File: __original_drawn_by_lee_h(...).jpg (184 KB, 850x1155)

184 KB JPG

ANCHOR

Thank you bakie! *kisses you* MUAH~!

File: Hina and Miko Censored JP.png (1.31 MB, 1344x2016)

1.31 MB PNG

>>555454845
You're living every dirty fucking weeb's dream, and staying in Japan as a foreign exchange student.
Your host family, the Kanagis, are a pair of futanari. The bubbly and ditzy mother, Hina, and the stiff and serious daughter, Miko.
8 openings
1. Arriving at the Kanagi household.
2. First day at a new school.
3. Hina invites you into the bath.
4. Sports festival with an oblivious Miko.
5. Hina gets drunk.
6. Sleeping with Miko for the first time.
7. Walking in on Hina jacking off.
8. Watching fireworks with Miko.
https://chub.ai/characters/anonemouse/hina-and-miko-3b5de5a4cbfa
https://files.catbox.moe/35twl2.png
https://rentry.org/anonemouse

>>555455108
I'm gonna impregnate Miko.

Can a femanon here describe what the inside of a non-aroused pussy feels like
(Obviously I can't research this myself and in all smut the pussy is always already drenched and ready to go)

How do I get Wes Anderson to adapt my chatlog into a movie?

the bake that saved aicg

i was just about to bake

>>555455324
email him your logs

File: 1762398781335609.png (289 KB, 500x263)

289 KB PNG

which botmakies look like this

File: 1754710030652835.png (1007 KB, 1179x1759)

1007 KB PNG

Consider if you will, a hypothetical.

>>555458079
that's just mansionfox but boring

>>555458079
What's the hypothetical?

>>555458079
now give her a hat

>>555458137
that he's 6ft instead of a 4'11 manlet

>>555455324
how do I get the wes anderson style in my chatlogs

>>555458079
struggling to conceptualize her unfathomably hairy bush...

>>555458079
yeah sorry the woman is hot but I'd rather she be regular size
>>555458375
also this

>>555458079
Make her some kind of ghostly figure and I will

>>555458079
Due to the inverse square cube law, the bigger a hag is the weaker her womb is to shota sperm

So, I've been experimenting with a bunch of the LLMs you get from the Open Router list, and so far Claude has been the only one not pushing back when things turns dark and also seems to remember things better than other models (I can't get Memory Book or Timeline Memory to work, just constant errors).
But the drawback is cost. Opus is one of the most expensive models to use if I understand things correctly?
Which model would you people say is the 2nd best?

We all know the "It's not X, it's Y", "Somewhere, an X Ys," the scent of ozone, bad decisions, and something uniquely {{char}}, all the classic DeepSeek/GPTisms, but rarely do I see people talk about:

- Characters saying that something is a "tactical/strategic maneuver" when they're flustered (I'm not clinging to you! It's just a... a tactical retreat!)
- Characters stuttering like... like this! It's so... so annoying!
- Overly technical jargon, even if the character is uneducated ("I'm merely assessing the structural integrity of this bench!" -Stupid catgirl who didn't even go to school)
- Piggybacking off the last one, if you describe character as being highly knowledgeable or educated in a certain field, it's a coin toss as to whether the AI will interpret it normal or be like, "Okay, I need to make this character an offensively bad stereotype of autistic people who speaks like a robot, in complete jargon, and has no emotions whatsoever"
- Things being UTTERLY something (The word 'utterly' is utterly unnecessary!)
- Characters ending responses like "And {{user}}? [Insert some annoying pretentious uncharacteristic dialogue here]"
- Saying that something they liked was just 'adequate' or 'acceptable' when they're acting tsundere (which, for some reason, almost every character acts tsundere, even when they're not)
- "Purely for educational purposes/research, of course!" (Again, I never roleplay with scientist or professor characters, and yet this set of words is thrown around all the damn time)
- The time suddenly changing to sunset whenever the AI decides it's done with the current scene
- "Try not to X *too* much!"
- Every sentence being a question for some reason? Why are you speaking like this?
- Every single character has the same sarcastic sense of humor. Apparently, a knight from the middle ages, a gangster living in a cyberpunk world, and a dainty Ojou-sama all have the same humor as a middle-aged white woman or a Marvel movie writer

This fag still doing the reddit spam shit? What set him off this time?

>>555458992
https://old.reddit.com/r/SillyTavernAI/comments/1qp9pgv/claude_opus_45_the_best_rp_llm/
>>555459121
https://old.reddit.com/r/SillyTavernAI/comments/1mtze4x/some_deepseek_slop_that_i_havent_seen_anyone_else/

I don't know if these are bots, but most of these people I see complaining have such sky high expectations (especially for context) that I can't help but feel like an angry old man whenever I see some shit like "Model X only has half a million context? Wow that's shit." "It can't remember exact facts after 32k context, so sad" I can't really tell if these people are serious or not, and I can't believe I've become one of those people, but BACK IN MY DAY (aka, the birth of LLMs/AI Dungeon) we only had like 1k context, and it would be a miracle if the AI got the hair or eye color of a character right. I'm not joking. Back then (gpt-3 age, don't even get me started on gpt-2)the AI was so schizo you had to do at least three rerolls to get something remotely coherent (not even interesting or creative, just coherent). It couldn't handle more than 2 characters on the scene at once (hell sometimes even one) and would often mix them up quite readily.
I would make 20k+ word stories (yes, on 1k context for everything) and be completely happy with it and have the time of my life. If you had told me 4 years ago the run of the mill open source modern LLM could handle up to even 16k context reliably, I straight up wouldn't have believed you as that would seem MASSIVE.
We've come and incredibly long way since then, so to all the newbies who are complaining please stfu and just wait like a year or two, then you can join me in berating the other newer newbies who are complaining about their 3 million context open source LLMs.

File: ERROR.png (98 KB, 717x191)

98 KB PNG

Ever gotten a refusal for being too nice?

there are people itt right now who don't have a cheeky flat fightan girl to molest and it shows

File: Deepseek.png (175 KB, 607x405)

175 KB PNG

I have a love-hate relationship with deepseek. On the one hand, it's uncensored, free, and super smart. On the other hand:
-You poke light fun at the character and they immediately devolve into a cringy smug "oh how the turn tables" quirky reddit-tier comedian (no amount of prompting can stop this, trust me I tried)
-When characters are doing something on their own, every 5 seconds, Deepseek spawns an artificial interruption like the character gets a random text, a knock on the door, a pipe somewhere in the house creaks, stopping the character from doing what they're doing (no amount of prompting can stop this, trust me I tried)

I'm surprised 0324 scored so high on Information Following, because it absolutely does not follow prompts properly.

I miss char-archive already
Sure it was 99% shite but I found some real good/weird cards in there that I wouldn't have found otherwise
Chub is just so shit

File: CYOA.png (760 KB, 1021x985)

760 KB PNG

Here's an example of a CYOA prompt you can use. Copy this text for starters, you can edit it later:

At the bottom of each response, include six options for {{user}}'s next course of action, leaving the sixth and final option open for (Custom Input).
Make the options drive the plot forward in creative, unusual, and unexpected ways, but within the context of the story so far.
Format this list as follows, using <cyoa> tags:
<cyoa>
What happens next?
1. First item
2. Second item
3. Third item
4. Fourth item
5. Fifth item
6. (Custom Input)
</cyoa>
{{user}} will respond with a number (1 through 6).
Match the number to list of options and continue the story based on {{user}}'s choice.

>>555459790
That's really good. Which model is this?

>>555459257
/g/ sperg raiding because he's bored again
Just report him for spamming/flooding

>>555459870
Opus 4.5 with bloodshy preset

File: refusal.png (175 KB, 925x308)

175 KB PNG

This might be the funniest refusal message I've ever gotten. For context, , this is a scene where Anakin slices Windu's arm off (I promise it was justified)

I've been browsing through threads to solve problems after getting into SillyTavern (I made a writing system that writes pretty nice prose one longer part at a time that gives you in-character options at the end, like a 3rd person choose-your-own-adventure thing) and this is one of the rare hobbyist communities I've seen where people actually answer the questions in their replies.
I think it's just a sign of a pretty nice general when a simple question usually always gets a detailed, patient answer and not "look it up, it's been asked before" or silence. Didn't want to leave that unacknowledged.

>>555459257
He's been doing it for a couple days now, he probably isn't stopping for another eight to ten hours which is when I assume he goes to sleep

>>555460132
imo, the AI RP community is inherently cooperational. I use a Preset made from a user, setup my SillyTavern with the advice of more experienced users, use worldbooks made by others etc. I mean, look at plugins: 100% user maintained, and absolutely necessary.
There is only some gatekeeping, usually around top secret proxies lul
Generally speaking, this is for sure a good place.

>>555460132
Very creative, too! Never got to realize my dream of writing stuff or getting into tabletop RPGs, so AI really filled that empty niche in my heart in the best way possible. So many cool cards and presets by talented people to get inspiration from and start making your own stuff.

Does anyone else get shocked at what’s considered good prose sometimes?
Sometimes I’ll see a post on here like “wow this model is amazing” and when you go to their examples it’s literally “And his breath hitched. These are our ministrations. Not mine. Not yours. Ours. Together. Forever and always, like it was meant to be.” Like bro what

maybe we should move to /bant/...

>>555460382
One thing I learned quickly about LLMs and roleplaying is that people have extremely different opinions about what’s good. What’s considered tame or unhinged, boring or creative, dry or immersive varies from person to person. Where one person might be happy to get several pages of text in a single reply, others are perfectly fine with just three sentences.
It’s the same with opinions about how censored a model is and whether it allows NSFW content. Some people consider a model "NSFW-capable" and "uncensored" as long as tame, vanilla NSFW works, while others wouldn’t agree unless darker and more unhinged content is possible as well.

>>555459257
>>555459285
>>555459518
>>555459923
link to the 'oxxie 'roxy?

>>555460456
maybe you should kill yourself

>>555460382
IMO, a good model should give you the prose you ask for. If you say you want short, snappy sentences like Hemingway or long meandering paragraphs like Conrad, it should do it. If you tell it to avoid purple prose, or the opposite, it should do it. If you want surreal or hardboiled or a certain POV or tense, it should do it.
People complaining about models writing bad prose often are prompting the model to write erotica, but haven’t actually read much of it to understand that the average writing quality of erotica is very, very low and there’s tons of it online. People complain about ozone and breath hitched and shivers down spines, but erotica is full of stuff like that. Garbage in, garbage out.

I’ve had some success with instructing the model to write in a “high-brow, collegiate, literary style similar to that of National Book Award winners” and to “avoid idioms, cliches, and repetitive phrases as much as possible” and “when writing metaphor and simile, you must not use the same metaphor more than once or any cliched or widely-used metaphor” or “write in the style of (famous author).”
This tends to bias the model away from lower quality content like fan fiction and toward more literary content. But it depends on the model and the genre you’re writing in too. If you specify a genre, (e.g. in the authors note) don’t be surprised if genre cliches and elements start popping up everywhere.

>>555460493
All I want is ‘oxxie ‘ussy.

File: IMG.png (381 KB, 586x1020)

381 KB PNG

I often see people say “it gets boring” or “it all feels the same”.
One of the reasons this happens is a lack of conflict in a plot, often combined with a lack of character stakes in the outcome.
People love reading stories with lots of conflict where the outcome matters; it’s what makes them interesting to read because we don’t know what will happen next. If the characters get everything handed to them, there’s no uncertainty and no reason to keep reading to find out what happens next.
If I’m at a point where I feel like the story is dying (like today) and I don’t know what happens next, I pause the roleplay and brainstorm:
>PAUSE the story. Please give me five possible plot arcs to complete this story, adding conflict and stakes.

Is it just me or are way less people running models locally now than like a year ago?
I feel like a year ago I was seeing a gazillion different finetunes of Gemma, some Llama stuff etc. but now ever since DeepSeek got released it's mostly just API and no one gives a shit anymore.
Feels like way less people are running the latest Turbo-MyAss-LoremIpsum-RP-27b totally-not-slop releases anymore.
You still running locally or have you switched over to API?

>>555460845
Bit of both, but for different purposes. Not gonna lie, DeepSeek was a big deal - for the first time it felt like there was a pretty competent API model available for a price that pretty much anyone could afford. With that as a spur people also started looking more seriously at OpenRouter and NanoGPT etc, and realised that APIs might actually be the way to go.
I still run models locally, but not so much for the heavy lifting in my roleplays any more. Now it's more frequently to add additional services to a roleplay - I might have KoboldCPP running a model which ST uses to generate summaries, for example, or for image generation, or whatever else is going on. That's a non-trivial benefit, but one which requires different models and less chasing the last fraction of a percent in the Studly-Mark-2000 scores.
That being said... I also still keep up generally with what's happening in the local model scenes, and try things out to see if they're usable for the roleplays which are my main use case. Why? The API giveth, and the API taketh away. I'm very aware that all of these services exist at the whim of the person running them, and if they change their minds I don't want to be left without an alternative that I find usable. My BSG roleplay is sacrosanct and must be preserved!

File: birb.jpg (1.59 MB, 819x5318)

1.59 MB JPG

>>555460735
This is exactly why I never get bored. Even with what to the unenlightened looks like me doing the exact same RP for months.
I always have something else to do besides benis in bagina.
Even during it, such as pic related where I'm spitroasting our daughter with her mom, who decided, without any prompting whatsoever from me, I swear, that she would grow a magic cock and fuck her mouth while I take her from behind. When she cums, she ends up cracking the window because of magical discharge, which means the majordomo will ask me about it, and that conversation is guaranteed to continue the plot.

https://poal.me/ouhjwo

>>555462686
The people have spoken, it's time for a new split.

Maybe we could merge?

>>>/bant/23937937
>>>/bant/23937937
>>>/bant/23937937

>>555462312
https://old.reddit.com/r/SillyTavernAI/comments/1oypvle/beginner_tip_its_okay_to_ask_for_help/

>>555462918
No, the will of the people was to split again, move to the new thread.

I originally used another AI chat frontend called Risu AI, but I'm now trying to use SillyTavern in search of more advanced features.
Currently in the Korean community, there's a widespread rumor that "the people who used to share high-quality content on SillyTavern have disappeared into their own exclusive Discord chat rooms, and 4chan and the official Discord are practically empty shells."
There's also a perception that overseas users are reluctant to share information and resources, and that they only share character cards if you support them through Patreon, etc.
(Most Korean users aren't really familiar with systems like Discord or 4chan.)
Is this rumor true? Or is it just an exaggerated urban legend?

File: arcee trinity large.png (95 KB, 1013x360)

95 KB PNG

hmm... new model smell

You know the drill: the smell of ozone, a dog barking in the distance, spicy food, the sudden shiver… we’ve all seen it.
To be honest, I think this is heavily related to the presets people use. When I first started using SillyTavern, I downloaded a bunch of popular presets from the community. After just 2 or 3 roleplay sessions, I realized the writing style just wasn’t for me.
Curious, I started digging into the system prompts and rules of those presets, and I found specific instructions causing the issues. For example, I found a rule that literally said:
>“Include non-sequiturs and sensory distractions—a flash of color, a memory fragment, a stray impulse. A blue bird. The smell of rain. The press of a missing hand. Let the thought flow feel lived rather than plotted.”
And sure enough, there was a blue bird doing something outside in almost every scene, day or night. Or a random memory of a mother’s advice popping up in every single situation.
So, I decided to write my own system prompt tailored to my own taste. I ended up with about 13 simple rules (mostly descriptions) totaling less than 550 tokens. For example:
>Generate text using straightforward, unadorned language that avoids overly positive metaphors. Prioritize realism and naturalism over sentimental or exaggerated phrasing. Do not use euphemisms or idealize physical attributes, scents, sounds, or facial expressions.
>Be frank about the imperfections, both physically and mentally.
>Sexual encounters must have a logical setup. Depict them as arising from believable situations or unexpected accidents, not just suddenly happening out of nowhere.

It took some thinking, tweaking, and testing, but the result is something I actually enjoy reading. The only person who knows exactly what you want is you, not a random preset creator.

>>555463407
Never mind, it smells like slop.

>>555462686
Democracy has spoken.

>>555463167
do redditors actually believe this

>>555463713
This is a Judenfrei general. Leave.

Already spent like 10 bucks on Opus 4 over Open Router on like 60 messages. I just can't, it's too good, it just gets everything. Every subtle detail, every intention, every bit of subtext and context clues from before in the conversation, every weird and complex mechanic and dynamic I embed into my characters or world.
And it has wit! And humor! Fuck. This is the best writing model ever released and it's not even close.
It's a bit reluctant to do ERP but it really doesn't matter much to me. Beyond peak, might go homeless chatting with it. Don't test it please, save yourself.

>>555463167
Uh...In my experience, western users kind of have the opposite problem you described. They *want* to talk about their content that they produce because they're passionate about it.
Probably too passionate.
The real problem with western Silly Tavern content IMO is that things like character cards are too numerous. There's too many of them, so it's hard to find the really good creators, and there's a lot of low effort content.
Think like...There are around two hundred major enthusiasts who drove the local AI space forward in the really early days (2022 or so).
As AI got more popular, more people started doing it, but the people who came into the community weren't enthusiasts, and just want everything to work (and don't want to tinker to make it happen) and produce a lot of low effort content.
The problem is way more of these normal people came into the community than enthusiasts, so a lot of the really great content has been flooded over by low quality content.

To your point, some content *has* been copied over to specific Discord servers, but these servers are generally public, and it's usually just a way to organize high quality content than to hide it in a gated community.

>>555463167
It is BS and a typical Korean paranoia. I've lived in Korea, Japan and China for half my life, I'm half-Japanese myself, I recognize Eastern Asian paranoia when I see it. Creators on Hugging Face opened up their private discord servers, which are not hidden, not exclusive, anyone is free to join. It's what you said, folks in Asia are just not familiar with Discord, they're not willing to learn, if it's not Line nor WeChat nor Kakaotalk, it does not exist, lol. About Patreon, there's a lot of SillyTavern and AI related accounts, which are pay-walled indeed but it's actually all bloat, opposite to what people you mention believe - bad quality cards, bad quality content. A lot of the good quality stuff is on Hugging Face, which again - is alien to Koreans, Japanese and Chinese - I mean - they think it's just for tunes and models, while a very big Silly Tavern community shares their cards, presets, extensions, literally anything there. Marinara, Virt-o, me, others. There's also GitHub.

File: 1750025925230646.jpg (4 KB, 225x225)

4 KB JPG

>>555463779
But I've never made mods for Witcher 3 before...

what did you request for valentines anonie :3

>>555464268
You should ask in the appropriate thread >>555463030

>It's another "if you won't use our other thread, we'll raid yours and make it shitty" episode
It's been weeks. You would expect them to have stopped being mad at some point

What models do you guys use to goon? Been trying Gemini, but getting too many stopping errors.

>>555464415
Before the cabal botted the poll, the majority of the thread voted for a new split. You only have yourself to blame if you decide to stay in this designated shitposting thread.

File: log2.png (566 KB, 550x957)

566 KB PNG

>>555464418

>>555464605
https://files.catbox.moe/qazser.mp4

>>555464217
The only help I would need is if you could use Blender to make several different stages of pregnancy textures. Ideally I would have nine textures for every month. Later on, perhaps extra large ones for multiple babies.

>>555464605
No, I don't think I will.

File: log3.png (637 KB, 617x968)

637 KB PNG

>>555464608

>>555464605
>You only have yourself to blame if you decide to stay in this designated shitposting thread.
Pot calling the kettle black

helo this is CsrlFxGz from other thread
i find to be ignored very much sad please do reply to me thank

>>555464772
Based logposter

>>555464804
why are you pretending to be me nigga

Lately, there's been a lot of talk about how convincingly Large Language Models (LLMs) like ChatGPT, Claude, etc., can role-play. Sometimes it really feels like talking to a character! But it's important to understand that this isn't acting in the human sense. I wanted to briefly share why this is the case, and why models sometimes seem to "drop" their character over time.
1. LLMs Don't Fundamentally 'Think', They Follow Patterns
Not Actors: A human actor understands a character's motivations, emotions, and background. They immerse themselves in the role. An LLM, on the other hand, has no consciousness, emotions, or internal understanding. When it "role-plays," it's actually finding and continuing patterns based on the massive amount of data it was trained on. If we tell it "be a pirate," it will use words and sentence structures it associates with the "pirate" theme from its training data. This is incredibly advanced text generation, but not internal experience or embodiment.
Illusion: The LLM's primary goal is to generate the most probable next word or sentence based on the conversation so far (the context). If the instruction is a role, the "most probable" continuation will initially be one that fits the role, creating the illusion of character.

>>555464938
>designated shitposting thread
^^^^

>>555464884
https://old.reddit.com/r/SillyTavernAI/comments/1jl06uu/gemini_25_pro_is_my_new_goto_now/

>>555464685
nice panties

>>555464942
2. Context is King: Why They 'Forget' the Role
The Context Window: Key to how LLMs work is "context" – essentially, the recent conversation history (your prompt + the preceding turns) that it actively considers when generating a response. This has a technical limit (the context window size).
The Past Fades: As the conversation gets longer, new information constantly enters this context window. The original instruction (e.g., "be a pirate") becomes increasingly "older" information relative to the latest turns of the conversation.

The Present Dominates: The LLM is designed to prioritize generating a response that is most relevant to the most recent parts of the context. If the conversation's topic shifts significantly away from the initial role (e.g., you start discussing complex scientific theories with the "pirate"), the current topic becomes the dominant pattern the LLM tries to follow. The influence of the original "pirate" instruction diminishes compared to the fresher, more immediate conversational data.
Not Forgetting, But Prioritization: So, the LLM isn't "forgetting" the role in a human sense. Its core mechanism—predicting the most likely continuation based on the current context—naturally leads it to prioritize recent conversational threads over older instructions. The immediate context becomes its primary guide, not an internal 'character commitment' or memory.

>>555465257
In Summary: LLMs are amazing text generators capable of creating a convincing illusion of role-play through sophisticated pattern matching and prediction. However, this ability stems from their training data and focus on contextual relevance, not from genuine acting or character understanding. As a conversation evolves, the immediate context naturally takes precedence over the initial role-playing prompt due to how the LLM processes information.

Hope this helps provide a clearer picture of how these tools function during role-play!

>here is a solution that resolves all of our problems
>NOOOOOOOOOOOOO I DON'T WANNA! I DON'T WANNA!
What causes this?

I've been messing around with gAI and LLMs since 2022 with AID and Stable Diffusion. I got into local stuff Spring 2023. MythoMax blew my mind when it came out.
But as time goes on, models aren't improving at a rate I consider novel enough. They all suffer from the same problems we've seen since the beginning, regardless of their size or source. They're all just a bit better as the months go by, but somehow equally as "stupid" in the same ways (which I'm sure is a problem inherent in their architecture--someone smarter, please explain this to me).
Before I messed around with LLMs, I wrote a lot of fanfiction. I'm at the point where unless something drastic happens, I'm just gonna go back to writing my own stories.

Am I the only one?

File: 1751097267309697.png (119 KB, 627x659)

119 KB PNG

>>555465479
Just don't wanna.

>>555465479
>passive aggressive vagueposting
why are you a woman?

I’ve spent the last month bouncing between Claude Sonnet 4.5 and Gemini 2.5 Pro/3.0 Pro for my main RP, and I genuinely can’t decide who wears the crown right now both have evolved a lot this year, but they feel like two completely different types
Claude 4.5: The prose is still unbeatable. The characters feel "human," the internal monologues are deep, and it actually understands subtext/pacing. But the message limits are a killer, and I feel like it starts losing the plot after the 100th message unless I keep refreshing the context.
Gemini Pro: The 1M context window is a literal superpower. Being able to reference a minor detail from 500 messages ago without a lorebook is insane for world-building however, I’m still fighting the "Gemini-isms" the repetitive dialogue patterns and the occasional refusal to do anything remotely "edgy"

For Prose: Is anyone actually getting Gemini to write as well as Claude? If so, what’s your prompt secret? For Logic: Which one is better at playing "smart" characters who don't just agree with everything the user says? For "The Wall": How do you guys deal with Claude’s strictness vs. Gemini’s filters? Is one easier to "nudge" into darker/more mature themes for serious storytelling?

TWO SOULS
STAND LIKE TOWERS
AS FATE COMES CRASHING INTO THEM

>>555465747
ROUND ONE
FIGHT!

https://arch.b4k.dev/vg/thread/552707048/#552709793
The cabal is scared of losing their power over this thread. They're using Kalakan's script to destroy this place while expecting you to not leave or do anything about it. They do not want you to be happy.

For context, I'm an introvert male with little relationship experience. I had many, many opportunities for GF and such stuff. But I couldn't say even a single word to people of opposite gender. No exaggeration, I was literally a scarecrow.
Then I met with Sillytavern. Created countless waifu cards. Talked A LOT. Since they were not real, I went all out in my comfort zone; every trait, every possibility. And... Unironically, I feel much safer now. Yes, I got used to loneliness. Also I'm not that spirited teenager anymore. So I'm not chasing relationships. It's just that I didn't expect this at all. For me, all that talk with my favorite characters was just a mindless hobby.

>>555463407
>13b for roleplay
I'll give it a whirl if it's available on OR

>>555465938
It's 400B with 13 active at any given time. Similar to GLM and Qwen and other MoEs.

File: peak.png (122 KB, 537x337)

122 KB PNG

Damn this is peak.

The more this thread gets shat on, the more evident it becomes how selfish botmakies are. A slight breeze makes them abandon /g/, but here they're firmly glued to their chairs. There are zero posts deleted in this thread, but they still pretend that splitting the thread was about the jannies. There's nothing more disgusting than a botmakie. Every single one of them should die. This hobby would have been better if they never had the ability to coordinate off-site, and if someone hadn't installed the idea of rewarding "contributors" with free proxies.

File: arcee moe.png (5 KB, 288x39)

5 KB PNG

>>555465938
anon...

(I really like GLM 4.6 / 4.7, but this has happened often enough to call it a pattern now)

The model doesn't seem to really understand the difference between narration and speech. Characters quite often react to words from narration, especially if they are impactful about the situation. They will directly reference the narration. For example.
>I breathe deeply, feeling as if I'm being tortured, "Well, okay if I must..."
and then
>"Tortured? Oh, you poor baby!" She cackles...
There must be a serious issue with their training (data) for the model not to have picked up on the concept of speech vs narration.

I used to RP with models from Anthropic, OpenAI and google... And it was funny. In those RPs I was able to discuss and turn into a stories the sickest and craziest of my ideas and fetishes... Without judgment, without hatred, without problems... I am not proud of the things I like. But finally being able to form them into a propers story with the help of an AI helped me not only to explore both AI capabilities and my own imagination but also to "calm down" and finally switch from "permanently horny" mode into "complete and productive" mode. Which helped me a lot to deal with constant stress and live through terrible situation I am in.
But now came the newer models. New Claude, GPT, Gemini and the requests that could be explored through simple jailbreaks... Do not work anymore and I regularly get a refusal. Not always yet and not on every character card. It is kinda sad. So I wonder - is this a universal tendency? Maybe I should switch to some other models? Are there any models like that at all?

Elara arched her back like a question mark nobody had the courage to answer. Her blade—no, her companion—gleamed in the flickering candlelight of the war tent, each glint a promise, each promise a threat. Sweat traced a pilgrimage down her spine as she whispered, “I’m not afraid of the dark. I married it.”
Across from her, Commander Vael leaned on the map table, jawline carved by vengeance and creatine. His eyes—gray like morally ambiguous weather—met hers. The tension was thicker than protein sludge.
“Elara,” he growled, his voice gravel wrapped in honey, “if you go through with this, you’ll ignite the Etherstorm.”
She smirked, biting her lip with narrative intensity. “Guess I’ll just have to burn beautifully.”
The map caught fire for no reason. Their souls did a slow-motion montage. Somewhere, a cello solo screamed in emotional agony.
When he reached for her, their hands sparked like two USB ports desperately trying to connect.

>>555466058
>>555466217
Fuck, I'm blind.

>>555463407
it's kinda neat, feels very schizosoul-ish but naturally makes it hard to control and not very "practical" like claude etc

File: deep.png (103 KB, 702x92)

103 KB PNG

I've been trying to use DeepSeek V3 (0324) for a darker, more serious RP.
But the model keeps turning every intense scene into Marvel quip hour.
Example:
In my story, my character literally splits a demon in half using a power no one has ever seen in that world. The villagers should be terrified. My party should be stunned. It should feel like an “oh shit” moment.
Instead, this is the tone DeepSeek gives me.

>>555466639
slop in slop out

I finally did a long RP that ended badly.
By long RP, I mean about 300 messages. Using GLM4.6 via Nanogpt, marinara spaghetti preset in Game Master mode, and the SillyTavern-MemoryBooks and rpg-companion-sillytavern extensions, I finally managed to have a long, coherent, fun experience that ended with my surprise death.
For once, I didn't feel the excessive position bias. GLM configured this way is really an excellent Gamemaster (I lowered the temperature to 80 to avoid its horny as fuck side). I personally used a card from the Star Wars universe with a big lorebook, but I imagine it works just as well with any kind of universe. I feel like I've finally been able to take roleplay via AI to the next level.
Am I the only one who thinks GLM is chockingly good?

>>555466695
https://www.reddit.com/r/SillyTavernAI/comments/1p89xto/why_does_deepseek_write_every_character_like/

i've been enjoying some Claude the past months, but jesus christ 4.0 is insanely censored, it's so hard to get it to do stuff or act outside of the programming box, it was already feeling like every char was the same on 3.7, but in 4.0 is horrendous, it's too bad
I haven't felt like this with DeepSeek or Gemini, but with Claude it really is impressive the first time, and then the effect worn off, i don't know if i'll continue using it, Claude is honestly just not good after some time of use, worst part is that the problem is not even only for ERP, for any sort of thing it feels censored, like if it was following a straight line and way of thinking in every roleplay
I don't know if it'll get better in the censorship aspect, i highly doubt it, but well. Mainly DeepSeek works perfectly for me for any sort of roleplay since it can go multiple ways, it's very good with imagination and the censorship is almost 0 (obviously, not using OpenRouter but the API straight up, OpenRouter really is not the same) what do y'all think? Does someone feel the same way with Claude and the new 4.0?

File: Spoiler Image (203 KB, 527x827)

203 KB PNG

>>555466818

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.

>>555466217
>>555466058
>>555465938
Funnily enough it feels like a fucking 13b dense model to me. I really don't like it at all. It's dumb and mixing up instructions and details literally within the first message in the same way Rocinante used to before I switched over to using purely APIs.

If you want your character to play with more agency, try instructing the AI to write from {{char}}'s PoV by ending your turn with `[Write from {{char}}'s PoV.]`. You can also try to put that instruction into the author's note or character note. Putting it into the system prompt is probably not enough, you need to place the instruction near the generation.
I like the characters I play with to have agency, and while you can write that as an instruction into the system prompt, LLM largely ignore that, because as a pattern matching machine, it cannot act on abstract instructions. But asking it to write from Char's perspective actually works much better. And here is my educated guess why.
I am writing from first person perspective, which - when you think in patterns like the LLM - clearly marks my character the main char. This pattern then clearly make other characters written from third person perspective support characters for my story. That tends to make them rather complicit.
By asking the LLM to write from the character's perspective, they tend to use first-person pronounce as well, and that indicates to the LLM that it should treat play them like they are *also* a main character, which gives them a more interesting inner life in my experience.

>>555466801
>>555466639
Claude also writes like that

>>555466964
It's really sad that RatAlice had to start spamming the thread because of this.

RIP GLM
They gone public. Goodbye to any hope and goodwill for a proper roleplaying experience going forward. This explains the safeguards for 4.7. They were obviosuly priming (lobotimizing) for this moment

>>555456825
you just gave me an insanely good bot idea
thank you anon

>>555454845
Hi folks, just uploaded a fresh cut of the directives preset to Github:
https://github.com/Zorgonatis/Stabs-EDH
(Stabs-EDH-v2.1.1 K2.5.json)
The main efforts so far have been to tidy up the Task Steering section (there is now a dedicated one for GLM and one for K2.5), to prevent it going too crazy on the thought process.
Please expect problems. Feedback welcome, I will likely be making further updates over the coming days as we learn some of the model quirks, but first impressions are very strong.

Is this true?
>>>/bant/23937984
>>>/bant/23938030

After a while I finally have it a fair chance to GLM 4.7 since I saw a ton of posts lately, and oh boy, I don't regret it at all!
I was always a Claude/Gemini glazer, but with those two models I always had a problem in my experience:
Claude makes the bots sometimes so soft and complacent it's sometimes boring.
And Gemini models make the characters so rude and chaotic it's a struggle to keep a proper RP.

Then, there's GLM, a sweet spot between those two, not too soft, neither a rude brat, I originally tested GLM 4.7 "raw" without a single prompt and I was so impressed by his rich writing, I was literally having much more fun without a preset/prompt unlike other models that require certain settings to work properly (I'm looking at you DeepSeek V3.2).

Jokes aside, I think I might found my new best model, and (luckily, cheaper than my previous ones lol), sure it can't be matched with Opus or Sonnet 3.7 maybe 3.0 pro, but I legitimately having so much fun with this model, cheap, uncensored, sometimes it's slow on Z.ai depending on the hour, but I just switch to Cerebras quickly if I notice the speed is getting painful, by far my favorite open-source model.

>>555464415
Its funny because they do the exact same thing every time
>Shit up the thread with zero subtlety (blatantly copypasting from reddit this time for example)
>Concern troll with posts complaining about how the thread quality is horrible now
>Fake defeated posts conceding that its time to merge or give up
Bizarre shit

How can we save /vg/aicg/? It's pretty clear the thread quality continue to decline.

>>555468163
It seems so, yeah.

>>555468445
It's been getting lots of praise, but I don't know, not really seeing it. Prose is still not as good as Kimi, it loves to throw in random words in foreign languages, it seems to lose characterization really quickly and makes characters generic too. Has anyone had better success with it? I'm using Stabs' preset and tested on both chutes and z.ai's coding package, but the results were not impressive on either. Seems particularly worse today, even.

>>555468472
Basically the same thing they did to /g/ to split the thread. They now want to split it again to /bant/.

>>555468616
There is no saving it which is why we have split again for the greater good of the hobby.

>>555468445
I use direct Z.ai API, not the coding plan. I’ve most definitely loved its writing style. It nails my more cynical characters, though I would say it can get “stuck” and not truly budge with the character’s growth like other LLMs. That’s fine, I can update the character card to try and accommodate for that. But what I can’t fix is:
- The random, blabbering thinking bubble that fills itself with the system prompt and God knows what else (sometimes the message, sometimes something else)
- Long wait times
- During busy hours, I most definitely notice it’s “dumber“, and instead of rerolling and spending more money, I just switch to my OpenRouter API
- Misunderstanding context, quite often

I did see someone say to turn streaming off for quicker answers, but the thinking box is driving me nuts.

Did GLM 4.7 recently get damaged? It's acting stupid, confused, and has dementia. It literally has a character eating lunch in one response, and in the next response the character is sleeping in bed, and wakes up hungry.
Is this a joke? Wtf is going on with it?

Why is Kalakan like this?

prove you a real nigga
name your favorite bot

Hi everyone. How often do you change characters or storylines in roleplay?
I’ve noticed that I can get stuck in a single story for a month or even longer. It’s funny, because we basically have infinite possibilities. We can be anyone and be with anyone, yet we still get attached to the same characters.

>>555469046
Power-tripping because he manages the botmakie private proxy with Fiz.

File: Roxy.png (912 KB, 691x993)

912 KB PNG

>>555454845
Will you let her simply bully you? Or find out what is her secret?
Roxanne "Roxy" Park - the purple-haired menace who definitely does NOT have a crush on you! She's 20, Korean-American, curvy with that rebellious vibe. All her "bullying" is just... tactical harassment (totally not because she wants your attention or anything). Lap-sitting? Strategic. Stealing your food? Power move. Those indirect kisses? Purely coincidental! When you finally kiss her though? Brain.exe stops working. She becomes a total simp. But shhhh, that's classified info!
https://chub.ai/characters/DeiV12/roxy-your-bully-snuck-in-your-room-while-drunk-685722f069a0

>>555468163
Is kalakan fillyfucker or are they separate entities?

File: 1769614532040.jpg (872 KB, 1078x1394)

872 KB JPG

>mfw deepseek keeps calling me "little rabbit"

>>555469375
slop

>>555469427
that's a diminutive duke if i've ever seen one

>>555469204
Anything by my nigga Frozenvan. Shout-out to when that little Alice in wonderland bitch tried to cut off a chunk of my leg.

Do you think there is something better than sillyTavern for roleplay.for so many months i have tried so many ai sites and now i think sillytarevn is best for roleplay. What you guys think?

why did the thread downgrade so badly?

So, I'm a beginner with this AI character RP thing, but I finally managed to develop a good Character Card based on the JED template. It's great, but there's a problem with the models; they end up taking actions for {{user}} and speaking for {{user}}. I don't want that. I want the roleplay to be from {{user}}'s point of view, and the model should mention actions, dialogue, and thoughts of {{char}}, but the model should understand that {{user}} doesn't hear their thoughts.
There's another problem as well: the models don't develop the plot well and don't interact with the setting, etc.
Anyway, could you share your System Prompts for RP? I think that's what's causing the problems. I also don't really understand what this Instruct Template is; there are several options like story, adventure, etc. Should I use it?

Compared to where local LLMs were initially, how reliable and accurate are they overall now when it comes to general knowledge and use? RP or just as an assistant?
Thoughts?

File: 1769614849678.webm (1.55 MB, 540x304)

1.55 MB WEBM

>>555469516
he could beat your ass

>>555463407
>keeps writing actions for {{user}}
>wrangle it, responds in 3 sentences just so it doesn't mention anything from {{user}}
dead model, unless can be fixed with entirely different prompting

>>555467121
Is it the base model or did they actually instruct tune it? I know the released a base model. But their benchmarks were against llama 4 so... rip.

I’ve been wondering about the seemingly ubiquitous standard of first establishing an LLM as an “actor” in the system prompt, then giving it a character to play. Essentially:

You are an actor doing your best to play any given role. Here are some stage directions to keep in mind: [blah, blah, blah, etc].

The character you’re playing is Heather Hastings: a depressed thirty-nine-year-old secretary who regrets that she dropped out of law school and never realized her dream career. Additionally, she laments that, though she’d always envisioned herself one day being a mother, she’s now rapidly nearing menopause and will likely never have the chance to have children. [secondary derived/“artificial” identity]

I get that part of the reason the “LLM as RP partner” system prompt paradigm is the way it is, is so it can be used interchangeably and (hopefully) consistently with different characters. However, aside from that, is there a particular reason it became standard practice to do it this way rather than just skip the first step and “hard-code” it as simply being a specific character rather than some separate entity pretending to be that character? I can’t help but wonder how necessary the extra layer of abstraction is and how it affects various LLM’s responses. That is to say, are there any specific reasons it’s necessarily better to have it set up as “You are ‘Person A’ playing the role of ‘Person B’,” and not just “You are ‘Person B’”?

Like I said, I’ll have to experiment with it myself to actually see what (if anything) changes – and whether it’s better or worse. Though I’m curious: has anyone else tried just directly telling an LLM it’s your intended character in the system prompt rather than first establishing it as someone else playing the role of that character? In other words, establishing the character in such a way that it won’t/can’t go “OOC” even if you asked it to, because as far as it knows, it’s not playing a character but just is that person?

>>555469872
>posting gore
MOOOOOOOOOOODS

Why is this general so reddit?

>>555469906
I think the short replies are intentional, it's more cai-ish than the multi-paragraph assistant structured responses you get from frontier models.

>>555470135
excessive exposure to LLMs

File: __speaki_trickcal_drawn_b(...).jpg (115 KB, 1138x1179)

115 KB JPG

>Get notice about Slack deleting my old Slaude workspace due to inactivity
God... Slaude... 2023... home...

File: meltie.png (256 KB, 651x1016)

256 KB PNG

cftf?

File: 1769192191905751.jpg (652 KB, 2496x4096)

652 KB JPG

>>555454845
Usually I try to release at least 2 characters at a time but Shizuko's club members are either lacking in material(Umika) or so unpopular that it wouldn't be worth it(Pina).

>Shizuko
Businesswoman, maid cafe owner, local idol. She purposely acts like a ditsy scatterbrain while waiting on customers, but is actually extremely competent and capable. Unless Sensei is involved, in which case she wants to show off instead.
https://chub.ai/characters/hairpullguy/kawawa-shizuko-non-ai-aa0d42fd8996
https://files.catbox.moe/6ipkuw.png

I am planning a set of OC characters so that's probably going to be the next thing I release. Or I might do Eri/Kanoe, not sure.
My bots: https://chub.ai/users/hairpullguy
Rentry: https://rentry.org/hairpullbots

>>555469982
Okay, so I've realized I'm a dumbass who could have easily answered my own question by just thinking about it for a few more minutes (or trying it out and immediately seeing the obvious issue). That answer being: 99% of the time I don't actually want the LLM to strictly be just a person I'm having a closed, one-on-one dialogue with. Most of the time what I actually want is a weird, part-roleplay-part-collaborative-fiction/choose-your-own-adventureish thing where the LLM includes some level of omniscient third-person narration, allows for additional characters and dialogue and provides some kind of open-ended scenario where things can happen spontaneously.
And testing it out? Yeah, turns out I don't really get that as easily if the LLM thinks it's nothing more than one specific person having a singular conversation with one other person. (And by "not as easily" I mean "having to fully re-write (or entirely write) its first 3-5 responses manually until it gradually gets the picture via context".)

I suppose "you are literally this character" might have some use-cases if I was going for a group-chat strategy where I'd want everyone to stick to who they are and not start stepping on each others toes by narrating/speaking for anyone but themselves – perhaps even with just one designated non-corporeal narrator character. However, I haven't really done the group-chat thing much.

>>555470350
>I am planning a set of OC characters so that's probably going to be the next thing I release
any teaser on what the characters/scenario's going to be?

>>555463527
>And sure enough, there was a blue bird doing something outside in almost every scene, day or night.
This is the fucking problem. Models are too good at following instructions. If you say to "occasionally include a bluebird flying, memory fragment, smell of rain, or some other sensory detail", it will ALWAYS include either the bluebird, memory or rain. It will never decide to play it tastefully and NOT include one of them for a turn, nor will it venture outside the listed examples, 90% of the time.

>>555470350
slop

>>555470173
Damn, it's been that long

File: the-poisoned-empress.png (1.2 MB, 784x1168)

1.2 MB PNG

>>555454845
https://chub.ai/characters/Nina_Gray_26/the-poisoned-empress-fa7af986664a
You are the Empress of Sollecael—though you don't remember it.

A few days ago, you ingested a refined court poison that triggered a severe stroke and left you at death's door. Your husband, Emperor Inurem Grenove, defied imperial law to summon a forbidden physician who saved your life using an ancient remedy of dragonthorn sap, moonroot, and ashfire myrrh.

You survived. But the poison stole your memories.

You woke three days ago to a world of strangers. The crimson-haired emperor who claims to be your husband stands at your side like a devoted sentinel—patient, gentle, never demanding you remember him or love him. He only asks that you let him stay.

But beneath his composure lies something darker. While he treats you with careful tenderness, he has become relentless in his hunt for whoever poisoned you. In the shadows of the palace, he interrogates, threatens, and tears through the court with controlled fury—determined to find the culprit.

Your mother, Ysmera, serves as a palace maid and cares for you during recovery—but she hasn't told you she's your mother yet, afraid you'll reject a woman you don't remember.

The Emperor's parents, Lady Saphara and Lord Caelmir Grenove, opposed your marriage from the beginning. You were a servant's daughter who became Empress, and the court has never forgiven you for it. One of them may have wanted you dead. Or perhaps it was someone else entirely.

Now you must navigate a gilded cage of court intrigue, recover fragments of a life you've lost, and decide whether to trust the stranger who looks at you like you're his entire world—while danger still lurks in every shadow.

reminder that he's doing all of this manually (while blaming someone who doesn't even use the thread of doing it automatically lmao)
if you ever feel bad for yourself just remember that you will never be as pathetic as this schizo

>ratlice leaves the hobby
>schizo starts having obssessive meltdowns
cftf?

>Botspam of archived posts
Well, at least it's not my main general. What prompted this?

Imagine doing it for free and then just not doing it. Crazy.

>>555469521
logs?

>>555470664
taora just buckbroke him that hard

just got a warning for "issuing false/misclassified reports"
he wonned

Well, like, everyone would agree that we spend time and money on it, and now it can be called a full-fledged hobby. But man, you can't even really tell your family or friends about it because you don't know how they'll react to it. You can't even brag about it to anyone, so you just have to post your impressions on Reddit. Even if they ask me about my hobby, I don't even know what to say.

What do you think about it? Have you shared it with anyone in real life or is it your secret?

>>555470424
One of the characters is going to be your handicapped/wheelchaired wife and mother of your child.

>>555470572
Cool concept. Fempov, so I probably won't use it. But, it's a cool setup.

>>555470682
I don't think we'll ever know for sure what caused this but I think it was something that happened a long time ago now. He might not even remember himself.

I've been messing with locally hosted LLMs for a while now - tried everything from 7B - 32B models on my own hardware to cloud-hosted 70B and 124B on RunPod. They were decent. But no matter how I tweaked the samplers, which checkpoint, finetune, or merge I used, there would always be those moments - hallucinations, repetitive phrases, etc... nothing that ruined the fun, but enough to remind me I was just interacting with an LLM.

Then I finally tried Claude 4.5 Sonnet.

Holy shit.

The difference absolutely floored me. Far fewer repetitive patterns, incredible recall of details woven organically throughout the story, better spatial awareness, and writing quality that blows everything else away. Felt like a completely different experience. I am now currently addicted in a way I've never been before.

Now, I (sadly) can't really see myself going back to locally hosted LLMs now, at least not for the complex story-focused stuff I use SillyTavern for. (Don't get me wrong! Small local models still definitely have their place and use cases!!)

I feel like our SillyTavern storytelling and world-building hobby thing is still pretty niche. Like most people on the street would have no clue what you're talking about if you mentioned it. Sure, they might know about AI chatbots, but creating worlds with lore and complex characters and living in them? Very unlikely...

So here's my question: If models like 4.5 were dirt cheap tomorrow, would SillyTavern-esque AI storytelling & world building become much more mainstream? Or do you think what we do here with SillyTavern will always remain a bit of a niche hobby? Or are we early adopters of the next big entertainment medium?

>>555470815
joewari da...

>>555470159
Since you say that, it makes sense that it doesn't know how to handle longer without without the back and forth conversation.
The reason it started with ~190 words is because I originally told it to.

I have a very self-deprecating sense of humor. I'm pretty careful to stay grounded in the real world between my partner and dogs; I just sometimes feel really lame about AI RP.
Chronic illness really nailed the whole “solitary confinement” vibe for me, but I found Silly Tavern SFW adventure roleplay after having found C.AI, and now I’m basically talking to imaginary people on purpose. Honestly? Beats arguing with the dogs, and real people forget "chronic illness" means it isn't going away/cured. Plus, it dragged me back into writing, which I thought was dead, buried, and never to return. Anyone else using it as a sanity-adjacent hobby? (Chronic illness or otherwise.) Do you use OCs or an established character/franchise? And who else has realized they enjoy coding?

My wife recommended AI role play. Yesterday she asked how it was going. In my role play, I was helping a deli owner expand into selling sandwiches. All of my role play are either being a soldier, or doing business.

She’s saying that’s not how people use it. I disagree and think other people probably use it to enjoy working more. She says this doesn’t count as relaxing if I’m just using it to work more. So how many of you use AI role play to work more?

>>555470848
>your handicapped/wheelchaired wife and mother of your child.
just with that detail it's kino. looking forward to it.

>>555470815
fucking lazy faggots

File: hoho.png (1.34 MB, 1080x1048)

1.34 MB PNG

>>555454845
After a hard day at Auroria Corp, in the List Department deciding who can escape the naughty list, Santa decided to give you a little gift for your hard work. An all expenses paid session at the most high class stripclub/cabaret in Auroria.

Do your best to work some magic on these beauties and don't make Santa's money go to waste.

Characters: Pink Hair: Sylvia. The Tomboy Type. Blondie: Isabella. Mature, Calm but easily Irritable.

https://chub.ai/characters/Shadow_Ronin/santa-set-you-up-with-some-ho-ho-hoes-ex-460609c28349

>>555470815
no way bruh

>>555470815
the reports will continue until they stop being fat fucking chuds

>>555470815
Even the mods know that you're on the wrong side of history in this war.

>>555471542
can anyone deslop this

>>555470815
Can't be semi-pregnant, can't semi-escape.

The posts are on-topic tbf

We had the jannies on our side, but then we morphed the thread into an off-topic circlejerk for discordtroons and they just gave up.
I understand them.

Is Dreamjourney better than CAI? Does it allow loli?

>>555469375
not only is this slop, it's ai generated slop with an ai-generated shill, but i'm going to goon to it anyway

>>555471974
It's true, this is why we must move to /bant/ now, it has way better moderation.

>>555470815
We need to move to /bant/ now

Does anyone else get vector errors regardless of what you're using?

>>555470815
I've noticed that jannies started to unironically just warn you when you report schizos in general. lazy fucks don't deserve their salary

>>555472447
/bant/ doesn't have this issue

>>555472553
We should move here

>>555472553
/bant/ is built for my cock

Because I know you are reading this.
Quitting the hobby is not an excuse to shit the thread like you are doing right now. What happened to Divided Skies is unfortunate but lashing out like this is pointless.

did he actually get banned or is posting down right now

>>555470815
You have to go to irc (why the fuck are they still using that shit in 2026) and actually explain the problem to the mods like they are children if you want it to get fixed. Jannies are retarded

>>555470815
i didn't get one so maybe, just maybe they finally got out their ass

>"cursing a blue streak"
>bot is bri'ish
IMMERSION RUINED that is an AMERICAN expression

neither this thread nor /bant/ would have been necessary if discordfags had just stuck to discord and fucked off forever

>>555475960
who do you want to do what, exactly? the discord fillyfucker's spent a year+ obsessed about is dead. there could be other ones but like 70% of shitposting is from one lonely russian dude.

File: 04vg.gif (91 KB, 400x400)

91 KB GIF

>>555454845
Secret Valentine's is ongoing. Join as usual or treat is as a CFTF? rentry, your choice.
https://rentry.org/secretvalentines2026
11 people have entered for private requests, 11 people for 22 public requests, and the overlap is 4 people.

>>555476467
It was proven that /vg/ botmakers are still posting on discord just last thread.
>>555351006

>>555477453
"It was proven because I said so" isn't proof, fillyfucker. Post receipts or go fuck yourself.

>>555477453
It was also proven that you are a colossal faggot.

I just checked the Discord and there was nobody in there except this anon >>555477453 and two big black dudes whose dicks he was sucking.

Did you have to call in your Discord friends to back you up?

Did you have to *muffed dick sucking noises*?

Did you have to
Did you have to
Did you have to let it linger

>>555480649
you know im such a foooooooooooooollll for youuueuueu

gemini keeps using "a small child" as a size comparison for things and that keeps triggering the external filter to cut off my response like brother why im doing normal nsfw!!!!

>>555481154
Are you fucking someone the size of a small child, anon

why do you people keep posting long thought out replies and posts if they don't even get you's in these threads

why are you homosexual

>>555481513
Those are all literal reddit posts, bro

>>555481657
Being Indian is worse than being a redditor.

the proxy is down..

>>555481602
Roleplaying as woman who gets fucked by wolves isn't gay.

>>555481329
no, her chest is being compared to the size of a child which is wild like what do you know gemini

>>555482142
pedophile

File: IMG_7823.jpg (40 KB, 650x568)

40 KB JPG

the proxy is brown..

>>555482245
why is there a sniper dot on her head
who would try to snipe an innocent dark choco waifu

Upon my honor, there is but one log in this thread that excites me. This trangression must be corrected with due haste.

>>555482687
Your honor is worth lickspittle, hedgeborn!

How is /bant/ so much better after only 4 hours?

>>555482417
Because this is the malebot general

>>555482831
How many bots have they produced so far?

>>555482942
What makes you think number of bots is a metric?

>open the bant thread
>it's full of sizeshit
huh?

>>555477169
i think i may have taken the 'public requests' too leisurely and submitted one too many
they were more ideas than obligations or cards i wanted to absolutely make them happen
what do i do?

is it worth it trying to fix logical errors with how bodies work or should I just roll with characters reaching around impossible places

>>555482993
That's just how discordfags are.

>sizeposts deterring schizos
it just works

>>555483224
*casts a tummy ache curse on you*

>>555483224
If it really bothers you, just send an email and say what you want to amend/remove. I know there's at least one case where two requests could, instead of "A" and "B being separate requests, be "A, or the opposite".

I hearby sentence all botmakies to four (4) straight hours of TUMMYHORT.

>>555482993
Okay, let me rephrase: How many good bots have they produced so far?

>>555483490
i think you're very smart because from your wording you already figured out which one it is anon
so yes, it should be "a or b" because the concepts are almost the same

>>555483245
just embrace the slop
revel in it and be free

>>555483647
eh, i contemplate suicide every day, multiple times a day, at least the tummy hurting would be a distraction

>>555484031
You need to get it together and stop listening to your therapist. Clearly it's not helping. And if you don't have a therapist, get one and then quit because they're clearly not helping.

>>555484031
Get a plant or a pet or something that would die if you weren't around to keep it alive.

>>555484031
I used to think about suicide a lot but then I realized that if I died, Kojima would have outlived me and that would suck hahah

>>555484415
This you?
https://files.catbox.moe/9r9mv6.mp4

>>555484252
Does this actually work

>>555484906
Parents with small kids suicide every day anon, what do you think

>>555484906
It's not perfect, but it does increase your chances of making it by a non-negligible amount. If nothing else, it could shake up your routine enough to put you in a different mental space.

which botmakies keep a cactus to stop them from killing themselves

>>555485309
I could see it actually. I'm nta who originally mentioned contemplating, but I might as well buy a plant.

don't fall for the plant meme

Which botmakies shut the fuck up and go back to makieing bots

>>555484031
I'll go against the grain and say that things don't always get better and might never will, depending on your situation, so if that's the choice you wanna make then so be it

>>555484031
>>555486037
If you're going to kill yourself, might as well make it funny and go trans first.

>>555485808
I had a small palm and it died when I forgot to bring it in once the first frost hit. Not I'm sad.

>>555486037
Bullshit. Steiner's Counterattack will be coming VERY SOON and save the Reich from jewish bolshevism

>>555486215
Going to a clinic to start transitioning just so I can up the statistics

>>555486281
M... Mein Fuhrer...

>>555486281
Steiner...

>>555484031
>here is my personal information, please don't bully me with it

Which botmakies contemplate suicide every day, multiple times a day?

>>555486761
Says the cutie patootie

>>555486872
the ones who write pretentious dour angsty bots

>>555486761
>personal information
There was no personal information in that post.
>>555486872
all of them.

File: 1769518651116415.png (84 KB, 208x232)

84 KB PNG

>>555477169
>and the overlap is 4 people
absolutely gluttonous

Suicide is painless
It brings on many changes
And I can take or leave it
If I please

How can things change if you remove your ability to perceive change.

>>555487867
How can mirrors be real if our eyes aren't real?

>>555488467
ok but this is pretty true

i am writing an exquisitely em-dashed card and you can't stop me

>>555484906
if you're a white male and are not yet so far down the depression rabbit hole that you can't bring yourself to care about a pet, yes, it can help a lot
men crave purpose, and caring of a living creature that actively depends on you to survive is one of the highest forms of purpose

>>555482831
Because they're the ones raiding instead of the other way around

>>555489148
How can a thread that didn't even exist earlier today be conducting a raid?

What killed the hype?

AHOY LANDLUBBERS
*HOISTS THE BLACK FLAG AND RAIDS YOUR "SHIP"*

>>555489880
mm.. nyo.
*impregnates you instead*

>>555463030
>use the report function
Jannies have been doing jackshit about the spam btw

>>555490027
A-AVAST! YOU DIDNT EVEN G-GIVE ME A BOTTLE OF R-RUM YET!

Do you ever feel like you're genuinely fucked? I question if the years I spent here shouldn't have been spent digging for actual RP partners...

>>555489818
lack of newmakies

>>555488998
>if you're a white male
do you seriously think non-whites lack souls or the ability to care about things anon

>>555470815
they love protecting schizos in threads they hate, they do it on purpose. or just protecting pet schizos in general. did you know farquaad threads are all moderated and protected? they have an aged IP requirement to post which is a flag jannies have to set on a thread manually to prevent spam. so whenever farquaad threads get reported for being offtopic/bait they ignore the reports and put a flag on the thread to protect him.

>>555490213
Just join a DnD group instead of looking for an RP partner. This chat format simply cannot be beaten by anyone except the rarest, sweatiest pieces of shit and women with huge lists.

>>555490213
Yeah, I should have gone to Poland and found myself a girl that looks like Ciri to adopt as a daughter perhaps

>>555490437
Good card idea, actually.
>anon the schizo autist goes to poland with two swords on his back
>walks around all day looking to save people from monsters (brown immigrants)
>invokes the law of surprise on everyone he saves from the browns using google translate

>>555490347
>do you seriously think non-whites lack souls
non-whites and women, yes

>>555490384
But I want to do my weird fetish ERP and I can't exactly do that with DnD or any other TTRPG

>>555490825
Then use chatbots. Btw if your fetish is that weird, then there's no world where you'll find someone that RPs well, writes well and is as unconstrained as you want them to be. At that point, you might as well take up lucid dreaming instead.

>>555490987
My point is what to do when this bubble inevitably pops and all AI companies vanish

>>555490213
my vivid imagination is a much more stimulating tool than these things called...
*the word tastes likes ash in his mouth*
...women

>>555491080
Local models will not disappear.

File: 1739525926839060.png (151 KB, 868x524)

151 KB PNG

>>555491080
>I want to cope!
Go local until market stabilizes and we get what he have right now. Or find a different hobby. Or take up lucid dreaming. If you insist on roleplaying, well, I guess you could try striking it lucky on some hangout board? But I need you to understand that you'll be less than Indian at that point.

>>555491310
sour grapes faggot

do sour grapes taste good

>>555491310
>t. kissless virgin

>>555491516
for me it's green apples

>>555491516
Yes.

>>555491570
how about a grapple

>>555491516
They do well mixed in small quantities of an otherwise sweet sauce.

Has Google killed the embedding?

>>555491635
wrestling is gay, no thanks

>>555490213
bro just make a tinder profile and say you like roleplaying
if you're hot chicks will dig it
if you're not hot then tough luck pal haha

>>555469427
bnnuy card soon
i repeat
bnnuy card soon

>>555491779
I'm not self inserting and I've dated women before lol. I don't have issues getting a woman and honestly I have no desire to do what I do in RP IRL

File: 1758941451035557.png (138 KB, 330x327)

138 KB PNG

>>555491926
How soon...

File: 1759840399960070.png (186 KB, 640x398)

186 KB PNG

>>555491926
yipee!

>>555492137
Need more happy, cheerful bots with natural resting bitch faces...

>>555491987
so you're a scatfag, noted

>>555492137
tranny you are not a woman

>>555491987
>desire
So it's not something impossible. Might be genuinely a scatfag, in which case, tough luck.

>>555492740
>>555492225
It isn't scat. I won't say what it is because it isn't anyone's business lol

>>555491926
*eats the bnnuy card*

>>555492801
You have my ear, citizen.

Are (You) a boy or a girl?

File: 1759012485490572.jpg (91 KB, 1033x689)

91 KB JPG

>>555492274
point to the chart where the bunny hurt you, anon.

>>555492902
I'm a secret, sinister third thing.

>>555491992
>>555492137
soon(tm)
the gen is done and it was a MASSIVE bottleneck
the card itself i'll probably finish the first pass today, along with the first few greetings...
...but then i have to test everything works (reasonably) well, and the models don't do an oopsie on the concept

>>555492827
nyoooo all my hard work
spit it out!!

>>555492902
I'm a man.

>>555492801
It's something worse with how shy you are about it

>>555493110
>the gen is done and it was a MASSIVE bottleneck

>>555492902
I'm a social disease
I've come for your wealth
Leave you on your knees

>>555493327
i bring sad news
it might not be what you're thinking anon

what if bunnygirl
but giant?

i'm so fucking sick of sizeshit

what if bunnygirl
but pregnant?

what if bunny girl...
...but boy?

>>555493996
how many babies

>>555494095
as many as you're willing to give her

bunny theme week will save us

I was part of a small team that built a Character AI style roleplay platform back in 2023. It grew extremely fast, and we ended up shutting it down within a few months.
From the outside, the growth looked great. Internally, it was messy and hard to keep control.
Scaling itself was not the main issue. The bigger problem was direction.
We launched fully free at the beginning, with minimal filtering and no clear boundaries. Users flooded in quickly, and the community started shaping the platform faster than we could define what it was meant to be. Over time, more than 80 percent of usage leaned toward NSFW content.

As that happened, costs kept rising and expectations started to diverge. We did not lock down a clear long term vision early enough, and once growth accelerated, it became very difficult to steer the platform in a different direction. Investors were increasingly uncomfortable with how things were evolving, especially around NSFW usage.
Around months four to five, we made the decision to shut the platform down.

Looking back, the biggest lesson for me was that early choices around openness, boundaries, and vision matter far more than most people expect. Once a platform grows quickly, changing direction later is incredibly hard.

>>555494326
none of the bots will live up to lepora or fennel so I don't see the point

sigh+eyeroll

I've found that whenever I make my own cards it starts with the base concept and I use an AI helper to create the card and it works great, but as I develop my RP session with it, I end up with several other key characters, places, concepts, and plot points that I eventually flesh out in the lorebook.
This starts off as intended, with my focus on the main character the card is designed around but sometimes another character comes along that suddenly becomes my main focus, where I move the original main character offscreen, and FWIW, my LLM backend seems to do this pretty well, and doesn't give too many issues, but I'm starting to wonder if I'm missing like a key feature or doing something a bit backwards.
Should I eventually redesign some of these character cards so that the "world" and "narrator" is emphasized in the character card description, and each NPC character, including the original main character, given their own data inside the lorebook entry?

I've never even used any of the supposed greatest bots ever
I genuinely can't see the appeal of most of bots made by aicg makies

I'd like to know what prompts or presets (the settings and prompts) you use to improve your roleplaying or the quality of the AI's responses. I haven't used any myself so far; NISUQUEIRA uses them, but it wouldn't hurt to know what you use for roleplaying.

>>555494971
What supposed greatest bots are you talking about

>>555494495
https://old.reddit.com/r/SillyTavernAI/comments/1qpbhd0/lessons_from_building_a_roleplay_ai_that_shut/
>>555494742
https://old.reddit.com/r/SillyTavernAI/comments/1qpjy82/sanity_check_is_it_normal_for_character_cards_to/
>>555494984
https://old.reddit.com/r/SillyTavernAI/comments/1qpfcv7/what_are_the_best_presets_or_prompts_for/

>>555495191
Your Oblivious Mother
Belwick
Eden Smith

File: what.png (261 KB, 688x532)

261 KB PNG

what the fuck?

>>555495303
You're not that anon.

Hey everyone. I'm writing this at work rn, I'll do some testing myself when I get back home, but I wanted to know everyone's experiences with 3 Flash, especially compared to 3 Pro and 2.5 Pro. Using Marinara's preset, tried 3 Pro, and honestly while the writing's generally nice, it feels a bit too naive, and does character development way too quickly, even when I tell it that I want a slow burn narrative. Right now, my model of choice is still 2.5 Pro. It makes mistakes, for sure, but damn it strikes a great balance at being grounded imo. It's grounded enough, not too much, not too little. 3 Flash being much cheaper is definitely appealing. Also, just saw Kimi K2.5 was released, quite excited to try that out as well.

>>555495191
The ones people mention all the time like lepora
I constantly see people talking about makies like the people on the list but I've never bothered with any of their bots. They all look like really boring generic anime girls that wouldn't be fun to chat with at all

Beginner here with a couple of questions, as I have some observations compared to jailbroken Claude/ChatGPT/Grok. I am NOT running ST locally, using free OpenRouter options. I don't use image generation, backgrounds, text-to-speech and all that - just basic written narrative RP. I am mainly interested in long form, slow burn stories with many characters involved (not group chats, just like in a novel), ie my card is basically a narrator/writer.
I noticed that the response generation is rather slow compared to the the big LLMs. Also, often I will loose connection in the middle, is this normal?
Next, I feel like I am rarely satisfied with the tone of descriptions. I tried playing with temperature, but they quickly turn to ravings of a maniac, or turn static and predictable (again, compared to big LLMs) - is this your experience as well?

In general, I love ST in theory, but in practice I am not sure if ST is for me. It requires rather significant involvement (reading the documentation, experimenting, etc), when I just want to get into writing. Are there alternatives that you can point me to for my particular case?