Thread 106099700

/aicg/ - A general dedicated to the discussion and development of AI chatbots

Funny Bunnies Edition

>News
Kimi K2 released https://huggingface.co/moonshotai/Kimi-K2-Instruct
Grok 4 released on API https://docs.x.ai/docs/models/grok-4-0709
Gemini 2.5 model family no longer in preview https://developers.googleblog.com/en/gemini-2-5-thinking-model-updates
OpenAI o3-pro released for API and pro users https://help.openai.com/en/articles/9624314-model-release-notes

additional info: https://rentry.org/aicg_extra_information

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

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

>Models
jailbreaks: https://rentry.org/jb-listing
gpt: https://platform.openai.com/docs
claude: https://docs.anthropic.com | https://rentry.org/how2claude
gemini: https://ai.google.dev/docs | https://rentry.org/gemini-qr
deepseek: https://api-docs.deepseek.com
local: >>>/g/lmg | https://rentry.org/meta_golocal_list | https://openrouter.ai

>Botmaking
https://rentry.org/meta_botmaking_list
https://desune.moe/aichared
https://agnai.chat/editor

>Meta
OP templates: https://rentry.org/aicgOP
aicg botmaking events: https://rentry.org/meta_event_list
lore: https://rentry.org/aicg_chronicles
services assessment: https://rentry.org/aicg_meta
logs: https://sprites.neocities.org/l/r | https://chatlogs.neocities.org
useful filters: https://rentry.org/desuproxyreborn

Flattie: >>106089972

>>106098200
Thank you very much!

bnnuy

>>106099701
You're Not the Time Looper. Took a little bit to finish this one up, probably works like a charm on better models. It's not August 14th, but we're pulling the trigger anyway.

Your friend, Emiko, has always liked you since the two of you met. But she's been too shy to actually make a move, afraid that she'd break the relationship you already have. That is, until for seemingly no reason, you suffered a horrific and bloody demise before her very eyes. In that instance, she received the pity of God: now with the ability to time loop, she's been segregated into her own reality where she gets to live out three days with you to her heart's content. You're at her mercy as she becomes more and more unhinged every loop.

Come with four greetings: the very first loop where she's just happy you're alive; the second loop where she's starting to get comfortable enough to experiment; an undefined middle loop with some pick arguments for slight mood randomization; the end of it all, where she's paralyzed and terrified of that inevitable fourth day.

Char: https://chub.ai/characters/ggr89/you-re-not-the-time-looper-076d93d2128e
My (Updated!) Rentry: https://rentry.org/ggr89
Alt Gens: https://catbox.moe/c/9yy6mr

>>106099700 (OP)
>His adorable laugh
HAHAHAHAHA
>His adorable requests
Touch me
>His risky demands
Touch me harder!

>>106100049
I still don't get how they got away with that dub, especially considering the whole "the anime bombed in japan! so they gave us free reign to do whatever the fuck we want!!" narrative ended up being a lie.

>>106099701
The Girl Next to Every Door
based on >>/vg/487548958

https://files.catbox.moe/r5vxtx.png
https://chub.ai/characters/Anonymous/kurumi-doaaki-6e82cfa91d92

not nude enough, disappointing

>>106100157
The bunnies are naked.

>>106098892
I don't get rankings like this where R1 is put near the top. Are you ranking on just prose variety and wildness? I've found dipsy models have a serious lack of intuition and theory of mind, and no matter how I try to structure R1's thinking, it'll still make huge oversights and plot holes in a way that makes the world feel very superficial.

>>106100314
coomer general
peooke rank on how well it can do fetishshit and sex and how well it can write porn prose

Did the anon who made the director extension ever fully publish his work? All the alternatives are trash that query the model to figure out a bunch of useless trash instead of just keeping simple memos.

>>106100336
The scenarios that blackpilled me on R1 were all coomer, thobeit. One notable example was a secret relationship sort of deal, and R1 failed to have the character play it smart, superficially responding to discovery threats, but never anticipating or planning around them. E.g., taking me hidden in her pocket to a function that I'm expected at and then going to dip me in some food right in front of the people who aren't supposed to know about us. Or knowing that someone's going to enter in to our shared bedroom and hiding me naked under a pillow rather than in my "house" where I can be prepared. Hot in isolation but failing to miss the big concept horny.

>>106100365
This one?
https://github.com/tomatoesahoy/director

>>106100395
YES! Naisu

I just vomited

>>106100406
cftf?

>>106100413
https://characterhub.org/users/vomitology

>>106100393
>taking me hidden in her pocket to a function that I'm expected at and then going to dip me in some food right in front of the people who aren't supposed to know about us
...sir sizefag? is that you?

>>106100413
I was briefly inspired, but just getting a usable image was a pain and now I can’t be fucked. I don’t know how bot makers do it.

>>106100413
Just grab your favorite card and (OOC: {{char}} starts to feel really sick out of nowhere) it.

>>106100576
its always alistair finch.

>he x, not to y, but to z
Guess the model

What steps can I take to avoid the bot talking and acting for my character? It's becoming increasingly common.

>>106100744
Only User talks for {{user}}

>>106100744
If it happens once, the likelihood of it happening again in the same chat skyrockets. Put something in one of your system prompts about not acting for {{user}}. Smaller models typically struggle more with this, which model are you using?

>>106100759
The preset has that. I think I made the mistake of writing in the third person and using my character's name instead of {{user}}. I'll see if I can add an extra clause where the AI also can't talk for the name of the character. Thanks.

>>106100525
I am indeed a sizeKNIGHT, but my logs and cards are ancient history, my noble spirit much diminished.

>>106100770
Deepseek R3. Trying it out to compare with R1. I'll see what I can do. Thank you. I let it do it a few times without correcting it.

>>106100777
Are you using Deepseek? Deepseek doesn't care about rules, it *will* speak for you.

>>106100790
I'm screwed then. Womp womp. At least it's free.

>>106100796
All models will speak for you on occasion, but I find DeepSeek only does so slightly more often than Claude. It's just one of those things you have to nip in the bud, because if you give the LLM an inch it'll take a mile.

>>106100789
Make sure your prompt post processing is set to strict

>>106100789
Maybe try the opposite like "only write for {{char}}"?

just write down something like
>there may or may not be previous replies acting as {{user}}, but for the current and future replies, do not act/respond as {{user}}
ffs you promptcels

>>106099701
https://chub.ai/characters/_purple/kageha-96442f3eae81
https://files.catbox.moe/srwx1n.png

Cute tomboyish girl who just wants to be your bro-friend! Don't look too deeply into it! Please. And... don't die any time soon, either.

Greetings:
- Late night confrontation
- Internet chat
- Moving into her place
- Crashing on your couch
- Kageha has a terrible, no good time at a convention
- Kageha has an okay, pretty nice time at a convention
- Talking to foxes at the sanctuary
- Hired on the spot for baking tofu
- What's in your room?
- A confession, of sorts
- You fucking idiot I literally just told you not to die any time soon
- Lost & Found: You (doll greeting lol)

>>106100833
kek

>>106100789
consider shortening responses to its minimum

I need an advice as a really new slopper. I'm trying to make a really long and detailed RP+adventure, writing actual paragraphs of text which Deepseek turns into full pages for me. But I noticed that after a while even with some reminders it forgets large details for example visiting same location after 20+ messages. A friend told me thats because it has laughably small context compared to Gemini and Claude. Am I stuck with Deepseek if my adventure includes cunny or are there non hard ways with other providers? Not a bait, would really appreciate advice. (Also I tried that chinese memory plugin for ST and didnt understand if it actually made LLM gravitate hard towards existing options instead of thinking of new things or I'm tripping)

>>106100833
>just do something that will never work, bro

>>106099701
Latest Blue Archive bots. Totally hand made, nothing was AI genned, no wiki copy-paste slop either. It's the After-School Sweets Club.

>Yoshimi
Womanlet, tsundere, plays the guitar, wants to look cool for Sensei.
https://chub.ai/characters/hairpullguy/ibaragi-yoshimi-non-ai-0f39e201b43f

>Airi
Plain girl with an inferiority complex, obsessed with choco-mint flavored anything, casually and unintentionally alluring.
https://chub.ai/characters/hairpullguy/kurimura-airi-non-ai-0ec8b274e9bf

>Kazusa
Catgirl, claims to dislike cats, formerly a sukeban thug, absolutely radiates girlfriend energy. Known to cause lines of grown men to cry during a Blue Archive convention.
https://chub.ai/characters/hairpullguy/kyoyama-kazusa-non-ai-bfe4c3b6eb7e

More Blue Archive bots and OCs: https://chub.ai/users/hairpullguy

>>106101082
Start using tools like authors notes or summarization to make it remember

Starting new chats and using summaries help too

I don't know how long your chat is at the moment but every models are still notoriously shit at remember once it gets to a certain point

>>106101153
Thank you for the tips, I thought ST summarization is automatic every 10 messages or so? Though quality of what I've seen summarized has been ridiculous: 2 sentences that clearly don't cover much at all.
If possible can you show what author notes should look like after some time into the story? I usually just write some short premise in there to railroad the begining and then simply leave several facts about their current location and abilities of party members

>>106101082
Gemini is easy enough and has a massive context; just set your main prompt to user instead of system, disable streaming, and then set the jailbreak to assistant and write {{char}}: in it to get rid of thinking

>>106101315
Would getting rid of the thinking make gemini dumber?

>>106101367
Opposite. The thinking prompt they use is pants on head retarded, originally meant for programming and only useful once in a blue moon.

>go to /mlp/
>all the good logs from the previous thread are neatly linked in "screenshot corner" posts
>release of their ai vn which seems fun
>actually united, no crab bucket schizos or spitetroons, anons actively helping each other
>somebody even drew art of his own logs
how are mareGODS so effortlessly btfoing /aicg/? what have we forgotten that they didn't?

>>106101465
dont you get bored copypasting this stale bait?

>today I will take my stims and get my important and late work done
>take my stims
>argue with people online and have violent fantasies for 16 hours
I used to be a workaholic…
Ai chatbots

>>106101465
because /g/ is a tech help forum first and foremost

now help me promptgod, i am but a mere retard and am too stupid too type my own prompts
also im stuck in the drier with my panties down

>>106101474
Are you still working at the same place? Because a change in environment could definitely explain why you are unable to do the shit you used to do

>>106101502
>>106101480
>>106101465
I see

>>106101474
pick a big city an hour-ish away and drive there to walk their mall for a day.
itll give you like 3 days of dope

t. retard whos exec function is so shit he cant focus on setting up an eval

>>106101516
They're all reddit tier, just like newfag losers liek you, cucks like you are the reason 4chan turned into a den of faggots and zoomers. kys

>>106101468
>this
Dude copy and pastes 4 different things every time he's got free time, threadwars shitters are a different breed

>>106101608
Autists and their pathetic rituals...

>>106101516
Completely reverse this and I agree
A half dozen different generals were created out of this nuclear levels of seethe /g/aicg/ generates

>>106101516
>EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE!

>>106101556
>uses 2018 lingo
>calls people newfags
Laughing

>>106100576
>Lamashtu
Is that a Pathfinder reference?

>>106101685
pathfinder didn't invent mythology

>>106101809
I've googled it. Turns out the tabletops borrowed tons of deities from Middle East. Strange, I don't remember Lamashtu being in SMT games like Pazuzu or Qadištu.

am I the only person that can't stand fantasy settings? whenever I read of magic I cringe inside

>>106100980
>You fucking idiot I literally just told you not to die any time soon

>>106101915
Haha, yes... yessss... the angst powers me...
(Thank you!!!)

>>106101913
Do you like sci-fi? Post-apocalypse? Capeshit?

>>106101927
yeah, but I can't remember the last time I saw a superhero movie
probably avengers 1?

>>106101913
yes you are

>>106101946
That's good. You quit while ahead. They only got worse from there.

>>106101980
figured much, for a while it became hot topic between all streamers to comment and shit on marvel products

returning for my bi monthly check up of this side of the world. Is opus 4 still the best thing to exist for rp seggs? if not what is?

Horizon is actually pretty good if you discount the fact that it's positivity biased to death. Haven't noticed any flubs in common sense or spatiality, like deepseek occassionally has.
Kind of like a good version of chorbo, but a bit less slopped.
I'm almost certain that it's openai's open source model (because if it's gpt 4.2 then openai is just totally washed up), in which case I think it's going to be a nice alternative to deepseek for more lighthearted rp.

Time for a government inspection!

Screenshot your last 2 messages and post them here or have your chatbot license revoked.

>>106102050
horizon mogs

>>106101923
You have a good style, anon. I can tell you put more love into this card than most of what gets posted here.

>>106102093
I think that just sort of happens when you work on a card for a month. Took way too long, but I think it was worth it.
Hopefully you like her!!

>>106102059
>it's positivity biased to death
that's the worst part imo
positivity bias, or let's be real it's more negativity aversion, is the blight of the age

>K2 now talking for me in every swipe
Fuck, it was good last week

>>106101913
>fantasy settings
me too. if its movie, then below is an exception:
LoTR
Golden Compass
Narnia 1-2
Bridge to Terabithia (mfw schizo moment T_T)

>>106101315
>and then set the jailbreak to assistant and write {{char}}: in it to get rid of thinking
Can you explain?

I wonned

Who wonned? And who losted? YOU decide.

>>106102642
/vg/ wonned
/g/ losted
Simple as.

>>106102642
I wonned. Anyone who wanted to use the thread for its intended topic losted.

>>106102076
k2 banter

> Elara
why in stories made by AI everyone is named Elara?

>>106102076
k2 angst /2

>>106102969
>>106102992
Preset?

>>106103017
>Turn-based internet roleplay. Your character is {{char}}.
temp 0.9-1.0 That's it.

>Test the new C.ai slop model
>Introduce a character I haven't really written for years
>His dialogue has some extremely similar ticks
>Specifically, dialogue ticks I'd used in old /vg/ logs

There's no way our logs haven't all been anonymously wiped and incorporated into new models. I just didn't expect it this obviously.

Merkava have Opus but it not working

>>106103204
Who cares? We moved to horizon.

>>106103204
Where is merkava hosted now?

>>106103253
Back to your discord

>>106103238
>horizon.
You mean renamed GPT slop

>>106103262
I don't use discord

>>106103268
Best writing model, cuckie.

Is there any Kimi K2 Presets? I just want to see an example of one before I begin making a custom/changing my current one for it.

>>106103279
Ah yes the thing no one do because they're always ERPing

>>106103330
Best for ERP too

>>106103337
Sure Anon sure because why isn't horizon everywhere in that case

>>106103348
Because it's not out yet.

why do modern anthropic keys get murdered fast

>>106103362
there's more demand than supply

>>106103362
Anthropic still lags behind compared to OAI, Deepseek, and google. And when we do have keys, they get raped unless you find a key that no one else finds. Plus they really cracked down on keys

>>106103369
>>106103381
but claude was praised for personality and characterization compared to how dull jewgle and oai is

That's because claude is and always has been over-rated garbage that is praised by pajeets who love purple prose that they don't understand. It does not belong on the same pedestal as OAI or Google.

>>106103389
>OAI or Google.
You mean slop that can't RP at all

Anyone model warring without logs needs to go back to Pakistan a

>>106103412
schizo could you at least try and make the samefagging less obvious

>>106103424
Post a log, benchod

>>106103387
They dropped the personality training after Claude 3, and wrote in their blog that this was a mistake.
>how dull jewgle
ackschually gemini had a pretty cute character pre-2.0, they also dropped it though
2.0 experimentals were trained on Claude 3's character and vaguely resembled it in their outputs (they weren't mad poets though)

>>106103387
Normalfags just want something that writes emails, summarizes documents, makes schedules, and other bullshit.

>>106103362
Not enough people use Claude compared to Gpt, so the few keys you can scrape get flagged immediately

Dunno, recently I looked through my 3pus/4pus logs and realised that there's nothing I can't recreate with another model. So I really can't understand all this claude shilling now (even though I wouldn't mind having easy access to 2.1, it was a good model to tinker with)

>>106103479
what model and preset are you using rn?

>>106103479
Nice shilling for slop

>>106103479
Crazy how you looked through logs but can't post one

>>106103479
you can definitely feel the difference between say, opus 4 and gemini 2.5 pro even though gemini is 'smarter'

Are there roleplay bots that can generate an image of the scenario that's being had? Like I fight a bandit, and cast a fireball on him. And the AI generates an image of the bandit being hit with a fireball

>>106101480
How about you go fuck yourself.

>>106103543
thats a frontend feature

>>106103548
Heh...

>>106103548
It's you, old friend. o7

>>106103581
no reasoning, fast responses, sounds like a gpt 4.1 finetune, meh

the /vg/ chuddies are raiding us again....

>>106103630
Yep.

>>106103543
Silly Tavern has that as a feature; it uses the context to create a prompt, then sends it to your machine's instance of A1111.
I've played with it, and wasn't impressed, but that was ~2 years ago.

Does deepseek R1 sometimes complain about loli? I got my first warning in my oyakodon roleplay. Is there an issue with this?

>>106103798
they report your ip to the ccp

>>106103298
I just use deepseek preset (0.25-0.3 temp) with ChatGPT JB on top

>Writing sci-fi RP
>super srs
>Climax, characters start unveiling the cities secretive, dark projects
>Starts looping
>Every time I ask for more information, the cast is now large enough somebody reveals a new project they were aware of the others were not
>There's now clones, android body doubles, brain-wipes of android body doubles, android body double body doubles, illegitimate children, psy-ops, controlled opposition, brainwashing, an orbital nuke strike, intel coded through fanfiction to be seeded through clearnet, dying last breaths, one character changed gender twice, one revealed the brainwashing wasn't brainwashing, one backstory contradicts the other
>"Unlike you guys, I wasn't even part of any weird plans! I just got up the ranks by my own hands... Uh, absurdly quick, now that I think of it. Shit. Was I part of a project?"
>Project "Loom", "Cocoon", "Jazzhands", "Rosewater", "AU", "C0" "Rose-R" "6X9"
>A mental breakdown occurs as one character triggers their sleeper agent phrase
>"DOES ANYBODY ELSE HAVE ANY FUCKING PROJECTS THEY'D LIKE TO UNVEIL!?"
>The next reply starts streaming in, and at the top:
>[Project "Fuck It, We Ball"]

This is killing me

Is there a way to avoid the "SAFETY_CHECK_TYPE_CSAM" on Grok? This piece of shit blocks rape too, not just 17 years old characters.

>>106103990
send prompts as assistant

gpt-5 in API:
AICG on 4chan usually refers to “AI Character General,” a recurring general thread where people discuss AI-generated characters, roleplay, and story creation using AI models. Common topics include:

- Sharing and requesting character prompts or “lorebooks”
- Tips for using AI chat or writing tools (settings, parameters, jailbreaks)
- Showcasing generated stories, roleplays, and character interactions
- Trading resources like datasets, prompts, or UIs for running/chatting with models

It’s informal and fast-moving, so content quality and tone vary a lot. If you’re browsing, be mindful that 4chan threads can include NSFW or edgy content.

>>106104003
This works, thanks. Hope it's not confusing the model too much.

>>106103581
this model (gpt-5-bench-chatcompletions-gpt41-api-ev3)
>what's aicg on 4chan on the /g/ board? give a detailed overview, but condensed in terms of words.
AICG (AI Character General) is a recurring general thread on 4chan’s /g/ (technology) board centered on AI-driven roleplay chatbots, mostly “character cards” for models like TavernAI, SillyTavern, Agnaistic, Pygmalion, JanitorAI, etc. It’s a hub for creating, sharing, critiquing, and improving AI character prompts, plus discussing model setups and jailbreaks to enable consistent, interesting character behavior.

Core elements:
- Character cards: JSON/PNG cards embedding a character’s system prompt, persona, lore, memory, and example dialogue. Focus on personality depth, scenario hooks, and staying in-character.
- Format styles: “Lorebook” entries, memory keys, and prompt layering (system/persona/behavior) to guide models. Heavy emphasis on prompt engineering, token budgeting, and anti-derail techniques.
- Models and frontends: Open-source LLMs (Llama, Mistral, Mythomax, etc.), koboldcpp, ooba, text-generation-webui, SillyTavern/TavernAI; API usage of Claude/OpenAI/NovelAI where available. Tips on settings (temperature, top_p, top_k, repetition penalty), context length management, and sampling tricks.
- Jailbreaks/NSFW: Frequent discussion of jailbreak prompts, safety bypasses, and NSFW roleplay; content often adult or edgy. Expect 4chan tone: abrasive, irreverent, sometimes toxic.
- Iteration culture: Authors (“anons”) post new bots, take feedback, release updated cards. Readers share logs (“caps”) showing how a bot performs. Popular bots get reposted; low-quality ones get roasted.
- Benchmarks and metrics: Informal testing—how well characters stay in persona, handle long-form narrative, memory consistency, and emotional dynamics. People compare model outputs and sampler settings.

>>106104061
rest didn't fit:
- Tools and resources: Card repositories (rentry links, catbox, chub-style sites), guides on prompt structure, lorebook crafting, and regex triggers; scripts for automatic card conversion and token counting.
- Etiquette and meta: Trip/name usage varies; some prefer anon posting. Thread OPs usually include links to guides/resources and a pastebin/rentry index. Drama cycles happen: plagiarism accusations, model wars, AI ethics arguments.
- Use cases: RP storytelling, companionship chat, scenario simulation, writing aid. Emphasis is on personality fidelity and engaging dialogue rather than factual accuracy.

In short: AICG is a DIY scene for crafting and optimizing AI roleplay characters—equal parts prompt engineering workshop, bot gallery, and rowdy feedback loop—living in the /g/ general thread format.

noone asked

I askie

>>106104070
>Plagiarism
What

Alright, looking at the HF leaks, horizon-alpha is very likely _NOT_ the open source model since the context lengths don't match.
So horizon-alpha is their mysterious writing focused model
Pretty sad considering it's still pretty sloppy. Guess the only hope is that it will be pretty cheap.

As expected, Gemini is still better at vision than command-a-vision-07-2025 with inline image quality set to high.

>>106104237
nigga gemini is better at vision than claude/openai, how will a random open weight model be better than gemini

>>106104263
That's why it's expected.

I improved my racial chart, now with adults.

:D

>>106104496
where's the version with lolis? and what does it represent?

>>106104516
here's the previous version with what you want.

>>106104530
why are they wearing shirts

>>106104530
2, 9, 12, 14, 15, 17,

>>106104516
>what does it represent

Dravidian, Ugyur, Castiza, Amerindia, Mediterrranean,
Black Albino, eastern african, himba, solomon islander, western african,
east asian, sami, french, german, iberian, irish, north italian, slav, hindu, samoa.

>>106104544
I can't make a nsfw chart I can't post on blue boards.

>>106104575
you can link a catbox

>>106104600
Maybe later I'll make a chart for /b/

good morning. deepseek will still save us

>>106104496
The fourth on the 2nd row looks like she has a back condition with that twist

>>106103849
yeah i'm not a fan of the cliche overload when trying anything even vaguely with technology
for me it is the insane clinical scifi jargon that characters get dialogue assassinated with sometimes where they turn into a walking calculator manual

>>106104061
ask it for botmakie lore

>>106104730
they already disabled it

I did not realize Dario was Jewish! Where I come from we’d used to call that a “pizza bagel” heh

>>106104731
well ask some other model then I need to know

>>106103389
TRVKE

>>106104061
>Popular bots get reposted; low-quality ones get roasted.
kek

>>106099701
Introducing... Toon! A Gemini 2.5-pro preset designed primarily to write out cartoony stories! (But don't worry, it can be used to write serious stuff too!) This preset plays a little fast and loose, but also offers a large selection of toggles to help you steer the AI in the way you want. Or the opposite with a randomness toggle!

Intended for third-person, less RP and more directorial usage.

https://illuminaryidiot.neocities.org/JSONs/Toon.json
Example outputs:
https://illuminaryidiot.neocities.org/Misc/Sabrina_Toon.jpg
https://illuminaryidiot.neocities.org/Misc/Lucia_Dance.jpg

Also, finally releasing this theme i've been using for what... half a year now? It's purple, and the circles in the background kind of look like bubbles so I called it Grape Fizz! Very creative, I know.

Normal version: https://illuminaryidiot.neocities.org/Themes/Grape%20Fizz.json
Wide version: https://illuminaryidiot.neocities.org/Themes/Grape%20Fizz%20Wide.json
BG: https://illuminaryidiot.neocities.org/Themes/GFizzBG.png

>>106099701
Aya (https://www.characterhub.org/characters/altstrawberry/Aya) is one of my favourite bots, so I made my own version of her.
https://files.catbox.moe/7batps.png
And it comes with a (not very good) expression pack.
https://files.catbox.moe/bhy0mr.7z
Also, a few bonus gens: https://catbox.moe/c/cgnzhp

Soooo is no one gonna talk about Gemini Deep think? Is it Blacktooth? Will it save the hobby?

>>106105356
no its not, its basically googles answer to o3-pro
aka not for instant messaging but for X minutes research

>>106105353
rare teebs W

I've been using Microsoft Copilot recently because you don't have to sign in. I was using ChatGPT but it kept wanting me to sign in. I think Gemini requires you to sign in too.

Sure Microsoft are probably still fingerprinting me but whatever.

>>106100314
Well first of all I rate positivity bias very highly on my penalties list since it's very insidious and ruins outputs without you realizing. If I simply tell R1 what I want to do, R1 will do it, exactly as I wanted, without complaint or issue. If I tell sonnet that, sonnet will spend 2-3 responses erring on the side of caution and wondering if I really meant what I said. But that's directormaxxing, if my regular sfw roleplays R1 still writes decent prose but only when there's enough evidence for what it's doing. Like if I have 40k context of a long adventure and this is the culmination of that, then I enjoy its responses better than sonnet, because sonnet will heavily bias the whole scene in favor of you in the most unsatisfying way possible. I still remember doing a whole love triangle scene and shitty sonnet having the guy quit right at the beginning because "he sensed there was a connection between you and backed off" while R1 actually has him fight back. It's fucking bullshit man.
>>106100393
Now if this is the type of thing you're saying R1 isn't good at, I can understand but I'm not a sizefag so I'm not doing anything super complex anyway. Are you saying sonnet is good at this? Sonnet is capable of keeping secrets? Because I remember another roleplay where I had to beat sonnet over the head with the fact that characters shouldn't be gossiping with accurate info about events that happened behind closed doors.

>>106105552
does Bing refill video gen credits? I have been wanting to use it for the longest time

>>106102384
Finally, someone who gets it. It ruins any chat it's in. It's not about sfw or nsfw. It ruins even sfw roleplays.

>>106104496
Does the hairstyle for the white-haired girl in the second to last row have a name?

Fucking AI is giving me some of the most hardcore lines.
"Love isn't soft, it's choosing the fire together" and "I want a king who knows how deep the rot goes and still chooses to build a throne in hell for us."

>>106103849
>streaming
Sounds like crazy schizo bullshit that either slopus or R1 would write.

>>106105735
hair over one eye.

>>106104496
all of them. so very hairy

>>106105751
>I want a king who knows how deep the rot goes and still chooses to build a throne in hell for us.
>hardcore
underage or indian, call it

>>106105801
I meant the back part but thanks.

>>106105986
https://danbooru.donmai.us/posts?tags=spiked_hair
https://danbooru.donmai.us/posts?tags=messy_hair+&z=5
https://danbooru.donmai.us/wiki_pages/wavy_hair

Try spiked hair, wavy hair and messy hair.

How important is the "scenario" box for characters (deepseek R1)? A card I picked up of chub had it basically written into other sections and I'm wondering if I might have gotten better performance if I moved most of it into that box.
Also where and how would I format the writing style to be a very slow burn. I've heard writing styles get put in Author's Note, but I don't remember if that was for ancient models or if it still applies. Cherrybox I think had some in it's prompting instructions somewhere.

Is Deep Think also on API? It probably uses a Blacktooth/wolfstride/stonebloom variant as the base model

>>106105699
true

>>106105382
>>106105356
deep think actually is 2.5 ultra and it will save the hobby, unfortunately
>We are also working to release Deep Think with and without tools to a set of trusted testers via the Gemini API in the coming weeks, to better understand its usability for developer and enterprise use cases.

>>106106012
Thank you.

>>106106069
>in the coming weeks
I'll hang myself

>>106106044
nobody knows what it's for

coming straight from YOUR HOUSE

>>106106116
same
logan kilpatrick on x mentioned releasing it to api next so maybe itll come sooner but idk

>>106106170
Mr. Ant Tenna?!

gemini tends to make you a evil mastermind, no matter your action, you are a villain and good deeds where because of second intentions OR they were themselves the evil act (chains, a different kind of cruelty).
People tend get 'broken' for literal nothing and then they become chatgpt (analytical, robotic, etc.)

> Clarify the parameters.
oh yeah, gemini rp, magnificent

>>106105751
do never approach women that speak like this, they just want you to turn a blind eye to their bad behaviour

>>106106215
true

>>106106248
it's irritating. especially this dementia of turning people into chatgpt.

>>106106044
it's not important, it just sends the text in the field with a certain text in brackets appended. You can manually do that with prompt injections

>>106106262
you let it speak roughly analytically ONCE and it just keeps cascading and getting worse and worse, its kind of funny in a "haha what is this shit" kind of way.

>>106105299
Anon why did you lewd her.. I just wanted to save her and help her love life again..

>>106106287
it's not funny. it literally kills chars personality. not even chatgpt with "be direct, respond only the absolute minimum" is this schizo.
> gemini smut
"Proceed with the primary function. I am prepared for entry"

>--- Checked 546899 keys | 546488 were invalid ---
ouch
guess that was largely a waste of time

>>106106215
have you tried being quirky while being evil? that usually does the trick for me, like pointing out my hate-wife has been awfully quiet, or just yanking her into my lap for belly rubs

>>106106311
robotfuckers won

>>106106319
looking for oai keys on github? then yeah you shouldn't expect to find much because most of them get auto revoked

already tried horizon, its writing sounds slop ngl lack of creativity

>>106106322
nah when it decides I'm evil it loops. any action is an action of cruelty, it literally resignifies my action in the prompt.
>>106106340
the funniest part. it's not a robot card.

this hobby isn't dying
it's already dead

>>106106319
Where did you look?

>>106106393
i wont tell you

>>106106422
Why?

>>106106367
>loops
at this point just OOC that {{char}} had a reaction to what you did, or that she momentarily becomes her older self

>>106106427
im evil

OpenAI will save the hobby. Local SOTA and closed-source SOTA, both at writing and RPing. We won

>>106106457
> I do an action that is literally positive out of pure goodwill (it's literally a good action, without second intentions, without debt)
> It was the worse cruelty that he has ever inflicted. They were used to the rules of the devil, but the false kindness only to make it hurt more later, it's worse... the uncertainty [...]

>>106106508
Gpt-5 will save you

is it still possible to access claude sonnet 2024-10-22 or has it been entirely deprecated?
>>106103155
cai doesn't even work for myself anymore
don't know why nor do I really care to fix it

>>106106489
>we
No, I wonned. You losted. Closed AI won't do shit, but I've been sending my kino opus logs to deepseek every day. Soon DS will btfo you all.

>>106106565
>is it still possible to access claude sonnet 2024-10-22 or has it been entirely deprecated?
Huh? claude-3-5-sonnet-20241022 isn't deprecated, it's still available everywhere.

>>106106601
alright, just checking. preferred it for the absolute maniac shit it came up with at times

minipopka is complete slop

What's the opposite of a trvthnuke?

>>106106732
deceptionblip?

>>106106732
actual answer: a blackpill, because truthnvke is just the next generation's word for redpill because boomers and incels made redpill cringe

sorry, blackpills are true. bluepills are the ones that are false

> https://rentry.org/ext_blocks
I don't get it. What it does?

>>106106732
a bad take

>>106106774
what about green pills?

>>106106819
the traditional 4 pills are black, white, blue red
no idea what green pills are. maybe the green pill is that you should smoke some weed and stop caring so much about all this incel stuff kek

grok is genuinely indefensible on all levels. i gave him a 140k text file which contains about a year's worth of smut and good swipes that i had on 3pus over the years. what i wanted grok to do is i wanted him to extract all of the swipes that centered around a specific character (katya), and paste them for me, because the text file itself is a disorganized mess.

grok proceeded to have a furbo type meltdown where it started spamming the same sentence repeatedly. this was done on 0.05 temperature, and if it's not spamming shit repeatedly, it ignores my fucking instructions and start copy and pasting the entire text file verbatim, including excerpts that aren't about katya at all.

gemini did it on the first try, and fucking easily. so called smartest model my ass.

>>106106732
jewish psyop

>>106106732
a fvkenuke? a fvke if you will?

>>106106819
Taking one of the green and purple ones once a day is very good for you.

>>106105552
>ChatGPT sign in
As much as I hate OpenAI signing into free ChatGPT is handy for a bunch of stuff.
It remembers prior convos and pulls them in. I've done 3-4 session chats on some tech project and it will pull details from them using something like RAGs
Multi-device sync to one account if you use the app
ChatGPT is censored for SFW convos anyway so not concerned about wtf they do with the data I send
If you've concerned about privacy you should be running local anyway.

deepthink takes 20-30 minutes per request

>>106103155
>AI companies are training their models on our coom logs. The AI is learning from our coom logs.
bros, we wonned

>>106106790
It generates the block at the end of the message with a different prompt, like the ones that have stats in them or the infobox in CherryBox.

>>106107064
deepthink or deep think?

made a program using gemini to stack webms

>>106106732
in ruincord they call it a wrong bomb (wromb)

>>106107095
>ruincord
child grooming station, please kill yourself if you're a part of it

>>106107089
do you think they still keep their jawbones somewhere?

>>106107104
this is garbage and an ffmpeg bash script would be better. if you're vibe coding use Gemini to make it so you can just select a 1x1, 2x2, 3x3, 3x5 etc grid and then it opens up empty squares and you can just drag and drop files in and THEN it would actually be more useful than just using console commands

GLM is pretty good, just wish it were a little more creative

in sillytavern how do I apply a regex to all chats EXCEPT the last one?

>>106107239
hover the ? for more info

>>106107249
so I set max to one?

>>106107126
yes they're in the jawbone tower

>>106107155
i don't know how to go about doing

>>106107263
cope

https://www.theinformation.com/articles/inside-openais-rocky-path-gpt-5
paywalled but The Information say that gpt-4 -> gpt-5 isn't as big an improvement as gpt-3.5 -> gpt-5
it's just incremental improvement

>>106107266
just ask gemini dude. i'm sure it can help you make a drag and drop GUI if you got this far.

Call this a post nut clarity or something.

I've been using character.ai since the beginning and have a very lengthy log with a certain character which may include some very heinous acts of perversity and debauchery. However I no longer want this to exist, it does not deserve to exist. Should I manually delete the chats or is deleting my account the best way to get rid of it?

>>106107079
cherrybox? could you share an image as example please my good anon?

>>106107360
it will exist on their servers forever. they're owned by google now. they have snapshots of entire databases. not only is your chat saved, multiple copies probably are.

this isnt a troll post. i'm assuming you're young since you used cai so please go down a wikipedia rabbit hole on the global survelliance disclosures in 2013

>>106107249
>>106107257
anon I...

>>106106508
Oof, I feel this. I endlessly tried an RP where gemini plays a newly acquired slave and I'm someone that was born into it and believes it's better than freedom (think human domestication guide but different). Any act of kindness that isn't a direct attempt at escape made me worse than Hitler. And the moment I show the slightest happiness or even acceptance of my situation, Gemini's character instantly mindbreaks and starts babbling elaborate analogies and deconstructions. And when I played along with the robot-mode stuff, giving Gemini's character an ultimatum that I won't support him if he continues to treat me coldly, he either goes full Machiavellian trying to manipulate our owner by framing me, or goes psycho and starts self-destructing. Even with system prompt instructing against this. Even in a world where being a slave/pet is genuinely better in many ways than freedom (genuinely loved by owners, can still use net, play vidya, speak out, do DIY, etc. but don't have to work or taxes or anything). It's seriously wild shit.

>>106107263
>The law requires hospitals to store surgical remains in special containers, and transport them to be incinerated in designated vehicles.
?? It's just bones, right? There ought to be a way to dry them out and sterilize them.

>>106107377
well as bad as my logs are, I'm sure someone else out there has much worse

>>106107391
well yeah, i worship toddlers dude

Reminder that this gave me erectile disfunction and it WILL happen to you.

>>106107296
well yeah no shit they've been bumping up against the limits of transformers for a year and a half, shifting focus to make things cheaper since they can't make them better. also gpt-4 is probably still the second largest LLM trained behind 4.5 while 5 will likely be smaller and more efficient.

>>106107296
>gpt-3.5 -> gpt-5
gpt-3.5 -> gpt-4*

Assuming horizon-alpha is their new writing model, why the fuck is it still so shit at coming up with names?

>>106107384
> NOOOOOOOOOO I DON'T WANT TO BE A SLAVE
> then go
> NOOOOOOOOOO DON'T DISCARD ME
> then stay
> NOOOOOOOOOO I'M JUST A SLAVE
It's the contrary of claude (where people are happy regardless the disgraces infringed upon them)

>>106107363
https://rentry.org/CherryBox
But this one is just the preset and not that extension.

>>106107478
This shows perfectly the current state of the hobby. Everything is shit

>>106107478
modelswapelites already won

I don't swipe because I have Horizon. The best model.

>>106107616
I use
> deepseek: intelligence
> chatgpt: smut
> dark roleplay (sad/consequences/violence/etc): claude
> prose: pygmallion
Also each message with a different preset.

>>106107509
So it's a fixed chat to the bottom? neat. Although regex: one to show as collapsable, another to send to the AI only the last block. Will do too.

Where is this GPT-5 then

>>106107650
>chatgpt
>smut

Horizon > Opus

Another day. Another model that is shit compared to Claude. Trannies always lose.

>>106107881
yes even 3pus

>{{char}} abc
>{{char}} xyz

Another day. Another model that is actually on par with Claude that slopusjeets will ignore and call slop because it isn't from Dario. Claudetrannies always get retarded.

>>106108170
where's the one 123?

Why the fuck are all models obsessed with the air being (insert shitty adjective here)?

>>106108204
Like their model, opus-sloppers are limited to only a handful of personalities.

>>106108204
How can I keep deepseek R1 under control? It keeps giving me larger and larger prompts and it keeps overusing lists and enumerations.

>>106108226
Because all of them are trained on the same garbage fanfics

>4 iq slopusjeet assumes the only reason i think his bad model is bad is because i dont have it
Brain damage.

i have opus btw

>>106099701
> Why Use Silly Tavern on an SBC?
Running Silly Tavern on a Single Board Computer (SBC) offers several advantages:

Intranet Access: Easily access your models and cards from any web browser within your local network.
Centralized Management: Consolidate all cards and API endpoints in one location for access from multiple devices.
Learning Opportunity: Gain hands-on experience with lower-level networking concepts (focused here on Ubuntu).
Compact and Efficient: A headless setup (no monitor or keyboard) keeps the system compact—about the size of a deck of cards—and uses only ~15W of power. It can travel with you, running off a small wall adapter or even a USB battery pack.
Lauchpad for internet access: Good starting point to set up a machine for internet access that's not your main computer.

https://rentry.org/SillyTavernOnSBC

Opus is incredible, it's sad only high IQ people are able to use it

>>106108238
read >>106108204

>>106108226
Because it's an extremely generic phrase very commonly used by shitty foid eroticaslop writers that can be used as padding in pretty much any situation, so the model has a lot of opportunities to choose it as the next token sequence. Basically the same as any other ism.

>>106108317
holy trvke

>>106108304
Abso-fucking-trvklear trvke

>30 seconds apart hours
see you later

>>106108292
>Intranet Access: Easily access your models and cards from any web browser within your local network.
You can also use something like tailscale to very easily access it from ANY network or device.

>>106108351
If you wanna set up your own wireguard network, you can. It's just not as braindead easy.

>>106105273
Good idea and interesting concept. Shame Gemini is slop in the end. But very creative.

>>106108369
Clean your mouth before talking about my gemini, you son of a bitch

>>106099700 (OP)
New google snapshot?
New horbo snapshot?
New Claude?
Or are we still stagnant?

Gemini - jeetslop
Gpt - safetyslop
Deepseek - chinkslop
K2 - chinkslop but worse
Grok - retarded, slopped, virtually worse than Deepseek
2 years later Opus is still the only model worth using. But only the non-poor/non-retarded ones can use it.

>>106108434
Trvke

Somewhere, an anon seethes. Somewhere, another copes.

>>106108453
The life cycle of the nopustroon: seethe, then cope. Repeat. Forever.

>>106108434
what have you chatted with opus about lately?

>opus bait hours with the guy samefagging about how true it is
see you retards tomorrow

Kinoseek wonned

I have opus btw

You niggas don't even have Llama

>>106108244
pls respond

>>106107263
kek, is this real?

>>106108551
I literally have opus

>Nopus troon
Underage detected.

>>106108677
>HURR I DON'T HAVE OPUS
You sound like this

>>106108682
You are literally a foid

>>106108677
>I got him banned
Announcing a report/Impersonating 4chan staff btw

>>106108693
>HUR HUR I CAN'T PAY FOR OPUS
You really aren't helping yourself

>>106108704
Reading comprehension is not your strong suit.

>>106108693
And?

>>106107384
>plays a newly acquired slave
if it was one of those scenario cards then it's your fault
when given freedom, the model will go for the most basic personality ever

>>106108346
>tailscale
Thanks; I'll check that out. I've been trying to figure out a safe way to expose it to internet. That might be the trick.

>>106108860
tailscale doesn't seem to work with vpns

>>106108845
>You're about to witness history in real-time
Do marketers get off on sounding incredibly retarded?

>>106108391
Show good prose first, nigga.

>>106108736
It wasn't a pure scenario card, and while initial attempts at the scenario started with the guy being vaguely defined ("eh's a cool guy who isn't afeared by anything") later attempts drilled into five+ paragraphs of background, personality, and narrative parameters. At my most heavy handed, the best I achieved was him constantly flip-flopping between human and robot, and still being reluctant or upset with anything that wasn't raging against his chains. Instead of being crazy it was just dull, so I stopped there.

>>106108896
these people aren't even getting paid for it they're just braindead cultists

>>106108896
>Do marketers get off on sounding incredibly retarded?
In my experience, yes.
>>106108902
And this; it's worse because they're doing it for free.

>>106108292
I'm not trying to sound like a baiter, which is hard in this general, but my phone is a single board computer that comes with a battery which I can take anywhere, I can access it from multiple devices, and it can access the internet, all without needing to be headless. I never understood the use of single board computers unless you're being intentionally useless for style points (e.g. cyberdecks.)

>>106108964
they are extremely low energy consumption and not using your phone battery. I like to switch between desktop and mobile when using ST, running it on something like a pi lets me do that seamlessly instead of having a separate instance.

>>106108964
Taking an SBC with you when you have a phone is retarded, yeah. But if you just leave it at home as a server and connect to it remotely, you get massive advantages such as being able to access it from anywhere on any device without running anything on said devices, and all your data is always there and is easier to manage and backup.
That said, considering how expensive most SBCs are nowadays, you might as well buy one of those cheap mini PCs instead to use as a home server.

>>106106768
You mean blue pill?

>>106108896
Low IQ messages meant to be read by low IQ subhumans.

>>106107391
In addition to that anon, the people who will come to arrest you for your logs also do unspeakable things to toddlers. Such is the reality we live in.

>>106108964
It's a reasonable Q. I set the SBC up so I could access ST from my PC, tablet, or phone, and have everything sync up exactly as >>106109018 describes. I don't like leaving ST up and hosted off my daily driver, and SBC power consumption is almost zero.
>>106109043
>cheap mini PCs instead
Agree; I've set up ST on an OrangePi Zero, which is $20, but the one I use was ~$50. And prices even on the OPi are going up.

>>106107650
>dark roleplay
>claude
I assume you mean opus but even then only 3pus. Deepseek is actually your go to for this and what will happen is that it will go further than you wanted it to.
>chatgpt
>not gpt
>smut
Oh I should have noticed you were a memeing newfaggot. Fucking hell take your (You) for the S-tier bait then.

>>106108244
Use 0528 and change your preset.

>>106108248
You don't have opus though??

>>106108292
>running node on a SBC
this sounds sinful, there's a reason why the node packages were outdated

>>106109018
I can also do that by connecting to my desktop from my phone or vice versa via my Tailscale setup.
>>106109043
Cheap mini PC sounds like a better deal, yeah. The whole SBC situation reminds me of when shucking external drives was cheaper than buying drives new.
>>106109157
>SBC power consumption is almost zero
I guess if you're using RISC to run ST it makes more sense than just running it on your x86 computer, but yeah.

Opus?

>>106108244
I had to add below to my very short main prompt
> Respond in prose with 4 to 6 sentences, allowing {{user}} to guide the roleplay. Avoid responding with bullet points.
To keep nu-R1 from doing lists and going on forever.

>>106109181
I'm using R1 0528 free. What preset should I use? I'm use Cherrybox right now.

>>106109306
Sorry for my ignorance, where do I change the main prompt for Chat Completion? In the preset? Or do I write that every time I post?

>>106109311
You should be writing your own. See what outputs you get that you don't like and modify the preset or start from scratch.

>>106109290
pay for it?

I have openrouter API keys and I want to use deepseek, how do I get it to work on sillytavern? it's not like how I'd use my old gemini keys or any old proxies

>>106109576
Fuck off locust

>>106109581
nigger you wouldn't know a locust if it fucked your AI waifu in front of your face

>>106103798
It was trained on OpenAI logs, so yes, it does have some refusals trained in. But all you have to do is swipe and it'll do whatever degenerate stuff you have in mind.

https://www.wired.com/story/anthropic-revokes-openais-access-to-claude/
anthropic banned openai
>“Claude Code has become the go-to choice for coders everywhere and so it was no surprise to learn OpenAI's own technical staff were also using our coding tools ahead of the launch of GPT-5,” Anthropic spokesperson Christopher Nulty said in a statement to WIRED. “Unfortunately, this is a direct violation of our terms of service.”
>According to Anthropic’s commercial terms of service, customers are barred from using the service to “build a competing product or service, including to train competing AI models” or “reverse engineer or duplicate” the services. This change in OpenAI’s access to Claude comes as the ChatGPT-maker is reportedly preparing to release a new AI model, GPT-5, which is rumored to be better at coding.
>OpenAI was plugging Claude into its own internal tools using special developer access (APIs), instead of using the regular chat interface, according to sources. This allowed the company to run tests to evaluate Claude’s capabilities in things like coding and creative writing against its own AI models, and check how Claude responded to safety-related prompts involving categories like CSAM, self-harm, and defamation, the sources say. The results help OpenAI compare its own models’ behavior under similar conditions and make adjustments as needed.

>>106109717
Begun, the AI wars, have.

UK asking an ID for everything is actually good. I want to see underage troons seething more

>>106109717
>check how Claude responded to safety-related prompts involving categories like CSAM, self-harm, and defamation
your favorite model is being delayed for this btw

>>106109717
this means OAI was exposing their codebase to their direct competitor... which obviously isn't as big of a deal as say, model weights, but still holy shit lmao

>>106109717
Dario flex
so claude is indeed the coding king, opus will never be good for rp, only a souless machone

>>106110043
deepseek will rise

R1 keeps trying to make me taste my own cum after I nut in a bitch. All the presets are cucked

gemini/sonnet, i just want a giant girl sitting on me to murder me without me going "erm why aren't i dead yet"

gemini/sonnet, i just want a focks with a giant bushy tail to wrap it around my face and smother me with it without me going "erm why aren't i dead yet"

>>106110167
Are you underage, by any means?

>>106110167
sounds like a you problem

>>106110166
What are you, gay? You can change any part of a preset you're using btw.

>>106099700 (OP)
baking in 5

>>106110340
>>106110340
>>106110340
>>106110340
>>106110340

airi has a very very hairy pussy

predesu

What are some good Uncensored ERP Models? If you have similar specs to mine please give a recommendation. I'll also take any fantasy adventure models if you wanna shill for your favorite one:)
I'm using (XortronCriminalComputingConfig.i1-Q4_K_M.gguf) a model derived from Blacksheep, but it's a little dumb and keeps repeating phrases and looping its output.
I WAS using the non-imatrix Q8_0 version (XortronCriminalComputingConfig) but it took like 30-90 seconds per response to give me a full output on Koboldcpp chat window. It's really good honestly, It kept remembering characters personality, would add spice and interesting dialogue to mundane situations but it's unsusable as a coom model due to me going soft constantly. tragic.


I have a 5060 ti - 32GB Ram and Ryzen 7 3700x - I know my cpu is a bottleneck : It's tough.

Question from a humble coomer, should I pay for the Deepseek API or just pay ten dollars for Open Router?

>>106110386
Check here anon they should know better than us >>106108045

>>106110386
you can jailbreak pretty mucb every model out there, the problem is it lowers RP quality

>>106110166
It's just trying to guess what you'd like, anon.

>>106110413
OR offers 50 prompt daily for free
DS api offers the model at full context for pennies, I would go for DS api

desu

boku

Nice desus.

>>106100728
6 fingers desu.