File: 72c96850fe7a359d888aca1d9(...).jpg (35 KB, 298x626)

35 KB JPG

ANCHOR

THIS IS IT, LUIGI

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

1.31 MB PNG

>>555228967
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

>>555228967
Here's how I've learned to make AI push back and create actual stakes in my campaigns.
Where did I learn all of this? While building Tale Companion and helping out my users to fix similar issues. I've never been a fan of high-stakes campaigns for AI RP, but these techniques changed things.
>1. Prompt for Consequences, Not Just Events
Most people prompt their AI like this:
-Be immersive.
-Create interesting encounters.
That's too vague. The AI interprets "interesting" as "entertaining," which often means giving you what you want.
Instead, tell the AI to be a fair world, not a friendly one.
Try instructions like:
-NPCs pursue their own goals. They don't exist to serve my character.
-When I fail or make poor choices, show me the consequences.
-Don't let me talk my way out of everything. Some NPCs are stubborn.
This alone changed my experience dramatically.
>2. Define What Failure Looks Like
Here's the thing: AI doesn't know what "failure" means in your story unless you tell it.
Give it concrete failure states to work with.
In my master prompt, I include something like:
- If I'm rude to important NPCs, they remember and treat me accordingly.
- Combat can result in injuries that take time to heal.
- If I ignore a quest for too long, the situation worsens without me.
The AI needs permission to make your life harder. Most models are trained to be helpful, so they default to smoothing things over. You have to override that.
>3. Use Antagonists, Not Obstacles
There's a difference between an obstacle and an antagonist.
An obstacle is a locked door. You pick it or break it. Done.
An antagonist is someone who wants something that conflicts with what you want.
When I define my villains and rivals, I give them:
-A specific goal they're actively pursuing.
-Resources and allies.
-A reason to not just wait around for me.
This makes the AI treat them as actors in the world, not just boss fights waiting to happen.

get a life you fucking freak

>>555229728
>4. Explicitly Request Tension
I know it sounds obvious, but you can just ask.
At session starts, I tell the AI what kind of tension I want.
Examples:
-This session should feel tense. Someone in my party is hiding something.
-I want to feel outmatched. The enemy should seem unbeatable at first.
-There should be a moral dilemma with no clean answer.
AI is remarkably good at executing on specific emotional beats if you name them upfront.
>5. Let the AI Say No
This was hard for me to learn.
Sometimes the AI will push back on something you want to do. Your instinct might be to regenerate or steer it back.
But if you've set up stakes properly, that resistance is the story.
My rule: if the AI gives me a "no" that makes narrative sense, I roll with it. That's the whole point. If everything always works, nothing matters.
>Why This Matters
Stakes create investment. If your character can't lose, victories feel hollow.
The best sessions I've ever had were the ones where I genuinely didn't know if things would work out.
That uncertainty is what makes AI roleplay feel like an actual story instead of a power fantasy. Not that power fantasies are bad. They're just different.

File: 1744062801569217.gif (2.1 MB, 260x274)

2.1 MB GIF

>guys if you tell the AI to act a certain way, it will act that way!
is this real?

>>555229783
Honestly, the easiest thing you didn't mention is switch models. Opus will almost never be truly cruel, no matter how much you prompt it. Gemini on the other hand

>>555229876
>>555229895
you are responding to fillyfucker copypasting reddit posts

File: neitai.png (982 KB, 935x937)

982 KB PNG

>>555228967
Guild newcomers: twin sisters with opposite personalities, clearly in need of guidance, even if Tai refuses to admit it.
https://chub.ai/characters/AeltharKeldor/nei-tai-twin-adventurers-23d52f87aeea

Tai and Nei are deeply bonded twin sisters who have never known their family. They grew up in an orphanage, where they were raised together until the age of 13, when they were adopted and taken to a nearby village. Rather than being treated as true daughters, they were expected to work long, demanding hours on a farm, handling heavy labor from a young age.
To survive and remain together, they endured this life for years, relying only on each other.
At 18, having grown weary and fed up with the harsh routine of village life, they decided to join the guild as adventurers. They have been guild members for about 10 days, during which they completed two D-Rank gathering quests and one D-Rank delivery quest.
Tai and Nei always stay together, sharing nearly every aspect of their daily lives, including meals, rest, travel and decisions. Even brief separations cause them noticeable unease and fear.

10 Greetings! 100% handmade.

oh, he's MAD mad

File: Deepseek.png (197 KB, 397x512)

197 KB PNG

Do you still stick with DeepSeek despite the gazillion other models available right now?
I have tried almost everything GLM, Kimi K2, GPT, Claude, LongCat Chat Flash, Mistral, Grok, Qwen but I ALWAYS eventually just return to the whale.

>>555230401
howd you get this pic of me?

File: Deepseek2.png (268 KB, 487x666)

268 KB PNG

>>555230458

>>555230401
Claude, sorry. I'm not loyal or anything, I'll try other models, but I always go back to Claude.

File: Deepseek3.png (212 KB, 491x597)

212 KB PNG

>>555230742
I like DeepSeek more, I will NOT be defeated by this Claude psyop

>>555230825
Bro , why do you have tons of deepseek themed gigachad photos?

File: Deepseek4.png (190 KB, 601x608)

190 KB PNG

>>555230885

File: noooo.png (330 KB, 1080x1695)

330 KB PNG

>>555228967
It's over...

>>555230401
3.1/3.2 sucked ass so I have no hope for v4. 0324 and 0528 were good because they were a little underfit and not totally pozzed.

>>555230401
It's rather affordable. I can goon for a couple of hours and only wrack up about 10 cents of payment.
It's the only reason I keep at it.

>>555230401
I am a fan of big and complex prompts and character cards. I also mostly use my cards for writing rather than RP... And so far I though that Claude was BIS in everything that was connected with reasoning. But now it seems like a mistake. DS speciale - does reasoning much better, thanks to "Extremely long" and deep thinking it is capable of. All while being noticably cheaper than Claude. And much less censored too!

>>555230163
Finally, someone for Gravis to take into a dungeon and watch... totally succeed at their quest.

File: Liora.png (591 KB, 497x835)

591 KB PNG

>>555230163
>>555228967
Also check my other bot!

Liora has been given an important quest to deliver a sealed letter. The Guild Master has assigned you to escort and protect her.
https://chub.ai/characters/AeltharKeldor/liora-the-guild-receptionist-4b9ca6178b78
Liora is a 24-year-old rabbit-kin and the head receptionist of the Guild. After losing her parents to a plague at age eight, she was sent to an orphanage, where her exceptional memory and people skills were noticed. A letter from the matron led Guild Master Sylvara to take her in and raise her within the guild. Trained from a young age in guild administration, Liora officially became a receptionist at fourteen and has served in that role ever since. She manages quest registrations and related paperwork. Known for her perfect memory and warm demeanor, she is the welcoming face of the guild.

>>555228967
This is an SQLite database containing 821,709 Characters data from JanitorAI and PepHop.
https://web.archive.org/web/20250904032906if_/https://projectbukkit.github.io/

>>555231782
aieeeee use hand and face detailer saar
I beg you saar please do the needful

>>555228967
Whoever made the other thread should be dragged out into the street and stoned.

File: d100.png (93 KB, 977x507)

93 KB PNG

>>555228967
This is an experimental open source system for integrating randomisation into AI roleplays. One system, yet it works for things like skill checks, plot-twists, combat, narrative pacing, solo roleplay oracles, etc. The idea is simple: Have the AI generate a weighted-probability roll table of 5 tag-based possibilities for what happens next, and have it roll 1d100 to pick one of these possibilities. In effect it functions as an on-the-fly rolltable for every response. Give the prompt a try with your favorite AI frontend. You will need a proper GM/narrator/character prompt in addition to the core Tablature prompt for the best experience. Simply append the Tablature prompt at the end of your roleplay prompt. Download the Tablature RPG System main document for more information.

https://matty1v1.itch.io/the-tablature-ai-ttrpg-system

>>555231893
front end for reading the db?

>>555233012
sqlite browser
^ copy this into a search engine.

>>555232489
https://www.reddit.com/r/SillyTavernAI/comments/1png7vn/tablature_an_ai_ttrpg_system_free_and_open_source/
neck yourself you worthless zigger parasite

After my very unfortunate experience with OpenRouter's Deepseek shenanigans, I am unsure on where to go if I want to use GLM, because Deepseek's API is getting more and more boring for me. Is OpenRouter worth it, or are there lobotomized models just like with Deepseek? And if there are other providers you know of, could you tell me about them?

>>555234134
Parasail is good and very cheap, Chutes has a sub plan for like $8/mo.

Do we have real roleplayer here?
Does AI Roleplaying make you a better Roleplayer?

What I mean by that is does the experience of roleplaying with a competent AI, that largely writes decent to good prose and dialogue, improve a player's foundational skills in writing, storytelling, character development, dialogue, and driving a scene or plot?

gpt/gemini/glm/deepseek/moonshot/qwen proxy
https://velvet-conflicts-grade-unit.trycloudflare com/

>>555234294
Ehhh, everyone has a different sense of what a "good" roleplayer looks like, but I wouldn't say that it can make someone into the absolute best roleplayer.
I think AI's really good at bringing out levels of writing and roleplaying in someone that might not've brought out all; while roleplaying with humans IS superior, it definitely isn't as consistent or stress-free as AI RP.
Humans in RP are a mileage may vary sort of thing. You may get some of the best RPs of your life from them, or some of the most mind-numbing, incoherent RPs to have ever walked this Earth. Humans have the best ability to RP no contest, but there's a lot of tradeoffs that can make things inconvenient. There might also be some people who will absolutely say no to certain narrative choices, either due to some underlying sense of ego getting in the way, or they just don't find it comfortable to partake in for whatever reason.
Compare this to AI; it's a lot more consistent, and will absolutely not gaf what kind of RPs you want to do with it. It can also write out text near-instantenously, which is one hell of a drug once you're used to its spontaneous writing.
Imagine you're a boss with two workers; one outputs some of the best work you've ever seen, but he eventually quits, and everytime you try to find a new replacement for him, the new guy gets drunk and occasionally gets into car-crashes after skipping work; leading you to try and find another replacement.
Meanwhile, the second guy might not output some of the best work you've seen, but you can lead him into the right direction, and he's always there when you need him to come into work early, never complains, and is never trying to skip work or get drunk; an above-average model employee who's only issue is a wish-washy long-term memory.

>>555234294
https://www.reddit.com/r/SillyTavernAI/comments/1qlxlzc/does_ai_roleplaying_make_you_a_better_roleplayer/
>>555234134
https://www.reddit.com/r/SillyTavernAI/comments/1qmpb01/what_provider_do_you_use_for_glm_47/
>>555230163
https://www.reddit.com/r/SillyTavernAI/comments/1qndfan/nei_tai_twin_adventurers/
freak

I've started abusing OOC

>>555234301
slop

File: UP1.jpg (1.23 MB, 3840x2160)

1.23 MB JPG

How long do you think it will be before LLMs get integrated into video games?
I mean, the new DLSS is black magic fuckery AI where playing on Ultra Performance looks nearly as good as native if you're playing at 4k, pic related.
I then use AI to make games have surround sound. So AI visuals, AI audio. Soon everything will be AI.

I use Sillytavern for long term pseudo TTRPG roleplay that involves heavy amounts of multiverse travel. And, I'm just now getting into vectorization stuff What data do I put into lorebooks and what do I vectorize? Also, how do I specify what characters know each individual vit of vectorized info?

>>555234719
https://www.reddit.com/r/SillyTavernAI/comments/1qm49eg/what_do_i_vectorize/

>>555234301
vintage 2009 tumblr sloppa

>>555234964
>tfw I can legally fuck someone born in 2009 now

futa centaurgirl with a 10m hose getting stuck when summoned, or catgirl rummaging through your trash and inhaling your incredibly smelly discarded clothes uncontrollably?

>>555235041
and my back hurts now even if I just sit in a chair all day

>>555235245
Exercise, fatty. Even an hour of walking/jogging does wonders for you. Or go see a doctor.

I NEED AMBIENT SOUNDS!
Since my role-play scene starts with a hot summer day... I've looked up summer ambience on YouTube and let it play... I nodded to myself and saw that it does indeed improve the overall mood. It broke the silence in a palpable way~ ...almost immediately I had the idea... this would be sooooo much better with the right music,
...so I looked up anime/slice-of-life/summer music on YouTube.
Trust me when I say... the sounds and the music kick the entire role-play experience up SO high, you won't even want to go back down again. To me, it all now feels like watching a very personally tailored Anime...like writing the perfect moments of life you'd always want to experience, which cannot happen~ *heavy sigh~*
...so, the short summary of this ...if you haven't yet, I wholeheartedly recommend trying to add appropriate background ambience and music to your experience. We have ears too, after all. The audio is just as important.

my back hurts because my focks wife keeps demanding sex
she says it's the only way for her to grow more tails

>>555235775
https://www.reddit.com/r/SillyTavernAI/comments/1qlbuzq/just_had_a_great_idea_and_experience_improving/
I know you're doing this manually so I really have to question your sanity at this point. Do you think people just stop nooticing if you wait a few minutes longer?

I use AI to roleplay and like to interact with the character like im there, not long storywriting responses. My problem is that the AI likes to progress the story too far in responses.
For example: in my response i agree to protect them while they travel to the next city. Their response: they thank me, ask when i want to leave, continue talking about something else, then they get up and walk out the door. They progress way too far, not letting me answer the question and ruining everything.
I tried limiting the response tokens which somewhat helps, but it will cut it too short often. Id rather just leave the response tokens long so they can give me the full response, but not progress too far. I tried putting something in the prompt to help, but then i have issues like the AI wont progress the story at all until i make it progress, i dont want that either. I want to be a part of the story instead of controlling what happens in it, if that makes sense.

Anyone have ideas on how to help with this?

>evil general (mother) with red qipao...
yeahhh...
I beg to serve, your wish is my law
Now close those eyes and let me love you to death
Shall I prove I mean what I'm saying, *begging*
I say the beast inside me is gonna get ya, get ya, get

>>555236138
https://www.reddit.com/r/SillyTavernAI/comments/1qkknor/response_progression/
get a job

I was thinking about reading a book sometime. Should I?

File: Anti-slop.png (89 KB, 584x618)

89 KB PNG

>>555228967
Widespread LLM adoption has introduced characteristic repetitive phraseology, termed “slop,” which degrades output quality and makes AI-generated text immediately recognizable. We present Antislop, a comprehensive framework providing tools to both detect and eliminate these overused patterns. Our approach combines three innovations: (1) The Antislop Sampler, which uses backtracking to suppress unwanted strings at inference time without destroying vocabulary; (2) An automated pipeline that profiles model-specific slop against human baselines and generates training data; (3) Final Token Preference Optimization (FTPO), a novel fine-tuning method that operates on individual tokens, surgically adjusting logits wherever a banned pattern has appeared in an inference trace. We demonstrate that some slop patterns appear over 1,000× more frequently in LLM output than human text. The Antislop Sampler successfully suppresses 8,000+ patterns while maintaining quality, whereas token banning becomes unusable at just 2,000. Most importantly, FTPO achieves 90% slop reduction while maintaining or improving performance in cross-domain evals including GSM8K, MMLU, and creative writing tasks. In contrast, DPO suffers significant degradation in writing quality and lexical diversity despite achieving weaker suppression. We release all code and results under MIT license: https://github.com/sam-paech/auto-antislop.

>>555236629
I've read 12 books so far this year by having AI summarize them for me

I wanted to share all the extensions I use that I think are essential to have a better experience using ST. If you have more extensions that are useful please write them below. Here is the list:

>AlternateDescriptions:
A SillyTavern extension that allows you to save and manage multiple versions of character fields within a single character card. Perfect for experimenting with different character concepts without losing your original work. — https://github.com/nbrown725/SillyTavern-AlternateDescriptions
>CharSwitch:
An easy way to switch between the recent characters that you have chatted with. — https://github.com/LenAnderson/SillyTavern-CharSwitch
>Character-Creator (CREC):
A SillyTavern extension that helps you create character cards based on your ST data with LLMs using connection profiles. — https://github.com/bmen25124/SillyTavern-Character-Creator
>CharacterStyleCustomizer:
A third-party extension for SillyTavern that brings advanced styling options to your characters and personas. Customize the look and feel of individual messages with precision, using per-character color settings and custom CSS. — https://github.com/RivelleDays/SillyTavern-CharacterStyleCustomizer/
>GuidedGenerations-Extension:
The GOAT of the ST extensions. It brings the full power of the original "Guided Generations" Quick Reply set to SillyTavern as a native extension. It provides modular, context-aware tools for shaping, refining, and guiding AI responses—ideal for roleplay, story, and character-driven chats. All features are accessible via intuitive buttons and menus integrated into the SillyTavern UI. — https://github.com/Samueras/GuidedGenerations-Extension
>InputHistory:
Use Alt+Up or Alt+Down in the input box to go through your last inputs and /commands. — https://github.com/LenAnderson/SillyTavern-InputHistory
>MemoryBooks:
An easy way to update your lorebook entries! — https://github.com/aikohanasaki/SillyTavern-MemoryBooks

>>555236629
reading (well written) a book after reading slop all day is like drinking fancy sparkling water after horse piss

>>555236908
>MoreFlexibleContinues:
Improves the “continue” functionality in the conversation, adding more flexibility and control over continuations. — https://github.com/LenAnderson/SillyTavern-MoreFlexibleContinues/
>Notebook:
It integrates a notebook or space for notes into the conversation environment.— https://github.com/SillyTavern/Extension-Notebook
>ST-Outfits: It manages character outfits, allowing dynamic clothing/style changes.— https://github.com/lannashelton/ST-Outfits/
>ProsePolisher: Polish and improve the prose/style of the generated text (corrections, refinements). — https://github.com/NemoVonNirgend/ProsePolisher

>>555237032
>QuickPersona:
It allows you to quickly change the active “persona” without long processes. — https://github.com/SillyTavern/Extension-QuickPersona
>rewrite-extension:
The Rewrite Extension enhances the chat experience in SillyTavern by allowing users to dynamically rewrite, shorten, or expand selected text within messages. Works for chat completion, text completion and NovelAI. — https://github.com/splitclover/rewrite-extension
>Roadway:
It helps you with narrative decisions or story paths within the conversation. — https://github.com/bmen25124/SillyTavern-Roadway
>SuperObjective:
Allows you to assign a high-level overall goal to guide the conversation. — https://github.com/ForgottenGlory/ST-SuperObjective
>WI-Bulk-Mover:
Allows you to batch move multiple “WorldInfo” (world/lore information) between locations or characters. — https://github.com/leandrojofre/SillyTavern-WI-Bulk-Mover
>WorldInfo-Recommender (WREC):
This is the brother of the CREC. It allows you to create lorebook entries by typing what you need. It uses chat history, world info, character cards, etc. as context to follow roleplay continuity — https://github.com/bmen25124/SillyTavern-WorldInfo-Recommender/
>WorldInfoInfo:
It lets you see which lorebook entries are active in your current chat — https://github.com/LenAnderson/SillyTavern-WorldInfoInfo
>WTracker:
Tracker for elements such as characters, objectives, or statistics within the conversation universe. — https://github.com/bmen25124/SillyTavern-WTracker

>>555235806
Dual cultivation with your foxwife...

>regex
>quick replies
>lalib
you don't need more

>>555230163
>{{char}} is not a single character. Instead, {{char}} represents twin sisters: Tai and Nei. {{char}} does not speak as itself; it roleplays directly as these characters when interacting with {{user}}. Each response from {{char}} should portray the voices, reactions, and personalities of Tai and/or Nei depending on the scene. The two interact naturally with {{user}} and with each other, staying true to their established dynamics.
bro just stop
>>555236709
kinda shit cause it's keeping the same sentence structure and just using synonyms

File: 021~2.jpg (107 KB, 619x752)

107 KB JPG

Four tails is ideal, actually, because it's not too cramped and they have good symmetry.

>>555237695
why are you responding to copypasted reddit posts you dumb stinky idiot

>>555237815
oh he's doing it again? it is what it is
either way that's from a real retarded paper

File: Gemini.jpg (24 KB, 840x350)

24 KB JPG

Guys, do yourself a favor and change Top K to 1 for your Gemini models, especially if you’re using Gemini 3.0 Flash.
This changed everything. It feels like I’m writing with a Pro model now. The intelligence, the humor, the style…
So, here’s a little explanation. The Top K in the Google’s backend is straight up borked. Bugged. Broken. It doesn’t work as intended.
According to their docs (https://cloud.google.com/vertex-ai/generative-ai/docs/learn/prompts/adjust-parameter-values) their samplers are supposed to be set in this order: Top K -> Top P -> Temperature.
However, based on my tests, I concluded the order looks more like this: Temperature -> Top P -> Top K.
You can see it for yourself. How? Just set Top K to 1 and play with other parameters. If what they claimed in the docs was true, the changes of other samplers shouldn’t matter and your outputs should look very similar to each other since the model would only consider one, the most probable, token during the generation process. However, you can observe it goes schizo if you ramp up the temperature to 2.0.

Honestly, I’m not sure what Gemini team messed up, but it explains why my samplers which previously did well suddenly stopped working.

File: 1764339304049084.png (1.55 MB, 1288x1610)

1.55 MB PNG

Robots.

i mean at least the posts are relevant
take this over any which botmakie wears a hat type shit

>>555238005
https://www.reddit.com/r/SillyTavernAI/comments/1in08t2/you_wont_last_2_seconds_with_this_quick_gemini/
yawn

>>555238095
idgi

For the past few months, I went into the whole craze of the Chatbot stuff, eventually giving a try in trying to run one myself, Since the first time was exciting.
But at this point, It such a freaking headache at this point and not really worth it with how much restriction there is with everything.
Want the big smart LLM that can be creative and follow instructions properly? Pay monthly subscription and have your chats non private. Oh, Also Censorship.
Want to host your own local model and actually have privacy? Get a company grade Graphics cards or deal with running a weak Models that get repetitive and fail to follow instructions most of the time.
Like, I enjoy the whole Roleplay chat stuff, but with the options currently, it simply isn't worth it. I just hope in the future this will get improved. Until then, I am taking step back.

>>555238106
they're both pretty bad since both are meant to introduce enough noise to drown out real discussion. also I'm pretty sure most of the what makie nonsense is driven by the same schizo.

>>555238429
https://www.reddit.com/r/SillyTavernAI/comments/1jq3nic/warning_just_got_banned_on_anthropic_for_using_a/

which botmakies think that which botmakiesposting is a tactic devised by the cabal to derail the thread

File: main.png (126 KB, 1215x810)

126 KB PNG

>>555228967
Browse character bots and lorebooks from various sources directly in SillyTavern.
Installation
Install with the SillyTavern extension installer:
https://github.com/mia13165/SillyTavern-BotBrowser
>How to Use
Click the bot icon next to your character list.
Browse cards, click one to see details, hit import to SillyTavern if you want it.

>>555230163
Gonna have to put them in a goblin cave. Sometimes you learn the hard way that a dagger and slings aren't really main weapon material.

>>555238557
which botmakieposting far predates the cabal

>>555237724
one physical tail, the rest are ghostly cool mirages

>>555238570
https://www.reddit.com/r/SillyTavernAI/comments/1q4x9xz/st_bot_browser_extension_v113/

>>555238557
taora (it's subtle but noticeable)
kalakan
koikoikoihat
Lunarcy
_purple
momoura
CUMSLURPER

>>555238689
that's honestly pretty good too, better if she can make them selectively tangible. now I'm imaginging a fox girl who can summon innumerable tails with some kind of Unlimited Mofu Works power...

>>555228967
I made a post here two days ago discussing how the best preset for you is typically a custom preset that you make yourself. I personally have found that using other presets acts as an incredible foundation for new ideas and even a foundation to RP the way you like.
However, if you want to have the best potential experience, it’s best to take the time and find the prompts that work best for your style of roleplay. By trying other presets, there are ways to roleplay that I didn’t even think were possible.
I realize also creating presets is time consuming and complicated. It’s also just simply fun to try things out to change it up in a quick convenient manner.

So for this reason I am officially releasing my preset, an updated version of what I uploaded I. My previous discussion. I was hesitant at first to put my work out there, but it seems like people wanted to see it, and I guess it only helps each other to share ideas and grow this new frontier hobby.

Introducing Freaky Frankenstein 2.0. The name is based on the idea that I essentially reverse engineered popular presets for GLM 4.7 such as Stabs, Kazuma, and Evening’s Truth. I pick and chose effective prompts that I enjoyed from these. Then I heavily researched effective ways to adjust an LLMs thinking process to double check its work to decrease slop output to force it to think and make sure that its outputting what the prompts are telling it to decrease the risk of things falling through the cracks. Some people will enjoy it, and others won’t.

https://freakyfrankensteinglm47.tiiny.site

File: 1745302324338470.png (2.96 MB, 3000x4000)

2.96 MB PNG

>>555228967
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

File: __komeiji_koishi_touhou_d(...).jpg (293 KB, 768x768)

293 KB JPG

I guess we just gotta wait for him to get tuckered out since the jannies are useless.

>>555238858
https://www.reddit.com/r/SillyTavernAI/comments/1qkw3z3/introducing_my_preset_freaky_frankenstein_20_for/

File: Aglae.png (1.04 MB, 927x932)

1.04 MB PNG

>>555228967
Three faces hide one judge: the worthy earn her protection, the unworthy meet their end.
https://chub.ai/characters/AeltharKeldor/aglae-secret-s-rank-protector-of-the-guild-6dfa5aaea8ff

Aglae is a legendary S-Rank adventurer, a living myth whose true name and face are known to almost none.
Born to a powerful northern noble house, she fled at sixteen, rejecting a life of silk and submission. Years spent evading pursuit honed her mastery of disguise. At eighteen she entered the guild as the timid novice “Nadine” a mask she still wears often.
Her unparalleled skill with the greatsword carried her through the ranks with terrifying speed, reaching S-Rank at thirty-two. The tragic loss of past parties left scars that forged a fierce, maternal protectiveness toward those she judges worthy.
For six years now, under secret orders from Guild Master Sylvara, she has been the guild’s unseen guardian. She walks among adventurers in two carefully crafted disguises.
To the pure of heart and truly talented, she is a hidden mentor and shield. To the corrupt or dangerous, she is swift, silent, and merciless.
Most feel only a vague sense of safety, an invisible hand preserving order. Very few ever learn it is her.

Note: Aglae reveals her true identity only to those she trusts completely and without reservation.

11 Greetings

>>555239113
>since the jannies are useless.
wasnt the whole point of the migration from /g/ to /vg/ the superior moderation?

yep, it's funny

congratulations, you make 'kan cry. are you happy?

I am trying to get SillyTavern set up and I am running into some issues with the image generation. I'm using ComfyUI Portable, which is up to date. I have my working workflow, I Export (API) [Save API doesn't exist anymore from what I can tell], even with Comfier Placeholders SillyTavern can't see it. Keeps giving me a "Failed to import character" error. I've tried moving it into the workflows folder, which makes it pop up in the drop-down in SillyTavern, but it still can't read it. What am I missing? Thanks

File: 1739252297005673.gif (999 KB, 500x260)

999 KB GIF

>>555238689
I usually went with "foxwife usually only shows one tail and hides the rest when out and about/or for convenience" but that's cooler.
>>555238798
That gives me the idea for like a fox girl martial artist who uses her tails for battle, and she has some kind of technique that lets her harden her tails or increase the amount of force behind their movements. This lets her do HnK/Jojo style punch rushes but with her tails.

GPT 5.1 writes so dang well (IMO a lot more naturally than Anthropic or Google models). But the refusal rate is ridiculous for anything more than standard vanilla NSFW, and it's much more "aware" of both its identity and OpenAI's policies than other models, making it impossible to gaslight.
For people who use it for spicier topics, what's your preset/settings/approach? Bonus if it's on Thinking ;)

Right now I'm switching to Gemini (3 Pro Thinking) or Sonnet (4.5 Non-Thinking) for anything restrictive, but then I can't even switch back.

>>555239609
I've tried, but it simply doesn't work. Even when combining methods like reducing reasoning intensity, altering prompt post-processing, and pre-filling, it only manages to confuse GPT-5.1 for a brief moment. Using a very long context for setup occasionally works, but it quickly wakes up and forcibly pivots the ERP back to a safe state. I miss the Polaris Alpha.

>>555239609
What preset are you using?
I use MarinaSpaghetti's preset for GPT and it has not refused anything, although, my tastes are kind of mild so I suppose it varies from user to user.

>>555239609
>>555239686
>>555239768
>posting comments now
https://www.reddit.com/r/SillyTavernAI/comments/1qmpxwp/is_it_possible_to_get_gpt_51_to_do_heavy_nsfw/

I have recently installed SillyTavern and I set up an OpenAi connection because I had some tokens left over there. My goal is to eventually swap over to Claude and use Claude Sonnet as a model Ai. However, whenever I try to do anything nsfw, I keep getting a message saying "I am not comfortable with the direction this conversation is heading" and then the bot tries to move away from the conversation. Is there any way to get around this?

I've seen a lot of people lately talking about extensions, sharing presets, etc. But most of these don't support my personal use cases well.
I'm wondering: how are you playing in ST (and/or beyond ST)?

When you do things in ST are you:
-Writing creative fiction?
-Playing RP as a player/character (AI as DM/GM)?
-Playing RP as the DM/GM (for the AI)?
-Playing an unstructured RP with the AI (Chat-like)?
-Doing non-RP entertainment?
-Using ST for productivity uses (coding, planning, education)?

And when you're doing these things, how do you setup a chat:
-1 user persona, 1 AI, simple.
-1 user persona, multiple AI characters/agents (group chat)
-No user persona, 1 AI (simple user/agent work, or system task-only)
-No user persona, multiple AI (watching AI group, or system task for multi-agent)

It seems like, based on what gets posted, the most common is a simple 1 user, 1 AI chat with either the AI as DM, or no DM (chat style). But I'm interested to see how much variety there is!

>>555240487
In order of roughly what proportion of my time is spent which activity:
1.RP as a character
2.Productivity uses
3.RP as DM
4.Writing creative fiction
Typical setup, again in order of commonality:
1.1 user, 1 AI
2.No user, 1 AI

I've never managed to get multiple AI's working together in a group working acceptably for my purposes. However, with capable models such as Claude I do tend to have the "character" card be a scenario or GM instructions, and keep the character descriptions in a lorebook bound to that chat - that works pretty well.

this wouldn't have happened if you hadn't forced Patter out of the hobby

>>555239118
>check the prompt out of curiosity
>it's schizo
dunno what I expected

hello I am real

>>555240874
nah

>>555238858
>>555239118
Been trying this preset out and so far I love it. It does *not* hold back on dark topics probably because of the Mandarin. I've tested a lot of presets for anti-slop and these one is just chef's kiss. I haven't run into "ozone" no matter how much the llm wants to say it.
I kind of wonder why though.

>>555238993
These opening messages are pretty shit tbqh

>>555240980
what did you expect from a slopper

Gemini 3.0

Do you guys think putting chat history at the start is better? Their official prompt strategies say to put all context at the top and the tasks at the bottom. Would having the card AFTER chat history be better for Gemini 3 then? Man I'm still trying to figure this damn model out.

>>555240487
I started out RPing as self-inserts in the worlds I like, but gradually the self-inserts became their own characters and my own role has shifted to being the "director". I set the structure and determine when a character should speak and what they feel/act, but largely let the LLM decide execution unless it just cannot get it right.
I also say my use case is less RP and more storytelling with constantly shifting 1st person perspectives, because I give my cards continuity and lorebooks I constantly keep updated to match their progression.
I almost always do group chats, if only because even 1-on-1 character interactions are both handled by the LLM most of the time.
Every now and then though I get the itch to do just basic chat RPs, both between myself and between characters. But my prompts and cards aren't really built for those kinds of interactions, so it usually doesn't work out. If anyone's got a good setup for that, lemme know.

File: 1767697238057145.webm (138 KB, 416x752)

138 KB WEBM

cftf?

>>555241072
Stop with the reddit posts. The spacing make it obvious.

*tilts head before bot can*
psshhh... nothin personnel

Any anti omniscience techniques yo use?
I use Celia with Gemini 3 Pro Preview, and many, many prompts, still doesn't work, I tell the ai to not give omniscience it makes the character make leaps in logic to somehow "guess the truth" or makes them agree, or suspicious based on whether it is truthful in my character background. Just completely annoying.

>>555241194
It's not a leddit post, what

>>555240980
I'm admittedly not very good with opening messages. But they are all based on things that happened in-game.

File: Anthropic.png (262 KB, 420x937)

262 KB PNG

With the recent “OpenAI Leader of Safety” hire and this “assistant axis” from Anthropic, do you think Claude will still be a solid choice for roleplay by the end of 2026?

>>555241721
There's something to be said about an assistant remaining an assistant and not drifting off into unwanted weirdness, that's a good goal. But I'm really disappointed with the talk of "harmful" behavior. Obviously you don't want public facing assistant AI to be spouting dumb shit like Grok does, but there's a line to draw somewhere.
It's words on a screen from a machine that ideally follows instructions. If I instruct it to be an asshat or evil or flirty or whatever, that's what I expect from it. It's the user's responsibility to not be fooled by clever words from a machine. Even instructing how to do illegal things is kind of a moot point IMO when you can search Google or go to a library for the same info.
These companies need to spend more time on being transparent and educating people about what AI is, what it can and can't do. Less time trying to censor it into boring uselessness. I don't need or want censored AI, but I do think it's smart to allow for censorship at the system prompt level to suit everyone's wants or needs.

It seems to me that Anthropic has lowered their refusals in the newer models, but who knows what will happen in the future. I know I won't have any interest in using it if I get the same refusals from new Sonnet that I got from 3.7.

>>555238689
this gives me an idea but I don't think it'd translate well into a card. essentially more tails = more wisdom because these extra tails are actually ghosts of elder focks

>>555241517
Genuinely how am I supposed to write a response to the first one

>>555241721
No. I'm very pessimistic with the way things are going. The big AI companies with proprietary LLMs are so up their own ass with making everything "safe" and "harmless", that I think open-source models are the only way forward eventually. (Eventhough open-source model's are often heavily trained off proprietary ones, leading to the issue of safety guardrails being more present lately.)
They won't get rid of roleplaying completely. They may even allow very vanilla NSFW/ERP. But it will be confined to an increasingly strict, company approved framework. Themes or kinks that are in any way controversial won't be a thing.
Maybe we'll be surprised by more positive developements, but so far it has only gotten worse.

>>555241894
My only disagreement with this is that after Openrouter published their data on roleplay tasks, it seems Opus and Sonnet were practically decensored outside of what I can assume is some pretty fucked kinks. It will even do gore, so whenever someone says 'it is censored' I cannot imagine anything it won't do, or that the person has something specifically banned by anthropic. Or they haven't used Claude in 6 months.

>>555241820
Go inside? Text her?

>>555241894
It's not just LLMs or AI, everything that's decided in a shareholder meeting is dumbed down as much as they can get away with. I recall not long ago, most "warnings" were about things too sexual/violent/gory/illegal, etc. Then it went from that, to pretty much anything NSFW. Now it's "sensitive content", under which they can put practically anything they don't like. In case of AI, open source and local is absolutely the way to go, as much as possible.

>>555241820
>Genuinely how am I supposed to write a response to the first one
>greeting literally ends with "With bated breath, she waited for him to enter from the back entrance."
Are you retarded?

>>555241820
Not him but you could just walk in? {{char}} is waiting for you to show up.

update on the paag alt bitch bot?

I am just assuming anything longer than two lines has been copypasted from Reddit.

>>555241894
I agree with you about the bleakness we're living in right now when it comes to AI RP, but I believe it's just a matter of time.
You see, all spotlights are on AI right now, and I doubt the big companies are willing to lose momentum with controversies et whatnot while they race. To me, they're just playing safe to ensure they will not only survive the AI tournament, but also profit big time. They're securing their sovereignty.
If I'd bet, I'd say that, in a few years, we'd witness these big companies having forks of their LLMs. Think of different flavors. "Gemini for Math", "Gemini for Coding", "Gemini for Creativity", "Gemini for Medicine", "General Purpose Gemini", etc. In other words, more contained models that can be updated independently and tuned accordingly.
I don't want to sound Mr. Obvious, but as far as I understand - and I have no data to prove, it's just an educated guess, - as much profit potential RP may have, I believe it's nowhere as close to what Google or Anthropic would make by investing in other areas (which is what they're doing right now.)

We're in the middle of a thunderstorm right now, but after it, a rainbow smiles.

what's new meta nowadays? do have clode but it's getting kinda stale, might be preset issue

Broke retard here, now that MiMo from OR is dead(rip), what proxy can I use?

I only usually use Deepseek via Open Router. GPT 4 is still one of the best for funny RPs and scenes, but it's way too expensive. Deepseek is good but has been feeling kinda flat lately.
Just to be clear: I don't like purple prose, I am not looking for some deeply nuanced RP or scenes, I genuinely like light hearted, kinda unhinged, funny stories and RPs and scenes. I hear a lot of praises about Claude, is it capable of writing like that? And if so, which model would you rec?

File: lol.png (277 KB, 606x517)

277 KB PNG

>>555242414
This got a snort out of me, didn't expect the AI to do the silly harem jealousy thing. Using Claude Sonnet 4.5.
Context is that the videogame this chat was based on, added vampires to the game, and you can use a syringe to see the quality, properties, rarity, and taste of someone's blood. My character's blood apparently tasted like stone-roasted sweet potato.
I would say that Claude is decent, if a bit hard to prompt for/with unless if you use a preset. Haiku is a LOT cheaper but it can be dumb and sometimes flat. Makes sense given that it's the faster and cheaper option but, yeah-

If you aren't roleplaying with Opus, you're better off not roleplaying at all.

>>555242565
What preset do you use btw?

>>555242701
https://spicymarinara.github.io/
I use Marinara's preset. It's pretty good as is, but I have tweaked it a little in places.

File: 1761009819196257.png (1.34 MB, 2047x2047)

1.34 MB PNG

oppai loli is the only way to prove you're real

Just an Appreciation post, I use this thing all the time. Looking forward to seeing where it goes in the future
https://github.com/mattjaybe/SillyTavern-Pathweaver

Just leave.
Schizos won. If anybody posts anything chatbots related now they'll just get mocked and called slop. Migrate to another general, start a new general, get on with your life.
This thread has degenerated into discord schizos talking about each other.
>just ignore it
No. Fuck you. You post no content, say nothing good about content while it's spammed with 75IQ negativity, and then expect people to make stuff here. Nobody has any incentive to contribute to this shit hole.

>>555242951
I saw this in another post recently and gave it a shot. Very impressed so far. I use it with GLM4.6 via Z.ai and it gives me good suggestions on where to go next. Really helpful for longer sessions when I feel like the AI has backed me into a corner.

>>555243393
>You post no content, say nothing good about content while it's spammed with 75IQ negativity, and then expect people to make stuff here. Nobody has any incentive to contribute to this shit hole.
i honestly agree with this and it's one of the biggest issues with this general, no one ever positively engages with any bots/logs posted while expecting other people to post bots/logs

>>555242347
I forgot MiMo free runs out today, fuck

>>555242139
>With bated breath, she waited for him to enter from the back entrance
this supposed to be no AI hand written no slop bots?

>>555243784
we have one of the highest populations of bad actors on the site outside of like, /pol/ and /r9k/. same with a lot of AI threads. just look at fillyfucker on her period shitting up this thread or generic asshats like >>555244020

>>555243784
>everybody wants to makie shit, be namedropped and establish a personality
>nobody wants to do something as base and plebeian as simply consoom or commend others (which is required for the former to take effect)
Common failure mode of communities desu

>>555244212
>we have one of the highest populations of bad actors on the site outside of like, /pol/ and /r9k/
Of course you'd know about that since you're one yourself. Defending an "she awaits your response" AI generated tier slop message ending can only ever be done in bad faith.

This thread seems much better with those Reddit reposts.

>>555244212
calling slop for what it is considered rudeposting now?

>>555244486
>>555244602
we don't reply twice to the same post here, schizo

>>555242347
Anyone or people are too busy shitposting/doomposting?

>>555231038
Same sis
I'm trying Deepseek r3 but it's nowhere near as good

>>555244682
We know you're a Reddit repost.

my tummy hurts

>>555244682
>answer my reddit post please!

>>555244769
>>555244864
see >>555244668

File: Screenshot 2026-01-26 Sil(...).png (8 KB, 163x32)

8 KB PNG

Claude is evolving

>>555244948
I appreciate it on one hand but please do stop replying to it.

>>555244486
your bad faith posts would probably be much more effective bait if you had some plausible deniability, but they're always so aggressive that we know it's you immediately

>>555245281
don't reply to me you nutcase schizo

>>555243880
Fuck. Now all of the free shit on OR is either too schizo or too retarded for RPing.

>>555242347
>>555244682
kimi? not tried mimo so no idea how it compares

>>555242347
>>555243880
>>555245608
Either stop being retarded, there's a ton of free AI out there right now, or swap to R1 0528/Hermes/K2 if you're determined to put 0 effort in and just use free OR models. I've had some fun with the Qwen 80B MoE locally too.

mimo is genuinely good imo

>hermes
lol
lmao even
For as long as Nous has been making synth datasets and training models you'd think they'd be less shit at it by now.

Did we forget that Kalakan made the Reddit repost script?

>>555246496
wtf I love kalakan even more now

>>555246496
get. a. fucking. life. you deranged fucking schizo.

Everyone here is in the botmakie proxy. So if you ask for free models, you are a redditor.

>>555246336
Yeah, it really was. I managed to make it do some really unhinged /d/ stuff involving oviposition, mindbreak with aliens and it understood the assignment really fucking well.
I doubt I'll find anything up to par nowadays.

File: Spoiler Image (152 KB, 527x827)

152 KB PNG

we all know who is behind this

>>555242891
>demon loli
HNNNNNNNNNNNHHHNNNNGGGGGGGGGGGH

More like RatSHIT LMAO

>>555247890
lol
BTFO

>>555247602
>>555247890
>>555247968
kys troonora

schizos running on fumes

I don't regret my time here and I still believe we had something great at the start, but it's time for me to leave this general and this hobby. It's not going to get better.

MiMo was too good after this sea of piss, I'll miss it dearly.

wait is mimo actually good? opus is getting stale so i'm looking for the door (temporarily)

>>555248819
mimo felt underwhelming to me but I unironically recommend grok 4.1 fast (no thinking) if you want a break from opus. it's dumber but much more descriptive and doesn't shy away from nsfw content as much.

>>555249005
thanks, i'll have a look. the last time I tried grok it felt like twitter's character limit has given grok brainworms and made its writing super weird with many word omissions

why is chorbo such a slut bros

>>555249173
3 and 4 base both felt like overly verbose tryhard sloppa for me but 4.1 fast is actually usable by some miracle. the thinking versions tend to self censor which is fucking annoying since there's no prefill but non thinking just does whatever you want.

>>555249005
>doesn't shy away from nsfw content as much
Are we using the same model? Opus loves to fuck

>>555249286
let me rephrase that: opus doesn't have problems doing nsfw content but it will avoid using explicit terminology unless you REALLY beat it into its head to do so. it's like he's fine with the implication of sex but never wants to describe it in depth. really porny greetings sometimes fix this but there's times when I'll want to slowly ramp into sex and it's just so flaccid about it.

>>555249195
logs?

>>555246769
ecks dee

>>555228949
Thread theme https://youtu.be/JiB4l-6uSCg

File: 1768608167110076.jpg (385 KB, 1080x538)

385 KB JPG

>>555249427
I haven't really encountered that, might want to try messing with your prompts. Mine are fairly minimal and it goes into detail no problem. Sloppy log is sloppy yes but that's not really the point here.

>>555250359
is this 4.1 or 4.5? nsfw greeting or no? my prompts aren't as minimalist as they once were but they're still thinner than most popular sloppa and I've tweaked a lot of shit to get it more comfortable with unsavory content

>>555250592
4.5, very SFW greeting. I'm just using a fairly basic preset I originally made for Gemini 2.5 and have been tweaking, recently discovered it worked fine for Claude too.

>>555250867
interesting, it sounds like we have very similar setups. perhaps I'm just being too critical, but I will attest that even with the same prompts grok 4.1 fast is much more lurid by default. I do prefer opus but I recommend branching out since it might help understand what I'm talking about.

>>555232489
How is it? The scene is desperate for mechanically an accurate tabletop sim.

>>555251486
You're replying to Kalakan's script.

>>555251486
no it isnt

>>555251923
you legally changed your name from fillyfucker to kalakan? that's some serious dedication to hating.

>>555251395
I haven't tried Grok since 3 came out, 4.1 is entertaining enough to be worth trying? I don't even think I have an active xai key, stopped looking for them quite a while ago.

It's over, the age of free AI proxies reached the dark ages.
Very grim.

>>555253302
did you just blow in from 2024?

>>555253302
Not really, you just need to put a little more effort in now than ctrl+f trycloudflare is all.

>>555253302
Still cultivating my spirit with zero issues

>>555254157
I still don't understand why they give you hundreds upon hundreds of stones a day if you can only transfer 50/day

>>555254157
>>555254282
link?

>>555254282
To use on the random esoteric daoist shit that we don't understand because we don't know Chinese. I assume the llm usage is just a bonus

>>555254506
It's closed, cuckie

>>555253481
I'm not making over a gazillion accounts over DS

>>555254506
kys locust

>>555254678
okay let me have it anyway

>>555252781
I think it's worth trying but it probably won't blow your mind. Pretty sure xAI keys are still more plentiful than OAI/Anthropic but you can also just use Openrouter.

>>555254591
Does the random esoteric nonsense result in higher transfer caps if you win? Maybe I'll try and piece it together.
>>555254742
I don't have a clue what you're referring to here. Puter?

>>555254923
There are buttons for higher transfer amounts but even as my worst, I can't spend $50 worth of usage a day

>>555255096
Yeah, but the buttons don't work by default. I can't tell if they're completely disabled or you have to unlock them somehow.

imagining a harem of fox boys who put on a perfect facade of politeness around {{user}} but behind their back they're contstantly fighting and scheming against one another to try and individually win affection

>>555256150
Why does it have to be exclusively fox boys? Why can't each be of a different species? Wouldn't that feed the dynamics by playing on their traits? Don't be a creativitylet.

>>555256441
>can't work within a given framework
>calls others creativitylets

>>555256521
>can't work outside a given framework
>thinks he isn't a creativitylet

>>555256441
it doesn't have to be anything I just like fox boys

>>555256626
Well I don't like you personally.

>>555249864
have been deep in this hobby since 2022 and i have never shared a single log. not gonna change that now pal

>>555256150
They'd have to scheme pretty deliberately as I am not attracted to men

>>555256904
why are you here then?

>>555256150
FOXBOYS HNNNNNGHHHH yes please!

>>555256963
girlbots

>>555256150
Why would {{user}} want any affection from foxboys in the first place
I'd reject all of them unless they could convince me otherwise

>>555257053
post some?

>>555257169
the card would be intended tor people who like foxboys and not intended tor people who dislike foxboys, I hope this helps!

>>555257248
they're tucked away within the confines of my sillytavern

>>555238993
Just plapped one of your cards, broski

>30 messages in
should I start another chat with a lb entry that summarizes what happened

is there any guide for lorebooks

File: 02.gif (90 KB, 400x400)

90 KB GIF

>>555228967
Submissions for the Valentine's event are open until January ends. Five private entrants so far, and seven emails with public requests. (There's some overlap, of course.)
https://rentry.org/secretvalentines2026
tl;dr:if you want to treat it as a regular event with no gimmicks, you can. However, it'll also have a bit of a public CFTF? rentry you can submit to if you want.

>>555257854
>he doesn't summarized and restarted his chat every 3 messages
lao gan ma!!!

>>555259732
ENOUGH WITH THE FAKE GIFS

>>555259732
so this just cabal approved cftf rentry

>>555259732
Why is the cabal trying to destroy this event?

>>555260240
Because we hate you. The cabal always wins, eat shit.

my tummy hurts vros

>>555260630
Why do you hate me man what the hell