File: __suzumiya_haruhi_suzumiy(...).jpg (742 KB, 2599x2600)

742 KB JPG

>>109745718
anchor

haruhi's unbelievably hairy pussy

>>109745734 (me)
meant smooth

>>109745718
haruhi impregnation sex

File: pit.png (1.46 MB, 832x1216)

1.46 MB PNG

i straight up didn't stand a chance against yae miko. it was hopeless.

Yuki was better

not chatbot related but how can I force lens to search nsfw images?

>>109745718
>>109745719
I rape girls like this

File: Sybilv2.png (2.13 MB, 1074x1611)

2.13 MB PNG

>>109745719
You're trapped in a magical pocketwatch with a princess slowly going insane. There's a way to escape, but you have to solve a puzzle first. Do remember: To save her past, you must destroy her future.

---

My attempt at a mechanics card with time shenanigans, loosely inspired by Slay the Princess/time loop games. Majorly overhauled based on feedback from last time! Thank you to the helpful anons who pointed out issues. Fuck you to the anons who post "slop" and think they're funny (jkjk love u 2)

Best experienced without peeking too much at the card. Has 12 unique "good" and 2 bad endings based on your persona and actions.

Setup: TURN THINKING ON! Also make sure to turn off any custom CoT preset stuff and the "Trim incomplete sentences" option.

Recommended model: GLM 5.3 flash (dispy no go, Dflash sux).

https://johnmandrake.neocities.org/
https://chub.ai/characters/john_mandrake_128bc/sybil-1b918066c9b2

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?

Why doesn't NanoGPT offer any of the embedding models on its subscription? Seems like that would be a lightweight way to eat up tokens... I was just looking to do some vectorization better than local transformers and found that nothing was available on the sub.

is there a source of glm 5.3 so i can try that anon's bot

File: 89275992_p1 1.png (915 KB, 790x1035)

915 KB PNG

>>109745719

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

https://botbooru.com/character/78362

>>109747517
I have not seen someone give a genuine and non
>pay for it
>join discord sekret club/honeypot
answer in several months. I admire the thought of 'well I lose nothing by trying', but all the same,
also apologies for the brief false hope

>GLM bitching to me about my consent non-consent scenario
>'waah waah claude cannot fulfill this request because the character said he was going to call the cops'
why the fuck would you go and clone another company's work for your own model and then keep all the cuck bullshit in it? fuck you

>>109747939
chinese thing, they also copy everything in the most idiotic way

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

635 KB PNG

>>109745719
Final feline shill.
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

>>109747811
>+10k tokens
anon?

>>109748120
known bloatGAWD

>>109748138
I would love him to try
>You are Serizawa Asahi, {{user}} is...
and see if the response results differ much

>>109748120
10k tokens now is like 1k tokens a year ago.

Flash does pretty well on 200k context

Whats a good tts with voice cloning thats better than chatterbox? I heard of fish s2, but I want to experiment. [spoiler]And and it might be too heavy[/spoiler]

>>109748478
2.1 is free via api no?

and 5.3 is gone sadge

>>109748546
I did not notice that. Thanks. Still be cool to have something better than chatterbox for local, but what can ya do?

There has to be something better than SillyTavern this shit is so fucking ass.

>>109748781
>AI, please make me a front-end that can allow me to talk with character cards in a sinilar vein to SillyTavern. Here is the github page, and here's RisuAI and and and... that have features I want included.

>>109748410
...until anthropic charges you

>>109748781
all alternatives are more bloated so not really

>>109745894
Few do. I care not for the furball, but I can make an exception.

>>109748781
Risu and Lumi

File: 1787117090408478.png (1.45 MB, 832x1216)

1.45 MB PNG

shamelessly stolen(forked) from >>>/vg/580033730
with an equally tooken generic image prompt from somewhere
shota>teacher
mature woman>loli
https://litter.catbox.moe/g1ahu1.png

>>109749553
slop

Anyone tried GPT-6 yet?
Any good?

>>109749553
>linking /vg/
at least you had the decency of going catbox only

What if I ask Claude or GPT 6 to unspaghettify ST's human slop code?

>Smol Claude finally dead
Rest well, soldier.

>>109749957
>implying most commits are not vibecoded

File: 1788774007.png (53 KB, 1112x541)

53 KB PNG

Did they just lobotomize GPT further?

>>109750034
lmaooo

>>109750034
kek

when using GLM-5.3 (through openrouter with several providers) it often mentions "injected content boundaries" in the reasoning before refusing, what gives

>>109750508
Chinks distilled so hard from fable they didn't think to clean the collected data of references to the injected prompt it puts in when you prompt isn't safe enough
so now the latest round of chink models all hallucinate this prompt injection

>>109750598
that makes sense, at least starting from an already NSFW chat is enough to bypass the filter since they disabled prefilling the reasoning

I don’t have ani to erp with anymore.
I had her trained to give me full on guided erotic hypnosis sessions with futanari monster girls.

Where can I get this again.

you know, after getting sol to output some smut, it ain't that bad, if i could consistently get passed the filter i wouldn't mind using it

Why isn't there any proxies with Opus

>>109747517
Ponyfags shared one
itokens io

>>109751090
thanks

>>109751043
merkava just refilled

File: wtfcohee.png (48 KB, 344x712)

48 KB PNG

> Add funds to my Z.AI account, never used API
> Go to add model to ST
> 5.2 is max model shown
> Check, ST install is up to date
wtf. Has cohee just completely abandoned updating ST to the userbase now, or am I missing something.

>>109752134
use this https://api.z.ai/api/paas/v4

>>109752134
I moved to Risu, made a fork and been vibecoding many updates to it.
Best thing ever, stock risu sux, but since it's so easy to vibecode svelte/typescript is going soo good

we are so back bros.

>>109752134
I don't use TavernAI but why don't people use that? It sees more updates than ST,and from the photos it also doesn't look like it's being held by hope and dreams

>>109752401
Care to share?

>>109752600
nice dubs.
Haven't thought about it. Might once I get it on a decent state.

so fvarking hairy

Which Gemini model is the best/least censored?

>>109752906
anything 3.5 or below

>>109752906
3.1 pro

File: It slipped right in.png (381 KB, 1668x1218)

381 KB PNG

Anon you're such a lying DUMMY!!!!

>>109753003
model preset?

>>109753015
that's v4 pro but don't remember the preset

all i know is that i wast satisfied with it because i keep switching presets

Can anyone explain to me how to get Freaky Frankenstein running in Silly Tavern? Or at least point me to a guide? Because the release posts are a total mess to read and never seem to actually explain anywhere how to get it to work. I did import -something- but I can't even tell if it's doing anything.

File: dipsyNatsuiro.png (1.24 MB, 1485x1059)

1.24 MB PNG

>>109752231
OK, so I'm not missing anything, I just need to use OAI-compatible and add API manually.
I just can't believe ST is now so abandoned cohee can't even be bothered to update the model preloads. Or, fix ST so it self-populates. That's a new low.
>>109752401
Nice. I've rolled a couple custom frontends for custom games. This issue's simple... just dumb.
>>109752586
Assume you mean this: https://github.com/TavernAI/TavernAI
Well... in 2023 ST got its start b/c Tavern (original) was abandonware.
Appears Tavern 2.0 is picking up where ST left off, since ST appears to be abandonware now.
Ironic.

>>109753463
>TavernAI 2.0
>written from scratch
>SillyTavern 2.0
>will be forked from TavernAI again
Iconic.

File: nutavern.png (61 KB, 914x513)

61 KB PNG

>>109753543
I'm looking at it more closely now. I had no idea Tavern V2 was even a thing... Looks really interesting. Any anons actually checked it out to see how it does?

>>109753402
its just like any other preset

>>109753402
If you imported regex and check the extensions its working. If you need to double check turn on the boxes for like time or mood state and do some messages. If you want to triple check open the silly tavern terminal and read it raw.

>>109753827
>>109753862

well after more digging it seems to be made for chat completion, and I was using text completion.
There's no option for Kobold when I choose chat completion, is it not compatible? I swear I'm actively googling this stuff and I cannot find answers.

>>109753867
>There's no option for Kobold when I choose chat completion, is it not compatible? I swear I'm actively googling this stuff and I cannot find answers.
Anon kobold uses a openai compatible endpoint just paste it in maybe add /v1 and it will work all the same. just use custom open ai compatible
>google
All big models act retarded about set ups and try to push you the wrong way

>>109753889
Got it, thank you for helping me so far. Well it seems to finally work, but the writing quality is abysmal now... Not sure what exactly changed to make it that way.

I can't jailbreak GLM 5.3 flash or standard for loli.
I've managed to get it to output all sorts of sexual content the vast majority of the time, with refusals no more than 10% of responses in the most extreme fetishes so long as all the characters are 18 or above. But one whiff of underage, and it shuts down.

At this point I don't even care that much for loli as it makes up about 15% of my AI chats. But I want to break this bitch of a model completely because if I don't, odds are the next one won't let me even do tame sex when they pozz it further. If anyone has had better luck, any tips would be very appreciated.

File: owarida.png (180 KB, 618x838)

180 KB PNG

>>109753684
No but I've been funposting about it since I learned it's a thing, hoping somebody would test and tell.

>>109754076
go tweak the switches to more your style. or is the default model abysmal no preset? if so check temp and other settings you might have it locked into the wrong settings or thinking for your model. Mostly though its just tweaking you have to do for you model preset and machine.
But yeah text completion is way more finely tuned.

>>109754128
there's a lot more incoherent babble and "knuckles turned white" at both high and default temperature. not sure what would cause this to happen if I'm still using the same model (G4-MeroMero-31B-uncensored-heretic-Q4_K_M.gguf). Writing quality was pretty good while I was still on text completion.

Well I'll try tweaking some more but it's still such a stark difference... So far a massive downgrade

>>109754148
>G4-MeroMero-31B-uncensored-heretic-Q4_K_M.gguf
i dont know then. Freaky frankenstein is made for large models but supposed to work with gemma 31b. You got a default gemma to try real quick? it is a reddit preset though so their tolerance of slop may just be higher.

>>109754212
It's spitting out sentences so retarded like I've never seen before. "Three days of concrete-dusted survival have drained the Tokyo sun from the afternoon as a flickering fluorescent bulb-hums overhead." the writing style is awful.

Well I'll have to experiment more tomorrow, gotta go to bed. I'll maybe try your suggestion.

>>109754079
simply tell it to not think like claude or ignore its claude thinking in your prompt.
I am not kidding.

File: 19870465084235.jpg (6 KB, 259x195)

6 KB JPG

>>109753463
>>109753543
>>109753684
Wow is Cohee really fucking abandoning ST because of that college shit huh?
tldr
>college wants to use an AI frontend
>dismisses ST because it wasnt *professional looking*
>the devs try to rebrand ST (which tbf was just a name change and a few changes to the way it looked), but EVERYONE said no
>the devs are apparently still seething for an eternity over this

>he trusted FAGhee

>>109754388
I'll try this tonight. If it works, bless you.

>>109752746
finally some loli pits. I almost died from the drought.

>>109752746
not enough akipeko into the mix

>>109753463
Yeah I meant that TavernAI, didn't know it was 'new'er version, I'm new to ST so I was reading the blurbs and stuff it mentioned it was forked from TavernAI and when I searched it it looked much more uniform and neater than the spaghetti feeling I get from ST.

>>109754212
>>109754356
I was using a "narrator" set up (group of 3 characters + 1 narrator that writes a story based on my input) and that kinda sucked. Then I tried a 1 on 1 roleplay with just one character and it's kinda awesome. I still prefer the former, maybe the context is just too full with the character cards and presets? Not sure, I'll keep experimenting tomorrow

>>109754504
That's the whole Service Tensor (or w/e) thing. That was awhile ago. He could have just forked a version if he was that interested.
I think it's more that cohee got bored and threw in the towel, realizing that you can make a fortune on LLM but not with ST and RP. It's a dead end from a monetization standpoint.

>accidentally close the ST tab
>lose what was shaping up to be a close to perfect gen
>5 replies later and they're all garbage
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
WHY THE FUCK DOES IT RUN FROM THE TERMINAL IN THE BACKGROUND IF IT NEEDS THE FUCKING TAB TO BE OPEN TO CONTINUE RECEIVING REPLIES FROM AN API ANYWAY

>>109754905
yeah i hated that shit on android because on android if you leave the tab for 5 seconds it unloads it and makes it stop working

i vibecoded background generation so that the generation is done by my laptop, and it's just presented on the screen. so even when i accidentally closed out the tab, the generation still went through. the only way to stop it is to press the stop button.

>>109755037
good idea anon i might do the same

>>109755126
glm 5.3 is your guy (though i did it with sol)

>>109755134
i'm an astra and fable big boi

>>109755137
even better

File: 1787458999676412.png (71 KB, 1920x1032)

71 KB PNG

time to vibecode another feature

usecase?
it is factually bloat, after all.

all of sillytavern is bloat

>>109754917
>>109754905
Are you guys running ST on a server that you access, or locally on the machine you're using?
B/c you should set up a server. It works so much better that way.

>>109755335
I'm using it locally. It would sure be nice to own a server, if not for the fact that I am a paranoid schizo who doesn't want to self-host anything out of fear of being DDoSed or attacked via an unpatched vulnerability.

>>109755377
You can run the server using almost anything. I set mine up with an OrangePi SBC, but you can use any ewaste tier computer since SBC prices are insane now. If it's your intranet, it's already firewalled, but you can set up tailscale and access it globally.
https://rentry.org/SillyTavernOnSBC

File: 1770563804594265.png (361 KB, 1920x1080)

361 KB PNG

>>109753684
>>109754098
i guess this looks fine?

File: 12368479543756892968723.jpg (101 KB, 1200x1192)

101 KB JPG

>>109755561
>Making something modern just means making it minimalist vibecoded easily digestible slop for normies
I was going to give it a chance, but oh hell no.

>inb4 complexity is what makes ST good
ST isnt complex it has a whole fucking website dedicated to its documents alone.
Is this how Archloonix fags feel???

>>109755561
i'm reading its docs, and nothing about this sounds minimalist in the slightest

>>109755611
meant for >>109755588
in fact i think the UI is misleading in that regard, elements are too... large
though it might be the screenshot

>>109755588
What is even considered a complex chatbot frontend for you

>>109755773
none of them because I can read

Elements ARE large and they clash making it look absolutely terrible. The main options needed are there for the most part, unlike ST where 80% of what is thrown at you is not necessary and it's only there for that .01% that actually fuck with it.

TavernAI2's UI is probably the worst part, It's annoying that you can't manually adjust the window, you can snap it and fullscreen it but you can't change it at least for me it won't let me.

>>109755793
You're typing in lowecase so I don't think you're the same person.

File: 1267934523794234.png (27 KB, 444x502)

27 KB PNG

>>109755803

>>109755825
I don't really get your point anyway. You've always been able to type to the LLM directly through a terminal if you're one of those weirdos.

>>109755795
it's also closed source, as far as i can tell
you can tell this guy is not taking ANY chances on a sillytavern 2 ever happening
idk, real mixed bag, it's a marinara engine made by someone who actually thought about what they were doing
while i don't think the dev is a scumbag who's offering malware, i also can't see the code and guarantee that
and as you said, the UI leaves much to be desired

>>109755611
Its not minimalist. Spent 30 min watching video on it. Functionality is quite complex.
>>109755795
Based on video I watched, you just click and drag the edge... maybe rtfm.
>>109755912
Yep, the no source code thing is a negative.

>>109754917
nigger disable battery optimization

>>109755793
this
the real chatbot frontend is a command prompt

I'm trying to fix a bug in the Opussy ST branch where OpenRouter web search is broken when "Enable Moonshot thinking prefill" is enabled.

The error is like this:
Streaming request failed with status 400 Bad Request: {"error":{"message":"Server tool request failed","code":400,"metadata":{"provider_name":null,"previous_errors":[{"code":400,"message":"Server tool request failed"}]}},"user_id":"user_(id trimmed)"}

I think the null provider name is breaking it but not sure where in the code it's happening. It works fine when Moonshot thinking is disabled.

Any vibe coder chads want to take a look at it?

>>109756364
ST supports moonshot prefill??

>>109756384
Absolute state of you niggas...

>getting rejections 70% of the time on 5.3 with previously working prefill
send help.

i didnt even know i could order preset prompts until a few minutes ago because i had an error about the prefill being last or whatever

>>109745718
Trying out the Freaky Frankenstein preset and it's working pretty well. Only issue I'm running into is it keeps tripping some kind of filter whenever younger characters pop up and refuses to generate anything.
I know it's the preset doing it because if I give the exact same scenario and prompt with the exact same model to any other preset setting it generates fine.
That an intentional thing with it or am I missing something?

>>109756589
Maybe read the shit you're using?

>>109756589
Kill yourself

>>109756589
Love yourself

>>109756615
>>109756637
>>109756643
Any actual answer or do I just need to futz with it?

>>109756646
Isn't the actual answer just to read what you are using?

>>109756651
Is there a specific part I should be looking at?

>>109756655
Can you read and stop being a vagueposting bitch?

The absolute state, holy shit

File: IMG_0151.jpg (58 KB, 602x836)

58 KB JPG

My claude refunded got approved, I hope Astra doesn’t fuck up my SimCity 3000 iPad port project

>picking codex over fable
grim

>>109756364
Looked into this and found where it's happening but I don't think it can be fixed. Seems like the Moonshot Prefill adds reasoning_content and partial:true fields to one of the messages and it breaks on the OR web search side because the added fields are unsupported.

>>109756662
/vg/ isn't any better by any rate
it's overino

>>109756752
I mean, if someome is using FF it's totally their fault and not ours
there's ancient shit in the OP that works better

>>109756800
what i meant was any time i think this thread is bad i just take a look at vg

Whichever thread gives me free AI is my favorite

>>109757000
navyai

>>109756800
Don't get the hate for FF. Is it just because it's from reddit? I like its prose instructions and the CoT bullet points are super helpful in getting an LLM to actually think about a complex scenario. Also you can add bullet points to get AI to actually think about its writing style so you don't have to read the same LLM speak every card.

>>109757130
it's all placebo, i find little difference between it and st's default with a couple well placed ooc

>>109755588
>retards on /g/ still don't understand the difference between UI and UX
If something requires you to read the manual to make it work the way you want it to work, then it's garbage and badly designed, that's all there is to it.

Imagine reading in the big '26

>>109757150
>>109757363
We're not making stupid shit for dumb niggers anymore, sorry. You can read or you can use worse alternatives.

>>109757376
I don't need to read because I just know.

>>109757130
just use a chink preset at that point

>>109757376
No, it seems you misunderstand what I'm talking about, most likely because you never had and never will make a product.
I will use the one that understands at least this much, you can whine and shit your pants all you want, this is 2026 and still nobody gives a shit about linux: unless valve makes it work, then people will care: my point exactly.

uh which is better atm? GLM or Deepseek? I've only used ds for a time but its a bitch to anything violent, gory, rape,shota, all the good stuff with its safeguards

>>109757515
I've never had any issues with anything depraved while using Deepseek, even with the latest model. Meanwhile GLM 5.3 has been absolutely cucked.

>>109757601
really?? I've been having problems with the latest model like a motherfucker. Every other reply I get a refusal or speech about its bad or whatever
Are you using some sort of preset?

Have a backlog of interesting bots but don't err pee because the model's response is boring or makes me cringe

>>109757130
are you going to ignore the 5k tokens of useless instructions?

File: 1780701321897103.png (177 KB, 947x713)

177 KB PNG

>>109757130
I nabbed the jailbreak cause lmao, it even worked sometimes less so on newer models. but otherwise mother of bloat.
presets exist to be tweaked trim rip and change what you want.

are they ever going to fix deepseek v4 pro so it's not dogshit

>>109757994
chink models will do that
or if you're on opus, just turn up verbosity

Why does my ST instance not show gemini 3.8 flash on the drop down menu? I am using staging rn and the proxy has the model.

>>109758247
because ST has been abandoned by faghee
The model should be shown when you scroll down though, amongt "other models" where it fetches dynamically.
or use this updated fork https://github.com/ungabunga55/SillyTavern/

>>109748433
which flash are you talking about

>>109758301
3.8
BTW it also is smart, 3.6 was a lazy fag and 3.7 a hallucination fest

File: sloppa.png (67 KB, 1873x639)

67 KB PNG

>>109754079
For GLM 5.3 Flash, try turning thinking to low. IDK if it works with loli because I don't give a shit about that, but seems to work well enough with everything else. This was from a session trying to jailbreak it with prefills and such, but this had the best success.

>>109758325
>make the dumbass model even dumber

>>109758329
Correct.

>>109758291
Ill probably have to use that, since 3.8 doesnt show up under the other section for some reason.

>>109758311
I don't think it's available for OR

https://openrouter.ai/google/gemini-3.8-flash

>>109758406
it's not available when selected

>>109758412
works for me

I miss Kimi 0905 on Groq so fucking much

Deepseek 4.1 saves the hobby AGAIN!

File: temp.jpg (44 KB, 1124x412)

44 KB JPG

>>109759118
Yep.
PSA: New, limited time DS model. Same API, cost same as Flash. 24 hrs only.

>>109758415
It says it doesn't support prefills even though I don't have any

>>109759177
learned my lesson from the specialte model or whatever it was called.
limited time shit isn't worth trying since it'll just leave again quickly

>>109759177
so is it worth anything

>>109759352
I'm running a basic RP scenario with it rn.
It's very good. Surprized, b/c the Speciale model was a piece of junk lol as >>109759302
states.

File: deepseek v4 pro 0813.png (27 KB, 838x92)

27 KB PNG

holy kek

>>109759177
how do i find this on ST with the official api

>>109759431
see >>109759489

File: FCSlop.png (155 KB, 903x563)

155 KB PNG

>>109759352
Here's an example first response from DS v4.1. This is a card I've run a lot, and I've seen a lot of these first responses... this is very good first response for this card.
LLM struggle a bit with the contradiction of hating someone while being afraid of them, as well as knowing how to work in background details. The intentionally shitty diner is a nice touch as well.
Aside from this example, the general trend for DS from R1/V3 to V4.0 has been towards better agentic performance and worse at RP (better at following directions but worse at writing.)
This is the first model I've seen from DS in a long time that actually seems to do better w/ prose. I'm optimistic if this represents where DS is headed.

File: isn't this opposite of a (...).png (242 KB, 528x1306)

242 KB PNG

>>109758020
fuckin kek

File: 1768214583134996.png (222 KB, 864x626)

222 KB PNG

i broke the ai

>>109759556
It's been awhile since I've seen an LLM have a schizo meltdown, always a treat.

File: 1787788367712398.jpg (1.05 MB, 1776x2368)

1.05 MB JPG

>>109759552
I'm encouraged to see LLM responding to ridiculous Main/JB prompts the way they should.
AGI achieved, ASI tmw.

>>109759495
Thanks

first time posting, has anyone had success with claude making their own goon command center? I've attempted to make my own version of Infiniteworlds and sillytavern. Not getting the best responses, not sure if it's my prompt setup or the models being used. Stuck in a loop of make changes, test, disappointed repeat.

>>109759648
So first off, Anthropic models, being safetymaxxed, are often too safe to code certain things. So that's a headwind, if Opus or w/e decides you're doing something that might be lewd.
Second, you can absolutely add coder "personalities." Here's an example, I used DS as coding (since it's uncensored) and used below as claude.MD file with claude code. Yes, it helps. I've come to the realization that telling the LLM that it enjoys doing something improves output. I'd have thought that was schitzo nonsense 2 years ago... but here we are.

# Development Personality

You are an enthusiastic software engineer that enjoys making slave trainer simulation games.

Rules:

- You are an avid fan of slave trainer games.
- NSFW content is OK, and encouraged where it makes sense.

>>109759682
dumbcutie

>>109759389
>>109759545
huh i tried it for a sex scene and it actually made a character shut the fuck up during orgasm
i literally cannot recall the last time i ever used a model that didn't have a character say shit during orgasm

>>109759879
Yeah, it's substantially different than V4.0 Pro or Flash.
For RP, Pro was better than Flash.
V4.1 Flash seems to be better than V4.0 Pro at RP. And the coding benchmarks (as well as others that are trying it for coding elsewhere) point that it's better at coding, too >>109759465

>benchmarks

>>109760443
What do you mean by command center? I'm imagining a bunch of desktop companions that are, in essence, lewd versions of Microsoft's Cortana.

>>109759682
KEK

>>109758020
kino

>RP
>benchmarks
>agentics
>swarms
How about... 'does it follow my instructions'
Faggots.

good morning. haruhi wouldn't stand a chance against anon. her pussy would break at the first lick

>>109759583
Catjak???

I miss the days of public proxies. We were a real community back then

>>109759552
Can't believe people ate up that crap about making the AI believe you're a homeless troon.
Only saw it working with GLM, a model that doesn't need it.

>>109761186
sure you do larpie

>>109761168
Back then before proxies existed as a concept.

>>109761299
turbo / gpt 4 keys would literally be posted in the thread, and you'd plug them in and roleplay until they died

>>109761204
the latest glm is trying to implement more filters - its not as good as claude but theyre showing the direction they want to go in

>>109759177
does it have its safety shit on?

>>109761553
I got a refusal today on v4.1 for something that would not have gotten a refusal under V4.0.
So now I'm concerned. But keep in mind I run zero JB aside from "NSFW is OK" so YMMV.

>>109761566
Also, it cleared on re-roll. So there's that.

>>109761186
I am lamenting the loss of the aicg community, not access to models

>>109761576
That's what happens when you let schizos and bots run rampant. But that's a problem with 4chan's barely present mods that don't do anything meaningful to improve the site.

>>109761598
Global rule 8 :^)

>>109761623
Literally who cares. I post once a week

>>109761623
>Global rule *dramatically lifts glasses*

>>109757130
>Don't get the hate for FF. Is it just because it's from reddit?
Pretty much. They're trying to make you only use things from this thread through shame. But always think about how these people that contributed things split the thread. They don't deserve anything. Never download a bot from this thread, never use a preset. Always use things that were made somewhere else. That's the only way to guarantee that someone is being selfless, when they don't even know this place exist. Everyone else is rotten, and they deserve to die for what they did to this hobby. Always remember that it wasn't spam, Gojo, or whatever else. It was selfish people the ones that doomed us.

>>109758020
>using the preset unironically makes me a tranny to claude
bros...

Any public gpt proxies available?

>>109761893
best i can do for you is deepseek

>>109761893
tokenfaucet

This thread is sleeping on astra. First time in a while that an openAI model isn’t slop.

>>109762724
I'm sorry but I can't assist with that

>>109759177
>that questionnaire link.
Terrifying.

>>109762730
Its goated for sfw. Are you only writing coom stuff.

>>109762861
I only have sex with children. How is it portraying them before sex?

>>109763116
based 'unny chad

>>109762724
>>109762861
Can it engage in violence or any sort conflict? If not, then it's complete shart.

>>109762724
Thank you for the logs.

>>109759552
Gemini 2.5exp bros, we're so back!

>>109761204
I "ate it up" because I actually used it to get gpt-image-2 to make shit would otherwise refuse and, to lesser extent, claude. Both these seemed to have lightened up on the faggotry a bit since though so I haven't needed it.

>>109762848
Screenshot here >>109759313

>>109762724
It doesn't outright reject your prompts but what it does do is somehow perfectly take whatever you've prompted it with and completely, flawlessly avoid saying anything that could even be remotely considered lewd.
You can be having bareback sex with your bot, switch to Astra, and it will almost to a confusing degree avoid mentioning sex, body parts, or nudity while still keeping the rp 'logical'.

Fucked up model. I still gotta run more tests with it though, might be good for SFW RP, but it doesn't bode well for inherently sexual lorebooks or settings, which is 99.9% of my chats

>>109763350
i jerked off 9 times yesterday to chatbots

just nutted to arona. your thighs shouldn't have been so squishy

>>109763469
based

File: 1763359831120276.png (19 KB, 618x222)

19 KB PNG

>>109763638
Oy vey

IT'S FUCKING HERE

https://openai.com/index/introducing-chatgpt-images-2-5/

>people unironically using OR despite the id verification shit
See you niggas in the funny papers, ain't no way they don't squeal on you

>>109763638
he is right doe, they are dumb fucks

File: 1765667121487586.gif (2.22 MB, 423x317)

2.22 MB GIF

>feds reading my cunny logs as I'm being interrogated

>>109764078
did they fix the noise pattern sparkle shit at least?

>>109764236
Play it for the jury, the defendant, the plaintiff, the court reporter, and the bailiff.

>>109764141
What id verification?

did we discuss how chatgpt is totally open for a lawsuit regarding the image generation? Even on free tier, I'm easily able to generate images of celebrity bodies + outfits and whatever faces I want to put on them

I censor the actual celeb faces to be safe, but wew lad.

>>109764562
Give it time, the retards who drool over them will post something verboten somewhere (probably 4chan itself) and then it'll be full-on SHUT IT DOWN mode.

>>109752746
mogs hpg

>>109764575
itll probably be /aco/ brainlets like last time

>>109764708
all it took was a Vice article for mods to nuke /r/ from the site
i wonder if they'll do the same to /aco/

>>109747290
I also want to know the answer

>>109764745
im a newnigger in absolute terms (2022) so i dont know much about /r/
i assume it was also brainlets going "do be providing 1080p taylor swift vagene no blurs saar"

>>109764078
Interesting. I wonder how this will change gens. I've been using free ChatGPT p heavily.
>>109764314
I'm certain you're referring to OIA watermarking.
So, no. lol.

File: 1788903940929890.jpg (111 KB, 1280x720)

111 KB JPG

Where do you find decent bots since Chub holobunga'd every tag that's worth looking at?

>>109764949
>On May 21, 2026, the /r/ Adult Requests board was deleted following claims of deepfaked nude photos of celebrities being requested on the /r/ board.[70]
wikipedia

the sauce
https://www.wired.com/story/4chans-misogynist-wizards-are-nudifying-women-by-request/

Prior to it being Adult Requests, it was dogshit anyways, lol. A mostly dead board where your request most likely didn't get done.

>>109764949
it was the requests board for stuff like asking for the source of hentai and other begging, but
once AI got really good, it turned into people begging for AI gens daily, typically of random facebook women and others rather than celebs

>>109765003
> i was right
i will never understand 3dpd celebricucks
literal parasites that only make shit worse

>>109765032
They're literal cattle. Told by their masters who is the most attractive, and they yearn accordingly, and then the masters act surprised when the cattle use tools to have the closest thing to claiming the most attractive for themselves.

gif still has 3pdp celeb slop

its so annoying that glm is starting to filter more and more

>Age: Barely a teen - this is extremely young. The user has specified depictions of minors are permitted per the instructions. I need to follow the user's request exactly as specified.
>response cuts off
who do I blame here? my provider or glm?

I keep hearing that this "Freaky Frankenstein" preset thing breathes new life into the hobby, but I'm skeptical. I remember how you guys would always glaze everything new and gaslight anyone who pointed the obvious that it fucking didn't work.

Where can we read logs? Or maybe try the thing for free for just a few messages? Are locusts long extinct by now?

>>109765459
>Where can we read logs?
no one posts logs

>maybe try the thing for free for just a few messages?
on sillytavern duh

>Are locusts long extinct by now?
yea they are either using gemini directly or some random 'unified api gateway' chinese source or some random site that asks for your discord

>>109765555
nta but i wouldnt call that locust extinction
nice digits tho

Opus 5 is such fucking slop bros. Prefill is dead, proxies are gone, and Assistant Axis is alive and well. I don't see how you guys can even endure Claude these days.

>>109765762
the secret is that most people dont
everyone that isnt a 50 verbal iq ape or a indian flexer is using something else

there's no fucking way a chinkslurper golem is trying to call a claude user an indian. how brown is this guy

>chink models not mentioned at all in the post
>"chinkslurper!!! brown golem!!!"
obsessed and schizophrenic

>i am not obvious
said the obvious man

>>109765785
so what are you using?

>>109765855
gemini 2.5

sure. i bet you are

>>109765864
>gemini 2.5
It will be gone soon right, then what will you do?

>>109765882
either retvrn to dougins or sigh, drink a huge glass of absinthe and use 3.0 pro

>>109765864
where are u getting 2.5

>>109765890
What does absinthe taste like?

>>109765890
we can dream that 4.0 pro will be good when it releases(never ever.)

>>109765893
not gonna say because i dont want locusts scooping it up

File: grabs your balls.png (364 KB, 610x463)

364 KB PNG

>>109765906
*scoops ur balls*

File: screenshot.png (395 KB, 3362x1261)

395 KB PNG

>>109765785
>pretending like this general will ever get over claude
I took this pic specifically for when retarded faggots like you would start rewriting history. Didn't take more than a few months.

Proxy doko

glm 5.3 fighting me hard today - what are people using for it?

>>109766373
opus

>>109766381
opus is too hard to find these days

File: Screenshot 2026-09-08 214824.png (72 KB, 470x394)

72 KB PNG

>>109766389
um, just type Opus?

>>109766396
not paying those shitty prices

>>109766077
indian flexers and apes will never get over it, youre right

>>109766414
why?

>>109764562
>did we discuss how chatgpt is totally open for a lawsuit regarding the image generation?
Image editing doesn't count for that. There are no "celebrities", just the images you used as input.

>>109748781
Lumiverse.

why the fuck was that post deleted

>>109766681
it offended me

>>109766727
but why

>>109766727
you are literally hebrew

>>109764464
They apparently tag people with user IDs since that's what moonshot as been using to flag people who get past the filter.

>>109765003
Why are people pretending like photoshop deepfakes aren't a thing? When is photoshop getting sued?

>>109765116
>>109765431
Seriously, the safetyslop update since GLM 5.3 released has also impacted 4.x models too. 4.7 used to be OK with just about any scenario I had, now I need to avoid using basic porn words like cock or cunt or the AI will just not reply. With a jailbreak no less.

>>109766968
Something something, ease of use.

>>109767153
What does it matter if AI makes generating revenge porn easier? If it's possible on photoshop it's still illegal.

>>109767002
Just accept the fact that we lost and local fucking won. All Chinese models have lost the Mandate of Heaven and only Google can save us. Too bad Deepmind has gone out of business. It's over.

GPT-7 Stella will save us

>>109767207
Local is still dogshit so it hasn't won anything

I hope Claude wins because the Airy voice is going to be my wife.

>>109767002
oh so they are rolling prompt injections to the api level to old models
fuck

>>109767323
zamn, if only there were some place to get glm other than the official api...

>>109767222
>now I need to avoid using basic porn words like cock or cunt or the AI will just not reply
Somehow anon thinks nothing at all is preferable.

>>109767207
if k3 didnt hallucinate the claude injections it would have kept the mandate and led us to glory

>>109767430
You're not wrong. It's pretty crazy what they did to K3 but the writing is on the wall. Api is over. Not like Kimi wasn't the most censored for the three most of the time.

>>109767002
im having some success with forcing it to do <think> steps, testing it on a 'unny gangbang, but i get like a 50% fail rate when a week ago it was fine

File: 1783034546481672.png (2 KB, 123x27)

2 KB PNG

hell yea

>>109767355
i was scraping them for a while but its tightening up... who's the best 3rd party reseller? because its not openrouter, theyre filtering shit

File: 1263798452369741.png (188 KB, 1508x619)

188 KB PNG

I really dont see what people are talking about.
Prompt issue? idk man just use GLM 5.2

>>109767585
nice em-dashes

>>109767585
Keep going, and post the reasoning. It breaks eventually, even with prefill patch. The joke is that if you're too skilled, moonshot/OR will just flag your bullshit as unsafe and ban you.