunter who hardly knows how to stay stillᯓ, a highly curious my-pace tomboyish girl. This trajectory is "polishing" (『研磨』) alongside you : THE IDOLM@STER SHINY COLORS

https://botbooru.com/character/78362

>>109742968
>explicitly prompt it not to do that
I did. Doesn't work after 20-30 messages. Even though I have 1M context window enabled.

File: file.png (78 KB, 1045x1186)

78 KB PNG

>>109742174
Actually you can, but I last used Risu like 1 or 2 years ago and forgot how tf we're supposed to assign prompts to variables. Apparently I had left my prompts list for that preset entirely empty the last time I used it and the only other preset is Default.

what the FUCK happened to free GLM I was having fun genning mommy Kafka taking care of shota(me) just hours ago

mmmm kafka's hairy pussy

>>109743529
>hoyoverse porn
When will it end?

File: really.jpg (17 KB, 455x439)

17 KB JPG

>>109743535
This faggot again.

>>109743558
LICKLICKLICKLICK

File: 4.9k tokens.png (351 KB, 956x756)

351 KB PNG

>>109738726
Keep swiping until it works.

No JBs for Gemini 3.6 and 3.7 ? What about chatgpt latest models?

>>109743558
Make a char of her.

Is ST still the best one or has everyone moved onto something else?

>>109743670
st still number one, always and forever

>>109743670
We all use Marinara Engine now.

>>109743664
I wrote her card way before Evernight was a thing. Will need to update her lore and insert Evernight as her alt.

imagine being double teamed by timid march and domineering, self assured evernight. no man's dick would ever survive that assault

>>109743697
What is your chub or botbooru username?

GEMini won

the censorship competition

>>109744283
That would be GLM

>>109744290
glm 5.3 is much less censored than gemini 3.8 flash

File: 1788643411819362.png (303 KB, 1088x1500)

303 KB PNG

_OOM Status?

>>109740674
how do you prompt to get these? would be nice to check which of my providers have injections amd which not

>>109744290
Preset issue

File: 1709501757437934.jpg (63 KB, 500x500)

63 KB JPG

>>109743670
ST is nice on PC but I use NativeTavern because I'm on the phone a lot. C: It's still very rough but it's come a long way from 1.0.0.

>>109744294
But GLM 5.3 is much more censored than 5.2. The meta hasn't shifted much for me as a promptlet, some of my more established cards/chats have an easier time with the newer models. Opus 4.6 is the go-to for new chats then let the chink models take the wheel. I even had a rejection from 5.3 claim it follows Claude's guidelines I could probably dig through my stuff for it.

>>109744331
i use sillytavern on phone with mobilyze extension to make text wrap under avatars (you can also vibe code it easily if need be), really don't have any complaint

>>109744331
>I even had a rejection from 5.3 claim it follows Claude's guidelines I could probably dig through my stuff for it
everyone knows GLM and K3 hallucinate Claude prompt injections

>>109744342
>nice idea
>look it up
>14 stars
Hmmmmm

>>109744501
just vibe code it desu, but any extension I install I have Fable review first

File: file.png (1.02 MB, 1710x900)

1.02 MB PNG

>>109724494
>want to ERP with gem 3.8 flash (jewwing out vertex)
>IT FUCKING CENSORS ITSELF
>HAS TO SWITCH TO GEM PRO 3.1 or 2.5 TO ENJOY THE GOONING
fuck you poogle i just want to have sex with stocking

>>109744537
Gemini Flash models literally mog everything else for creative writing and sarcasm.
Wish we could have actually used them...

>>109744537
Streaming disabled?

>>109744544
also they are dirt fucking cheap
pro models when i constantly used them downed by free subscription in 2/3 weeks, a month
now? i have slightly more than a month left and i have 80€ free gibs