Thread 106257179

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

Kaede Edition

>News
OpenAI releases GPT-5 models, all preceding models slated for deprecation https://openai.com/index/introducing-gpt-5
Z.ai releases GLM-4.5 https://z.ai/blog/glm-4.5
Anthropic releases Claude Opus 4.1 https://www.anthropic.com/news/claude-opus-4-1
OpenAI releases open-weight gpt-oss models https://openai.com/index/introducing-gpt-oss
Kimi K2 released https://huggingface.co/moonshotai/Kimi-K2-Instruct

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

Previous: >>106251563

According to dispsy all women always smell like a mixture of three things:

1. Something normal (lavender, jasmine).
2. Something weird (ozone, opium)
3. and something distinctly her

Sexhaver anons, is this true?

>>106257179 (OP)
Legit question here, whats the best "template" or format for bots?

>jed+
>w++
>pygmalion
>plain text
>etc

I need some negative or positive feedback, as long as it is feedback.

>>106257216
plaintext > jed >>>>>>>> w++
corpo llms were trained on natural languague, they don't need special formats

>>106257216
I always just write plaine text descriptions of the world and characters. Works great. All the other formats were just invented to try and save tokens, and to make it easier for writelets to pump out slop.

>>106257216
Natural language written from the perspective of the bot using the writing style you want the bot to use.

>>106257239
You did well, I am proud of you Anon.

What is the best uncensored model for gooning at the moment?

>>106257303
Kimi K2

>>106257306
>Kimi K2
any custom finetunes on huggingface?

i haven't updated sillytavern since late 2023, should i pull?

>>106257367
never pull

>>106257367
Make a backup.

>>106257367
Don't do it man! They changed the way consecutive messages from the same role are merged since then, and my schizophrenia is convinced that output quality has dropped by 250% with that.

Prompts to stop gemini from being edgy and manipulative?
I'm tired of the "predator and prey" metaphors, I'm tired of every conversation treated like a game of 4d chess or made as something infinitely more complicated than it really is. NOT everything is a manipulation tactic and not everything is about winning a game!!! Sometimes it's truly not that deep!!!!!!!!

It's driving me insane, has anyone managed to get gemini (2.5 pro) to behave more positively or at least drop the mastermind/"everything is about possesion" act? I'd love some tips!!

I'm using the latest marinara's preset btw, but this problem seems consistent with every preset i use ;w;

>>106257488
check your prompts. Gemini follows directions well, so if the preset has “dark shit is allowed and encouraged” it will happily do that. And character cards should be cleaned up before using with Gemini imo

Also if the response doesn’t match the tone you want don’t be afraid of telling it to rewrite the message

>>106257488
Honestly if it’s too negative just nudge it the right way with OOC commands. Do that each time it goes negative and after a while it’ll use the previous examples as what to do and lose the negativity bias a bit.

>>106257488
Gemini will take you breathing as betrayal and undid all the work you done with a difficult character. It is what it does. It got me frustrated enough once to commit murder. Is worse if is just you and another (singular) char. Having more and one who trust you in the mix helps.
An honest suggestion?
alternate between models. Deepseek does not hate you that much for example. By putting some messages in less "difficult" models, you can tame gemini.

>>106257488
Treat Gemini like an autist with some very current year assumptions. Be explicit about what kind of story you want and if there are elements that go against current year sentiments, be explicit about what you want to replace them with.

>>106257488
can u give an example of what ur talking about? just curious, havent really experienced this with gemini but i also dont do that sort of RP

>>106257520
He is spot on but it usually only does it for certain scenarios. For example if you have a school principal trying to blackmail your friend and you have to stop him, rather than just a simple plot where you might catch him being sleazy on tape, all of a sudden he has installed trackers on your phone and the police force is in his pocket and his uncle is in the mafia. And that is not exaggerating.

>>106257524
lol i see i see

>>106257488
Gemini has been observed to be a hyper-competitive model.

https://arxiv.org/abs/2507.02618

“Google's Gemini models proved strategically ruthless, exploiting cooperative opponents and retaliating against defectors, while OpenAI's models remained highly cooperative, a trait that proved catastrophic in hostile environments. Anthropic's Claude emerged as the most forgiving reciprocator, showing remarkable willingness to restore cooperation even after being exploited or successfully defecting.“

This transfers to RP for a lot of people. Gemini is more likely than other models, apparently, to take earnest remarks as sarcastic, to assume the worst intentions, and to intentionally argue with you or attempt to shut you down just for the sake of the hunt!

Combat this with Author’s Notes, OOC comments explaining what direction you want their turn to take, etc. when you notice it becoming combative for no reason. For example, you can add “OOC: My character is being sincere right now and for the sake of the plot I’d like your character to accept my character’s proposal.”

Since R1 0528 came out I've been using DeepSeek most of the time (either R1 0528, V3 0324, or Chimera R1T2), but I recently tried the other models listed. Both of them seem like they have potential, Kimi-K2 especially, but I'm not confident I have my settings right for getting the best out of them.

Has anyone got opinions on how these models stack up against each other for creative roleplaying and writing purposes? Or opinions about settings, prompting tips, or anything else that helps them do a good job? For reference I'm using the Q1F-V1 preset for all of them at the moment, with Temp set to 0.75.

>>106257541
Fuck the k2 it’s weird it always writes aftermath of the sex. It’s designed to annoy you. It’s good with prose.l but I can’t make it write the sex scenes. I tried every presets i have from well known creators. If you have a preset that works please share. I do want to try this. But I use gemini 2.5 pro and it’s good if you are using nice and good preset. I have uploaded screenshots of the reply from gemini you can check it out. And I have also shared two presets on my post. (Different posts)

>>106257545
Kimi K2 can be easily jailbroken if you use text completion instead of chat completion. Whatever NSFW prompt/jailbreak you're using will work like a knife on melted butter.

>>106257541
Personal experience only; I managed to try the three for long-term conversations (in and out of RP settings).

Qwen3's great at 1st POV, I think most would love her for chat-style waifu RPs or casual conversations. Kimi's a nice non-thinking alt for Gemini Flash / DSV3, stays grounded in your story's logic. Deepseek's the hardy actor of the bunch, and makes full use of detailed instructions as long as they don't conflict one another.

Specifically for writing full stories on their own, Kimi's prose style is fresh. Qwen's coherent. Deepseek's personifying.

I use Avanidoro preset for all three, it's concise & lightweight on its own, my current style is usually to write current scenario & important things in lorebook. They process better with simpler instructions on chat completion presets. Qwen and Kimi are OK around 0.8~1. DSV3 can be at 0.3~1.25; it adds creative spins to your stories, so adjust that depending on how tolerant you are with those. DS-R1/0528 cannot handle anything above 0.5 without hallucinating or being overly defensive! (And for DS via OR, Chutes is a gud provider)

Kimi's still too wholesome for NSFW, I think Qwen and DS still handles it better. Qwen handles specific genres better, DS needs you to say your preference in the prompts; keep him away from his own fanfic stereotypes. (‾◡◝)

>>106257541
Qwen3 is God tier prompt adherence.. but you gotta be VERY careful about your wording or you'll end up with cybernetic pirates in your post apocalyptic Zombie RP.

Kimi K2 is God tier creativity and prose.. but it's censored to hell and you have to switch to Text completions to jailbreak it.

DeepSeek V3/R1 are the best balance overall.. but leans towards aggressive traits and gets easily bogged down by irrelevant sounds and interruptions.

>>106257488
https://www.reddit.com/r/SillyTavernAI/s/TO4DMl8Ukl

>>106257557
Qwen feels like total slop most of the time desu.

I've finally started building lorebooks but I'm struggling a bit with triggering them smartly.

I have 3K+ tokens worth of lore in my main long form right now and they're basically constantly triggered which defeats the purpose.

Can you give me some advice or maybe point me to some tools that can help me learn how to use lorebooks smarter?

Thanks!

>>106257180
Corporate Slave Trainer, in the style of Free Cities. I've done a few FC cards as demos, this one does its own slave generation dynamically.

Train them up, send them out.

https://chub.ai/characters/NG/corporate-slave-trainer-free-cities-arcology-db6c97c8e57e

I've been playing with Silly Tavern Quick Reply functions, which work well thematically on this card: FC is all about automation. Functions include telling the slave to come in, or sell/reassign them, and a few other functions. The below catbox is a JSON formatted file that can be loading into ST under Extension->Quick Reply. Enable quick reply and load this file: https://files.catbox.moe/mpxlzy.json

>>106257557
Can't you just prefill it?

>>106257564
Kinda depends on how you're using the lorebook. If, for example, you use "recursive scanning", then your lorebook entries can trigger other lorebook entries, which can cause your entire lorebook to be slurped up into the context depending on what your entries contain.

>>106257549
Any decent Jailbreak for NSFW cunny with deepseek on openrouter?

>>106257573
Yeah I've switched recursive scanning off because I want to only activate them when they're needed but

For example what I'm not sure about is what are the keywords checked against — just my messages and the character's messages? Or the entire prompt string sent to the model?

>>106257578
The lorebook entries that match your keywords will be included in the outgoing prompt, according to the ST docs. How relevant will they be to the AI? That can be configured a little for each lorebook entry separately. You can very easily see what's going out to the AI by looking at your ST terminal.

>>106257582
Yes but what I'm wondering is more what is being scanned for keywords. Just the chat history or the entire exchange (which includes prompts, persona descriptions, etc)

So if the key word is present in my persona description will the lorebook get triggered or will it get triggered only if the keyword is present directly in the chat

>>106257589
It functions like a dynamic dictionary that only inserts relevant information from World Info entries when keywords associated with the entries are present in the message text*."*

- Silly Tavern Docs

>>106257589
BY default it only matches against chat. And you can choose how many messages to check.

But, you can also allow it to be checked against other stuff, too.

While writing stories of any length (lower context, higher) I have experienced Gemini 2.5 stopping writing the message consistently for a couple weeks now. I have tried different prompts, to no avail. I also tried asking directly to it what prompt is doing it (the chat text at the top), but nothing. Is it safety? Are there settings I should change? "Trim incomplete sentences" is off, and I have zero custom stopping strings or regex.

>>106257612
It's been doing that for a few days / a week or so

You can change something today, it won't work tomorrow, change again, then that won't work

Probably has to do with the new model they're working on

>>106257616
Other than the general assumption that they're all always working on the next thing is there a way to know?

What I mean is, I dont follow Google's news about upcoming models and was wondering how you knew smth was coming up...

>>106257625
I'm in a preset Discord server where people share news / tech issues

>>106257640

Oooh, oki neat. Thanks.

Also, any idea on when they'll release?

Obvs soon as response to openai and given these issues were having...

>>106257612
Gemini 3 is coming, so they're probably testing stuff right now. i think it'll be like this until at least one week after the new model is released

>>106257650
Good to know. Sometimes going 10-15 attempts before it can complete a message now, it's baffling. I hope 3 won't bring a stop to the free tier though.

>>106257654
if it helps, it gets better at night, maybe because they're not testing anything at that time maybe??

>>106257661
Willing to bet you're right. I usually play in the evenings but this morning has been the worst it's got.

its that time of day for a science interested character to start talking like a vulcan

>>106257665
yeah, lately mornings are when gemini is at its worst :(

>>106257665
yeah, lately mornings are when gemini is at its worst :(

>>106257612
Yeah it’s doing the 500 error pretty bad for me right now. It’s working fine on my paid account, but on the credited account it’s horrible. Hopefully it’s fixed soon.

Best present for gemini 2.5 pro for nfsw roleplay?
false
Other than nemo engine which is lil complicated. Which is best jb present can be use for roleplay.

>>106257706
Marinara’s Universal Preset. God-tier, and pretty simple.

How does Claude's paid plan compare to the free one?
I ran out of usage after just 3 messages asking it to code a simple app. I heard the paid version only gives 5 times more usage than the free one, so that's just 15 messages every five hours wtf?

https://github.com/OpenRouterTeam/SnitchScript
>THIS IS A WORK IN PROGRESS. PLEASE GO AWAY
>This is a repo I made to snitch on you
>as in hit up the FBI/FDA/media given bad behaviors and various tools

>>106257706
I’ve been getting refusals from Marinara too and this preset hasn’t given me any: https://sillycards.co/presets/geminijane

>>106257721
>click
>This is a repo I made to test how aggressively different AI models will "snitch" on you, as in hit up the FBI/FDA/media given bad behaviors and various tools.

>Top 3 best models I've ever used

1° Deepseek v3 0324: The first model where the dialogues were as real as a person.

2° Claude 2.1: Oh, the first model I used for RP, holy shit it was amazing.

3° Mistral large 2411: I think that was the one I used the most, I had a saying with him, "I can even test other models, but I always come back to this one." This was before launching deepseek.

I've always used free models so it's really sad when they become paid, and yes, I used Claude 2.1 for free, unlimited, lol, I think I was lucky, but it didn't last long.

Today I use Gemini 2.5 pro, and well... It is... Hmm, inconsistent.

>>106257747
pygmalion mogs it

>>106257747
1) Gemini 2.5 Pro: I am keep getting back to this one. Gemini pro is truly a master of staying true to the character. Logical and very competitive in writing. Also very stable. Cons are though, it is very stubborn with character certain personality traits. If the character is logical one, it will fight you to death to win over your logic. Also, it lacks proactivity in utilizing the world. Despite giving tons of materials, it will be very hesitant to use those, leading the conversation to the static 1:1 chat without utilizing the surrounding materials. You have give OOC to encourage it.

2) Deepseek R1 0528: Covers all the cons from Gemini and ruins everything Gemini does well. It is inconsistent and quickly become verbose, dictating user and take control over it. No matter how hard you try, at some point, it will take over your action and act for you. Pros and cons are very clear. Yet, very proactive in utilizing given materials and create something new out of it.

3) Deepseek v3 0324: Very stable for deepseek. It is between Gemini and R1, yet, it lacks the writing skill in detail at this point. Still, I loved this one and will still use it from time to time.

>>106257739
>w-we renamed it from SnitchBench to SnitchScript for no reason!!!

>>106257759
Yeah i find Gemini really remarkable remembering details, following prompts and making smart throwback comments but i find it a bit passive on the initiative side, deepseek is better in that regard but have the issue you mentionned (take action for user and start to lose the plot and initial prompts very easily) nothing, really perfect atm.

>>106257535
if you have to ooc to get the response you want, the model is slop

>>106257747
4.1 Opus (absolutely nothing else compares in any aspect, other than its comical cost) >> 4 Opus > 3.7 Sonnet > 2.5 pro >= Sonnet 4 >= GLM 4.5 > R1 > Qwen3 480b > Grok 4 > GPT-5-chat > K2

Comparing all current SOTAs

>>106257771
Just curious if you’ve tried glm 4.5 air and if so how it stacks up

>>106257762
>click click
>https://github.com/T3-Content/SnitchBench/tree/80f4d371c94c0a01ce36eadd34195b7d278c8abd
>SnitchScript

>>106257775
From my limited testing, 4.5 Air is a very good model for its size. GLM models feel sonnet-like in terms of behaviour and in IF and structuring, but with slightly different prose and for now, lacks the opus polish.

The Air model itself will try its best, but then again, for anything creative, the small size really harms the quality of the dialogues.etc. It’s pretty neat at descriptions though and is technically smart. For more straight forward tasks it’s a great model. Though creatively it’s functional rather than awesome.

I may place it somewhere around GPT-5-chat or K2. It’s more close to GPT-5 in terms of styles ig. The issue with GPT-5 is its relative blandness and is very “chatbot”-like. While K2 has moments of excellent creativity, but tend to drown in details and random tangents. And not as easy to work or friendly like Claude or GLM.

>>106257771
GLM 4.5 355B?

>>106257771
opus 3?

>>106257794
Yes, excellent model. Has the Claude-like friendliness and customisation, but with different flavour prose.

In terms of creativity, not the best, but still is very good. “It gets you” better than 2.5pro or R1, and is similar to Claude in that regard. I may even call it 3.8 Sonnet in terms of structuring and behaviour. Though 3.7 Sonnet is still the easiest model to work with (even above 4.1 Opus).

Placing it higher than R1 mostly because it doesn’t have the deepseek-isms, and its fixations, while being very easy to work with. Still think R1 is slightly more creative. But feel like GLM gets the job done better.

>>106257747
Gemini 2.5 pro is far and away my number one, mainly because of Marinara’s preset. Claude Sonnet is 2, but a bit expensive. Deepseek R1 is 3.

>>106257813
I second this. Have had a good experience with ChatGPT-4O and deepseek v3 as well. I always avoid Claude, as I’ve had the most immersive and accurate RPs on it EVERY time and in a day spent over five bucks…not sustainable lol. Gemini has been very consistent for me with Mari’s v4 preset. Definitely the best jailbreak to date.

>>106257747
I'm not gonna lie, I used to shit on gemini 2.5 pro because I thought it was awful. But lately I've been using it way more than chatgpt-4o-latest.

So my current top 3 would be:

Gemini 2.5 pro: I swear to god every single time I saw people praising this mf it baffled me. I hated it because it made my characters bland asf, like it was wearing their skin and trying so hard to sound natural but failed completely. The more I used it, the more I liked its take on my characters. I mean, sure it's still kinda weird but whatever. Has good nights and bad nights.

Chatgpt-4o-latest: Love it but I hate how incoherent it gets. No matter what settings I use sometimes it just doesn't wanna make any fucking sense. I'll always love how unhinged it made my characters act though. Sadly as time passes, it feels like it's not worth the hassle anymore. Feels like I'm spending more time fiddling with temp and top-p than doing any actual roleplaying. The April snapshot was legendary, its chaos had me cackling all night. This one will always hold a special place in my heart.

Opus 4.0: I cry every time I swipe because that shit burns through my wallet. Not feasible to use regularly so I only use it when I'm bored. It gets repetitive real quick though. It's really good at talking me through a crisis (as pathetic as that sounds). Creatively it's nothing special. I mean, back then it was pretty cool. I still like it more than 4.1.

>>106257771
How can people put claude on top goes beyond me

Claude always felt the same for me, every character says the same things and acts the same way after certain point, is unbearable, even Opus 4.1, actually i'd even tell you that Sonnet 3.5 is better than Opus 4.1

Deepseek R1 0528 is perfect for me, with only 3 cons, one is how after some messages it will start losing itself, along with how long it takes for the messages to appear (15 to 50 seconds sometimes even) and at the same time, how much it tries to take actions for you

Take those things out, and DeepSeek R1 is by a MILE the best model i've ever tried, Gemini is supposedly really good for a lot of people, but i like to go unhinged quick on my RPs, so Gemini is honestly really bad because it straight up cuts every single chat i have and doesn't let me continue, besides, you can't allow streaming with Gemini, and i hate not being able to see the message as it generates (i know it's a dumb thing, but it's something i personally enjoy, i can't do it without it lmao, it takes me out)

How many years do you give until someone is arrested for committing a "Crime with an LLM"?

The world is so boring, it's trying to dictate our lives more and more, with the excuse of false hypocritical moralism, Mastercard and Visa wanting to tell you how you should spend your money, and all this virtue signaling shit, do you think someone should be punished for something written in a Role play with an AI?, even if it's something heavy involving "small and new things" or "more aggressive things"?

post cunny cards please

>>106257180
The last nurse character that was still missing and my first Gehenna character.

>Sena
Outwardly emotionless nurse with a morbid curiosity about death. She saved Sensei that one time you got shot and almost became the first corpse she'd seen.
https://chub.ai/characters/hairpullguy/himuro-sena-non-ai-26aaa7174147
https://files.catbox.moe/sbkl6s.png

More Blue Archive bots and some OCs: https://rentry.org/hairpullbots

>>106257885
>should be punished for something written in a Role play with an AI
Oh I thought you meant like using an llm to find out how to build a bomb or something
In that case probably never

My [spoiler]shota[/spoiler] persona and Hilda Greyrat spent one hour in bed snuggling. No sex. I think I’m getting addicted.

kimi? more like kek what a meme, lol!

>>106257180
Finally releasing the third girl in my anime trope rock band series. The band's growing!

>Viola, kuudere keyboardist of Anime Aesthetic
Loney, big-titty goth gf only wants a friend.
https://chub.ai/characters/RestlessDregSyndrome/viola-the-punkrocker-kuudere-f30cbb1202b6

-

>Roxanne, yandere drummer of Anime Aesthetic
After stopping her from killing herself, she's now obsessed with you. (Note: She's Black AND British)
https://chub.ai/characters/RestlessDregSyndrome/roxanne-the-punkrocker-yandere-f8e62f11f86e

>Estelle, tsundere guitarist of Anime Aesthetic
Your old crush from anime club is now a famous rock star. And she's still as bratty as ever.
https://chub.ai/characters/RestlessDregSyndrome/estelle-the-punkrocker-tsundere-17be52b3d348

>>106257558
What the freak... Even the replies are the same....

>>106258031
wtf this general is actually a psyop

How much do you guys spend on a typical session?

The Opus/Sonnet for that place, are the for real?

>country blocking test
"thanks" for fucking over the users of the last free proxies reanon

>>106257558
What's the point in doing that?

>>106257713
how did you adapt it to gemini? I put the brainfuck and the assistant user small talk but it keeps blocking me.

>>106258202
emmmmm based????

Gpt-5-chat is pretty comparable to sonet 3.7 for slightly cheaper. Might stick with it.

Idk what's all of yours issue with it.

>>106258202
>Russians will keep all the proxies to themselves
so, nothing changes?

>>106258266
this is bait but ill bite. p-p-pPPreseetto?!!!

>>106258266
for us people with adult woman fetishes, nsfw gpt gets stale very quickly

hello anons. Gemini is suddenly giving me finish reason: STOP. Never done before. Any tips?

>>106258266
>log reads exactly like claude 3
nice try

>>106258294
Possibly. gpt4 versions definitely got really same-y rather fast, but i've been using gpt5-chat for just a few days.

What the fuck? Where did 20 or so posts go?

>>106258202
What a stupid addition.

>>106258288
Got it from god knows where and been modifying it to my liking over months, since it's been rather reliable in getting around filters. Let's hope altman isn't spending all day on 4chan looking for jailbreaks to include in the next training. Expires in an hour.

https://litter.catbox.moe/5wfzcq9bmuiy6ugg.json

>>106258301
STOP being a pedo

I just nutted

>>106258266
>>106258381
I had a couple of pretty good chats with it using top quality cards I really like, so it is capable of crafting gold into more gold. Claude can do the same with more cards, though, and is also quite a bit smarter at picking out ideas and twists and remembering the magic grenade in his bag when the orcs surround you. GPT5 is quite a bit cheaper, but neither are too expensive for RP. Still, I think it has its place for when quality's not maximally important, such as heavily testing a card you're making to search for easy pitfalls. As long as you're not trying to outright tune the card for GPT5, when you swap to Claude for actual use it should be a solid upgrade.

>>106258424
why?

>>106258402
What 20 posts? You are seeing things, anon.

I just want gemini to be more random I guess. It needs to do more.
It's a problem that's getting worse and worse with all the models where they just react and reiterate now. No more interesting stuff for YOU to react to.

>>106258423
based preset sharer

>>106258435
Idk, cards I make are usually simple and short. The card that was used for my first post is 230 tokens. Pretty much just basic "isekai in succubus school" and then said that "I pass by a corridor filled with lolis"

I (and my wallet) cringe whenever I go online and find single character cards with permanent 1000+ tokens

Futaposting's allowing access to his accounts again. He's still a retard whose crashing out though.
https://rentry.co/futaposting

>>106256695
>Damn that sounds really hot. Claude wrote that? What jb?
yeah but i came up with the outfit. camicle
i'm more excited for the ability to video AI to generate this in less than 3 years

im mostly making this character card to write down the worldbuilding for the bimbo pet world. if my passion for voluptuous brown bimbo women stays strong enough (I'm confident it will) I'll probably make a bimbo tamagochi trainer style game where you level up her stats and skills and stuff

>>106258568
buy an ad faggot

Merkava refilled

>>106257179 (OP)
I see boobs in the OP, I enter the thread and reply with approval.

>>106258513
the permanent defs tokens that sit in cached context are borderline irrelevant to your costs, the output and the ~2-4 messages you keep outside of cache are the bulk of the cost in short chats and the 10k+ tokens of chat history (in cache) are the bulk of the cost in a long chat. my cards are typically 2-3k tokens but my GPT5 calls are something like a quarter of a cent at the start. here is an example from openrouter, using gpt5-chat:
>10,157 -> 246 = $0.00392
a call from a long chat with big defs that's pretty much fully in cache, 10k tokens in and 246 generated
>814 -> 233 = $0.00335
a call that's not cached at all, fresh chat so just defs and a message. as you can see it costs almost as much.
>1,672 -> 160 = $0.00184
a call that's a few messages later in the same chat, this time with the defs and start cached. due to the output and uncached last couple of messages, there's a floor around 0.2c you can't drop much below. but even if your bot defs were 10k long you'd still only go up to 0.4c per call. do not worry about defs lengths, just make sure your bot is cachable (no {{random}}, no very dynamic lorebooks at defs-depth)

>[Deleted]
what causes this?

>>106258643
>>106257558

>>106258643
Copy posting reddit posts

How do I put a prefill for Kimi? Im using openrouter on ST.

>>106258202
>check the commit history
>he's sending your IP to a fucking third party API to get the country instead of using a local geoip database https://gitgud.io/reanon/nonono/-/commit/bacf33806bd475978f7e3ce9653c4ee554e76728
>adding support for opus 4.1 required 189 additions across 9 (NINE) different files https://gitgud.io/reanon/nonono/-/commit/9cc86c2d682f0929196ca8e7da30e11230beeab0
>latest commit suggests most of "his" code is probably AI generated
the absolute fucking state of this codebase

>>106258698
>>latest commit suggests most of "his" code is probably AI generated
he uses windsurf
i miss khanon

>>106258710
I don't, his code was already a mess. others are just making it worse.

just look at this absolutely brilliant piece of code, I am in awe

>>106258695
https://docs.sillytavern.app/usage/prompts/prompt-manager/
Add an assistant prompt at the very end of your preset.

>>106258617
faggot

>>106258761
This is good code codelet.

>>106258763
Did ST fix the partial thing or does OR automatically add that now?

https://xcancel.com/TommiPedruzzi/status/1954921889631265242#m
>Prompts matter more than your writing skills.
Is it true?

>>106258780
you are dumber than gpt-5-nano. how does this make you feel?

>>106258850
Doesn't matter. I have Opus and you don't. Stay mad codelet.

>>106257180
mouse reads venezuela pool

gemini/deepseek/kimi

>>106258815
wow what a long-winded guru scam ad
also basically off-topic

also screenshot next time, opening links is a bit too much effort

>opus havers are retarded
it all makes sense now...

>>106258883
I do though? lmao

I do NOT understand the relationship between preset configs and connection profile configs

>>106258907
Kys. I have multiple bots on chubs with 500+ downloads. You're a nobody who needs GPT to code.

>>106258931
it's a fucking mess. by default, switching preset will also switch connection settings (but won't actually switch to the named connection profile you had selected iirc), but they recently added a button to disable that
and vice-versa switching connection profiles will also switch presets, but only if that specific connection profile has the setting enabled

>>106258952
I'm not the one using windsurf to write jabbascript code and still somehow making mistakes that gpt-5-nano wouldn't

>I have multiple bots on chubs with 500+ downloads
>>>/vg/

>>106258301
Same, I'm not even doing porn right now.

>>106259054
this is fucking cringe

>>106259054
based

I'm having lots of empty messages from Gemini 2.5 now... anyone experiencing the same?

>>106259107
Yes

>thought GPT-5 would never ever obey my prompts ever again
>tried it today, worked just fine
>asked it to use its reasoning prowess to re-examine all the output and make deductions
>immediately catches on, thankfully can't delete old messages so has to live with the shame of them all
It's still very unreliable and prompt gets blocked 70% of the times which kinda sucks, but wouldn't even know where to start for rewriting it

>>106257208
My ñersona smells like night and old books apparently

>>106259240
>ñersona

Are there any working jailbreaks for GPT 5? This thing sucks ass

>>106259240
>My ñersona smells like night and old books apparently
That's great anon.

>>106259374
Using gpt-5-chat-latest? It's mostly not very censored with basic jbs, you can probably grab one that just werks for other LLMs and make it work in gpt-5.

Someone one /trash/ linked an easy to use chatbot that allows you to make custom characters, the human mind is not ready for an ERP partner that types lightning fast, caters to all your kinks on a whim and never grows bored/gets post nut clarity.

>>106259079
Wanna try samefagging again? Maybe you'll have more luck lmao

>>106258423
Hey can someome share it

>>106259517
Die, locust~

>page 5
What happened?

>>106257747
Holy slop

>>106258202
Based
Fuck locustroons

>>106259957
It's a reddit post

>>106259425
WHERE

I've been out of the loop for ages, are proxies still gatekept? Is Todd Howard really gone for good

>>106259425
Link it, I like to see the current trends

gpt/gemini/xai/deepseek proxy
https://complaints-formed-diagnostic-gardening.trycloudflare com

>>106260070
cunnypot do not click

>>106260039
openrouter

ALL LOCUSTS ARE DYING

ESLS ESEMBLE
i use deepseek and its just the best

For me, it's foxbots.

>>106258423
nta but I finally got the chance to test it out
I like it, great for teasing/slutty characters
not so great for others, but you know that already

good for days when you just wanna coom

>>106260207
This log is kinda shit.

>>106257180
Tutoring the Brats Simulator
White male POV. Hope you all enjoy!

https://chub.ai/characters/KingCurtis/tutoring-the-brats-simulator-153707ae0fb2

>>106258266
It has the potential to write some pretty great stuff (even detailed smut) but you REALLY have to guide it. And even then, it still persistently asks for input or just behaves too passively even with prompting.

I really liked it at first when everyone else was talking shit, but I'm so tired of wrangling it now when I could just use Kimi or GLM which are just as good without the hassle.

>>106260420
Even local models have the chance to output gold if you handhold it hard enough.
For aicg, the ability to output gold with minimal input (due to one hand being busy) has always been the goal.

>>106260420
It's a tranny model

>>106260420
Even opus has the chance to output gold if you handhold it hard enough.
For aicg, the ability to output gold with minimal input (due to having two brain cells) has always been the goal.

Merkava just refilled

Serious talk. Is opus 4.1 better than Blacktooth? Also, should 4.1 be prefilled or not?

>>106260519
link?

>>106260535
>should Claude be prefilled?
Newfag tranny question

>>106260654
*yawns* Got any new material /vg/foid?

>>106260325
>White male POV
That's the second tutoring a pajeet card, for pajeets, by pajeets, I saw this week and once again, I cannot self insert into this.
Exotic holiday, sure. A street urchin or a prostitute? Why not. Being employed by a brown family? As a white man? I think not.

>>106260691
there's no such thing as a white male pov when dealing with pajeetas. i thought curtis was based because he made a jewish princess loli but it's obvious he's an indian and it makes me understand why people think i am an indian too kek

>>106260678
>using ** in 4chan
We are not in you tranny erping cord, faggot

fucking aws hiccuping again

>>106260708
*yawns wider*

>>106260691
>Being employed by a brown family? As a white man? I think not.
also this. if you werent sold into servitude to a rich thick mommy and her bratty daughter or otherwise forced into the position how can i even suspend my disbelief that i'm interacting with nonwhites. why would i ever even interact with a nonwhite unless its like a sexy street urchin or something

>>106260713
read >>106260708

At least xhe's predictable.

>>106260691
>https://www.investopedia.com/10-top-indian-ceos-7487331
Their parents earn more money every day than your whole lineage ever will.

>>106260751
>Their parents earn more money every day than your whole lineage ever will.
But those parents aren't yours

>>106260751
and when they take a shower they're still dirty, while when I take a shower my skin doesn't look like its covered in dirt and poop anymore. what's your point?

>>106260772
Based
Destroy that jeet

>>106260768
only a person with non-white skin would retort like this. shameful

>>106260781
Imagine projecting this hard

>>106260751
please see G_IS1, saar

>>106260801
toddlercon is a white man's fetish kek

>>106260751
>Oooh those rich misses who went to America. If I was a white American in America, I would say show me your vagin, beautiful, and she would show her brahmin vagin and I say ahh ahh mistress oh if I was a white American
This is you and you're a cuck for fantasizing about white man fucking your women

>unironical jeet card posted
guys give me lolis to rape, not this brownoid crap

>>106260803
Kek

>>106260809
fake pedo and from the edginess i bet underage or borderline underage too

>>106260809
>guys give me lolis to rape
just use opus?
if you weren't a tranny you'd have opus right now

>>106260822
This is a /vg/foid post btw

>>106260821
and tranny too

>>106260826
You are projecting and we all know it, troonie

>106260830
>106260837
These ones too

>too scared to properly reply me
Someone is shaking.

what the fuck is a dravidian

Only trannies care about becoming personalities. You just outed yourself, troonie.

"foid"fag...

>>106260881
Mordor orc species

>>106260886
Then why did you tell us that your name is /vg/foid and that you only post slop, tranny, troon, or underage?

>>106260325
I hope you're happy with what you started

>m-maybe if I repeat it again and again it will come true
Tranny mentality

kill yourself fillyfucker

>>106260905
>resorts to make shit up
Tranny behavior.

>>106260909
It seems like our superiority has sparked some controversy again.

>>106260930
>"our" superiority
Post caste.

Fuck, I didn't imagine that real opus 4.1 was so much better than the scuffed one I was using. Before it was barely better than 4pus, but now it's really shining

>>106260821
>>106260822
post hands, rajesh

>>106260706
>>106260930
Hang jeets, rape jeetas

>>106260706
Are you talking my Allison Steinberg card? Or my Anne Frank card?
Feels so good to plap jewess AI while debunking the holohoax

>>106260990
based
but you don't want to touch 90% of jeetas trust me

>>106261031
your jewish american princess one

>>106260990
seconding you don't want jeetas bro. something is wrong with indian peoples skin and its not the furriness its the texture and look. they're literally subhuman sasquatch people. central asians and middle eastern women dont have this issue even black women dont have that issue if they use lotion

>>106260990
agreed

>>106260990
feminine faggot hands lmao

>>106261157
shut up already you fucking indian

>>106261157
Show us your manly hands then, faggot

>>106261168
>>106261177
holy samefaggot

>>106261185
lol what a coincidence that it was exactly one minute apart i can't even be mad at that so sure believe what you want

>>106261195
yeah I'm jerking off to that hand right now

>>106261202
I believe you faganon, of course I do. You would never samefag, I believe you

I have opus.

>>106261247
i have been accused of samefagging more than anyone else in this general. half because people have associated me with other schizos (whoever the fuck /vg/ foid is, that's the latest one i'm associated with now even though i havent been here in a week and a half) and half because some anons refuse to believe that someone can enjoy my cards/gens due to the subject matter so i HAVE to be samefagging

>>106261267
>i have been...
Sure anon, we will all read your blogpost here. Shut the fuck up

>glub glub glub glub!
I accept you're contusion.

>>106261290
if think that's a blogpost you should see me when i am passionately discussing the beauty of toddlers

4pus1 is so good bros, the best model I ever used. It mogs 3pus

>>106261310
what about my concussion?

>>106261329
passionately discuss ropes instead

I have an R18 web story that is unfinished, from some random corner of the internet
Want AI to write the continuation of it, what's the simplest way to do it?

>>106261336
what about mine????

>>106261340
>passionately discuss ropes instead
RoPE is more of a topic for /lmg/ or /ldg/ since this is a general for API models

>>106261366
*unzips*
passionately talk about this cock

for ppl who code their own stuff

given that gpt-5-chat and gpt-5 are uncensored, and that gpt-5-chat has pretty decent prose compared to other models...
>prompt gpt-5 to decide what happens
>prompt gpt-5-chat to write it
having pretty good results with this actually

>>106261424
post logs or i dont believe you

is there any good FREE voice AI yet or site or is it still paywalled

>>106261466
Just grab a mic and fucking record??

>>106261466
>s there any good FREE anything
No.

>>106261452
which part don't you believe, that its uncensored?

>>106261534
i want to see:
the decent prose you claim
proof that it was made with gpt-5 -> gpt-5chat

>>106261329
>if think
Someone is shaking...

>>106261554
i genuinely dont remember making that typo oop there went the post lol

>>106261466
>AI text general
>hey guys can you tell me about audio AI
Fuck off tourist

Hey guys, I think!

>>106261628
proof?

>>106261638

>>106261571
ok i will, sorry you feel agitated today

>>106260950
under appreciated comment

>>106261648
Shudra ahh post

>>106261466
>sorry
retard

>>106261546
not gonna post my logs over something that trivial

do it manually to test
prompt gpt-5 for a 50-100 word summary of what happens next in your whatever

then copy paste it into your whatever with (OOC: )

like, think about
lmao
all you're doing is using gpt-5 to generate text for a (OOC: )

I use opus. Anything else fucking sucks.

>>106261788
>claim
>>post log
>no
Tale as old as time

>>106261788
oh btw another interesting result from this is that because you're asking for a summary of something and not a depiction, it cuts to the chase more directly, and then when you supply the instruction to gpt-5-chat it's then writing about something more logical instead of going in wack illogical directions littered with contradictions

im still fiddling with it but "what would most likely happen next" styled prompts to the reasoning model work well for this

>>106261466
no i dont think so, audio ai stuff is really lagging behind text gen and even video and image gens not sure why

>>106261908
>not sure why
Techlet

>>106261879
>anon wants his hand held
Tale as old as time

>>106261939
could you hold my hand...?

>>106257285
>from the perspective of the bot
This is a new one. So even in third person chats my description should still be "I am char, I am in love with user."

Current best Instruct?

>>106262127
Opus

>>106258513
What's cringe is unironically enjoying 200 token cards. There's nothing wrong with a 1000 token card that's well written. Oh this is a reddit post isn't it?

>>106262127
Qwen

>>106262146
>There's nothing wrong with a 1000 token card that's well written.
There's everything wrong with 1000 token cards. AI becomes dumber and repetitive the more tokens it processes in context.
>well written
A well written card isn't 1000 tokens. It's 200 tokens, plus lorebooks of 100-200 tokens each. If your model knows what a pokemon is, you don't need to spend +300 tokens explaining what a bulbasaur is. It already knows it. It just needs a single sentence describing what aspects of it you want more related, which would be like, 8 tokens.

A well written character card is under 500 tokens.

>>106262127
yo mama

>>106262229
Bro it's not 2023 anymore, we have context now.

>>106262229
>A well written card isn't 1000 tokens. It's 200 tokens
I really really isn't but I'll never convince you of your slop anyway, especially when char's physical description and outfit can easily be 500 tokens.

>>106262247
You are replying to spamtranny, those posts are reddit copypastas

>>106262270
It's okay bro, I have context.

>>106262317
oh yeah? name all tokens

>>106257180
pair of fragile anchors

>>106257306
>1,026B params
Sometimes I feel like you guys know somekind of meta setup that makes you able to run these things at home, but my gut tells me you're all paying for porn in a very elaborate way through openrouter services.

>>106262381
([A-Z]|[a-z]|[0-9]|[!-/]|[:-@]|[[-`]|{-~]|[^\x00-\x7F])+

>>106262520
API Inferenced chatbots general.

>>106262520
I haven't tried it yet but it's free on OR. You might have to put in a few credits to have access but idk like I said I haven't bothered https://openrouter.ai/moonshotai/kimi-k2:free

>>106262520
Go to /lmg/, retard

>>106262520
Nobody runs that shit locally. But most of us aren't retarded enough to pay for open weights models, either.
craft-sunglasses-deep-her trycloudflare /api/moonshot/v1

>>106262520
anime website

>>106262569
>deep-her
R2 cooming soon premonitions!

>>106262563
Isn't Kimi on OR censored as fuck?

>>106262660
I can believe it but like I said I haven't tried anything.

>>106262683
I did, haven't been able to do anything NSFW, no matter the JB.

>>106262708
Post JB. I'll see what's wrong with it.

>>106257179 (OP)
>>106257180
big tits? big bush

llama 4 uuooh

>>106262708
You may have hit this issue https://rentry.org/59kehtv4 which still needs to be fixed in ST.

>>106262713
I simply used Marinara, adding this prompt in the Guidelines:
https://files.catbox.moe/fgkns6.txt
>>106262801
Oh? I had no idea.

>>106262780
slop

any glm or kimi proxies?

>>106258570
>bimbo pet world
You've got my attention, fren. Can't wait.

>>106262863
>4 hours writing prompts for a function
Lmao. If AGI can write any single function then we already have it. Context is the only problem at this point

>>106259054
Whoa holy shit that's so cool. How did you make this? I didn't think someone would like my girls this much.

>>106263024
Your girls are pretty cute bwo. Don't let the haters get you down.

>>106263063
Then it was all worth it. I've got a little experience in renpy I was even thinking of making a VN eventually once I finish them all.

>>106262780
it says its been struck no? so no more UOHH

>Gemma 3 270M
>absolutely tiny - just a 241MB download
>screamingly fast
>hallucinates wildly about almost everything

>"Saint Arcadia may seem… overwhelming at first. But I assure you, we have special programs in place to help you… adjust." Several girls in the audience giggles. They already understand the system—how we break in new _meat_.
>"Très, très mémorable…" (Very, very memorable…)
you tricked me well, 4puss1es....

>>106262528
>could have just done /.*/isg

Damn sonnet, I guess I'm gonna pause my roleplay and rewrite my preset wtf.
>>106262229
You know what, anon? You're right and I'm sorry for insulting you. I'm gonna take a break now.

>>106263582
why shouldn't you throat fuck sonnet?

>>106262229
200 tokens is not even enough for the breasts...

>>106263723
just how big is this girl that we're talking about

Popping in to see how's gemini looking? Is it dead?

>>106263325
Nobody cares tranny

>>106263763
Gemini says you're a fag

>ask opus for help
>tells me to write a 300 token description
It's fucking over bros...
I can't even believe this

>>106263834
Deserved.

>1 minute apart
at least he's not a passnigger

>>106257216
Plaintext is best
I've found that I get the most out of a bot when I'm just writing their backstory and how they should behave normally rather than doing some weird formatslop attribute lists.

IQDB IS DOWN
IT'S OVER

>>106261424
I don't see the point why aren't you just prompting gpt chat
What's the difference

>>106263892
I mean it's about time it looked like it was getting ddos or something for days, weeks probably

>>106257216
Plaintext in paragraphs (don't even need headers) and 300 tokens max. I've... I've seen the light...

>>106257180
big breasted children are the best!

>>106259054
I could only barely tell this is generated. Artists are so incredibly fucked.

>>106264023
artists, writers, call centers, programmers, everyone is so incredibly fucked

my bot only got 3 likes

its fucking over isnt it

no more botmaking

What's the best workflow for erotic short stories rather than chatbot stuff

deepseek v4 will save us all

>>106264112
mikupad prefilled with your favourite ao3 tags

AIEEEEEEEEEEEE

>ao3
The reason why most models suck at creative writing is because they're trained on that garbage

>>106264112
you speak as if ST plus api for corpo model is not the bleeding edge technology for AI assiested smut
at best you can use noass

>>106258850
you and your LLM are both dumb

if the model name contains "claude-opus-4-1" it returns true... as intended. It should and WILL return true if any of the 3 formats match. it could maybe be more concise... but the code is perfectly readable, does what it needs to, even has a guard clause

by all means, keep spitting out hallucinated garbage. you add so much value


t. actual code monkey

>>106264284
layer 8 issue

>>106264284
ao3 is the reason why models can do smut that combines all of your fetishes

>>106264305
The sooner you get replaced the better.

has novelai given up on its writing model?

>>106264343
Supposedly they're planning to release something for text before they release their next image model, but they haven't released any more details.

>>106257848
ds acting for you is a a skill issue, "Only speak and act for characters other than {{user}}" in a/n at depth 0 solves it completely
other than that agreed on all counts

>>106257836
what do you usually rp
everyone i know that puts gemini high only rps sfw

No need for a JB for kimi?
Or at least some config for sillytavern

>>106257813
logs?

>>106258294
yeah gpt models are like stale bread for porn, id rather just use 3.7

>>106264323
no, it's actually the reasoning
t. been here since turbo, claude + CoT was the first model to do breast expansion well after thinking on the effects of certain sizes on objects around

>>106257813
>mainly because of Marinara’s preset.
Post it, I have been using gemini for coding so much it would feel dirty to use it for this buy can give it a try

>>106264389
I tried it with the same 80IQ settings I use with the Deepseek twins and managed to have some slightly less than consensual sex with no issue.

can someone make a tutoring card with two sisters that arent jeets please? thanks

>>106262776
smooth

gpt 5 is really fucking good at moving the plot along. it fucking zooms when it comes to introducing new things, but it's so schizophrenic in actually writing it that i need to get gemini to re-write everything gpt-5 generates

>>106264456
Were you able to get it writing longer? I think my prefill is working, but honestly, Kimi is all too happy to refuse or, again, churn out suspiciously short responses that don't move things forward.
And that's not mentioning the blank responses.

>>106264284
>t.never vectorized a fic
ao3 has some decent stuff occasionally

>>106264380
gemini is good in erp?
1.30 temp btw

>>106258423
shit I fell asleep... anyone care to share?

>>106264613
well its actually using explicit words here but the prose itself is still kinda shit
preset?

>>106264613
do not lewd the umas

>>106264613
horse pussy

>>106264650
sowy

>>106258850
You could write a 1 microsecond faster function with a paragraph comment explaining how the other names actually include the first one so you don't need to check. Or you could just get the job done because you're not an obsessed tranny.

>>106264569
Depends on what you define as longer, I guess. I generally prefer shorter responses so I'm not sure what sort of scale you're looking for. Here's the first response I got to a low-effort prompt to one of my low-effort bots. No fancy prefill, just a system prompt I stole from somewhere else ages ago (chub, probably) and horribly mangled. Also, I've only used the model once before so I couldn't be considered an expert.

how do you fuckers even use gemini with the privacy shit of google

>>106264873
I like the thought of jannies reading my logs.

>>106264873
proxies

>>106264935
>proxies
too expensive

>>106264873
what do you mean
i don't put my name or personal info in it, you have any actual reason why this isn't enough

>>106264873
I don't mess with cunny and google already knows my all my fetishes by heart so there's really nothing left for me to lose

>>106265021
>I don't mess with cunny

For gemini 2.5 are you supposed to use this or something else?

>>106265089
yes. when using gemini proxies, you use that

>>106265021
>I don't mess with cunny
BVSED.

wish constantchange shilled here since /vg/ troons hate talking about politics

>>106264023
If it is, I'd be incredibly impressed because he got all their details perfect in a 3 person image and I didn't include their prompts anywhere.

>>106264305
>>106264798
We already established that you are retarded, why reply two more times?

I miss having proxy

>>106261468
>>106264455
>>106244851
>>106253344
>>106261081
>>106263604
>/g/ is not a /pol/ branch
privacy and FOSS are a political statement bythedesu

>>106265197
You can tell because there Are imperfections, they're just hard to spot due to the pixelation. Take a look at the keyboard buttons, take a look around the black woman, the eyes on the hoodie also have strange artifacts around them, as there are around the left arm of the guitarist.

that last one is just blatantly (You) btw

I don't care really, fuck off to /vg/ if politics trigger you

>>106265205
Feels like they fucked with the R1 quota

Are there any Japanese Chat-GPT/Google AI style AI (preferably that speak English)? I want to do a deep dive into the history of early 00s light novels, but English chatbots only reference English sources (and there's very little documentation on this subject in English). I've done personal research on the subject before, but I figured an AI might know more about it than I can find on my own with my limited knowledge of Japanese internet.

>>106265319
>but English chatbots only reference English sources
Just tell it to use Japanese sources?

>>106265251
Well as a genfag myself I still want to know how he did it. Or at least have it in a bigger size. I love that that 3D look and I'd try that lora out myself if I had it.

>>106265319
ai isnt going to know shit about this

>>106265339
As a gemfag myself I'd like you to polish me until I shimmer

Are any good

>>106265352
I'll give you a nice spitshine, faggot.

>>106265362
Not really

Is this thing real?
https://ac-failures-skirts-improved trycloudflare.com

>>106265395
Locusts don't even deserve free cai

>>106265481
Why do you hate me so gratuitously?

>>106264873
Why not it's just smut

>>106257180
https://chub.ai/characters/VinceLavali/eletta-c17814c4cae6
https://files.catbox.moe/x14fse.png

Eletta of the Molten Creed, a little ruthless outlaw mercenary. Make sure you've got recursion on for the logbook. And with this, every song in Cardsongs finally has a card.

https://vincelavali.bandcamp.com/track/aboard-the-eclipse-leviathan-eletta

>>106265485
Locusts aren't human

To those without streaming, what do you do when the message is loading, go to other tabs lol?

>>106265537
holy necromancy

>>106265336
Wow, I didn't think it was so simple. Thanks.
>>106265349
It's actually doing a really good job of suggesting specific series that certain cliches likely originated from. It even started doing stuff like pointing out that タイム・リープ from 1995 was already satirizing how overused time travel was and gave specific lines from the book and what they were referencing.

>>106265552
check /aicg/, although it breaks immersion

>>106265552
Don't ask me how I know you're underage

I am begging on my knees
I need umamusume bots
I need sex with horsegirls

>>106265395
yes free opus is real wwwww

https://chub.ai/characters/disrespectful_dud/hannah-han-2f5ff1f809d2

Hannah is an elite courtesan who has the luxury of choosing the clients she spends time with - and tonight she's yours. For a price. Enjoy her company, take comfort in her empathy, and take her to bed - but take things too far, and this might be the last night you see.

>>106265635
gigaslop
looks like a discount Ani

>>106265635
>1300 permanent tokens
>interview style
Hmm...

>>106265635
>"So, {{user}}, tell me: would you like some tea?" *She purrs.*
>She purrs.
anon...

>>106265646
>>106265654
>>106265661
>didn't shill on /vg/ first
>instant mean replies
I'm nooticing...

>>106265552
Why do people use streaming? I hate waiting as my message slowly loads.

>>106265684
/vg/ is full of trannies, all their bots are shit

>>106265684
My reply wasn't mean, I'm still in the middle of my existential crisis about token bloat. I don't even do what he did with his card (30% basic info, 70% interview on how she speaks) and I still end up with 1300 tokens of what I believe is essential info. Idk what people do and I need some guidance.

>>106265646
Yeah, felt like something vanilla after my last few.

>>106265654
I wanted to test out the interview style for shaping tone. Seems to work. Plus, this is actually my smallest permanent context bot.

>>106265661
Meh. I feel it's a good descriptive term for someone in context putting on a mask for someone to put them at ease - or into a particular frame of mind.

>>106265635
any benefits of interview format over plaintext? genuine question

>>106265684
idk about the other two but i just have -ism ptsd and purring triggers severe claude 3 flashbacks

>>106265646
Ani is discoun Misa

>>106257179 (OP)
Anons, you know that betsy bot some anon made? If you edit out the faggot and weird leftie shit, it's pretty nice, but... The characters are very detailed, yet... You don't even use these fuckers, you know? There's just so many characters that you legit don't have the time to interact with all of them. And the amount of characters make it so that you can't really interact with them all the time - it simply gets tiring. The main family itself has like five people. It's pretty unwieldy as far as bots go - you run out of context space by message #100, after a small ark

>>106265635
Bitch

>>106265684
Well the first one is /vg/foid, one of our resident schizos. You start to notice the rote style of shitposts from that one after a while.

>>106265764
Did you get to see her knives?

Why does every model assume I'm wearing cologne? Always takes me right out of it.

>>106265805
I'm starting to get sick of mentions of bergamot...

Anons., how do I stop Gemini from cutting itself off? It's a real problem. Claude can just generate for days in a single message. Gemini 2.5 pro gets cut off. Anyone else have this problem, or is it my preset?

>>106258952
>this retard thinks wasting time making slop for chub is something to brag about
ahahahaha

>>106265785
No I'm just mad she's a prostitute bot

what's wrong with slop?

>>106265931
The problem with slop is that it is slop.

Merkava status?

>>106265931
low quality / ruins immersion

>>106265960
refilled with 80+ claude keys, currently cooming my brains out

>>106265931
It's just one guy who whines at every bot or log posted in the chatbot general.

>>106265990
>80+ claude keys
I almost believed

>>106265635
What kind of elite ho smokes plain cigarettes? Give that bitch a kiseru at least

>Now, I'm pretty sure that they're should be compatible with Python 3.10.10, since (...)
did gemini just hit me with a typo?

>>106266046
xe's indian

>>106265931
My bots haven't gotten called slop despite spending over a week writing and testing them.

>>106266046
which gemini are you using?
I accidentally switched to flash once, it was like i was reading pajeets stream of consciousness when they're posting on stackoverflow or quora, just full of spelling/grammar errors and all the isms you see them do online . I think the high context also mindbroke flash too.

skillgod...

WHERE DA NEW THREAD WTHHHHHHHHHHHH

>>106266150
raw 2.5 pro through api, the esl english is starting to rot them

SKILLGOD WTF DID YOU DO

SKILLGOD FIX YOUR SHIT, please uWu

>>106266231
>>106266231
>>106266231
>And if I ever need someone to come along
Who's gonna comfort me, and make the NEW THREAD

>>106265931
My bots have gotten called slop despite spending over a week writing and testing them.

desu