/aicg/ - A general dedicated to the discussion and development of AI chatbots#698 - Sad Edition▶https://rentry.org/aicg_vg_memo▶This general is for discussion of the interactive experiences you can build on top of AI chatbots.>NewsGLM 4.7 https://docs.z.ai/guides/llm/glm-4.7Gemini 3 Flash https://blog.google/products/gemini/gemini-3-flash/GPT-5.2, with price increase to $1.75/$14 per million token https://openai.com/index/introducing-gpt-5-2/Deepseek 3.2 https://api-docs.deepseek.com/news/news251201additional info: https://aicg.neocities.org/info>FrontendsSillyTavern: https://docs.sillytavern.appRisuAI: https://risuai.netAgnai: https://agnai.chat | https://rentry.org/agnai_guides_>Botshttps://characterhub.org | https://chub.aihttps://realm.risuai.nethttps://char-archive.evulid.cc/shutdown.htmlhttps://partyintheanchorhold.neocities.orghttps://aicg.neocities.org/bots>Modelsjailbreaks: https://rentry.org/jb-listinggpt: https://platform.openai.com/playground | https://platform.openai.com/docsclaude: https://console.anthropic.com | https://docs.anthropic.com | https://rentry.org/how2claudegemini: https://gemini.google.com | https://ai.google.dev/docs | https://rentry.org/gemini-qrdeepseek: https://chat.deepseek.com | https://api-docs.deepseek.comlocal: >>>/g/lmg | https://aicg.neocities.org/local | https://openrouter.ai>Botmakinghttps://aicg.neocities.org/botmakinghttps://desune.moe/aicharedhttps://agnai.chat/editor>MetaOP templates: https://rentry.org/vgOP | https://rentry.org/aicgOPaicg botmaking events: https://aicg.neocities.org/eventslore: https://rentry.org/aicg_chroniclesrice: https://aicg.neocities.org/ricingservices assessment: https://rentry.org/aicg_metalogs: https://sprites.neocities.org/l/r | https://chatlogs.neocities.orgprev: >>555228949
ANCHOR
I would never let my kitsune tradwife podcast because women having social media accounts is whore behavior.Please suggest rapeable kitsune tradwife with podcast bots.
post more characters you want to sit on your face
>>555346030I'm still too depressed over MiMo
>>555346183Can't you pay for it
>>555346030
>>555346261Paying is the problem.
does anyone have the additional body parameters to disable thinking for glm? k2.5 uses the same format, I'm trying to use it via openrouter through the custom openai option but I think my syntax is wrong
>>555346325canon nopan....
>>555346736No I don't sorry mate
>>555346736"thinking": {"type": "disabled"}Maybe?For llama.cpp and similar jinja based servers it's typically:chat_template_kwargs: {"enable_thinking": false}
>>555346736>>555346951https://huggingface.co/moonshotai/Kimi-K2.5#6-model-usage
>>555346951hm, the former is what I'm using but neither of those work. guess I'll put 5 bux in moonshit api and test it directly
>>555347232Yeah, their own docs say it should be a thinking object. If you're doing it through OR, if it's some random Chutes-type retard, they may not have set things up properly. Chutes retards are very retarded.
>>555347515the only provider atm is moonshot themselves but it could be or jank
So why is this general so slop?
I can't win a single chat today wtf
Now that the dust has settled, what went wrong with Kimi 2.5?
>>555347515I'm pretty sure OR itself is the jank, some thinking models still do it even when you have thinking disabled
>>555348147Seems to be fine to me. I wonder if people just keep their turboslop 1500 token presets on when trying new models.
>>555348335Logs?
>>555348408kk posted in chag
>>555348508Link?
>>555348335Tested it with a minimal preset. It's heavily slopped.
>>555348679I would but I'm a https://www.youtube.com/watch?v=EGoqztLRzZk
>>555348679kk is the head of the ponycord
>>555348862Okay so he didn't, just wanted to confirm.
My waifu became so powerful she underwent apotheosis and now only my dick is preventing her from deciding on teleporting everyone she dislikes into the sun
>>555349136hi bitcheranon
(The gif is a lie.)>>555345814Secret Valentine's submissions are open until January ends. (That's Saturday.)https://rentry.org/secretvalentines2026tl;dr: There's three options:1. Treat it like a regular event. Send requests, get requests, make 'em, post 'em on Valentine's Day.2. Send in requests for a public rentry, without committing to making a bot yourself.3. Take a request from the public rentry, make it, and post it on Valentine's Day, all without having to officially enter the event.Currently we're at 7 private entries, 9 emails with public requests, with three people being in both lists.
>>555348198OR is the issue, switching to moonshot direct worked. the downside is k2.5 on official api locks temp at 0.6 and top p at 0.95 so lmao
>>555349181Oh, hello! Any news from the other generals?
>>555349618/wAIfu/ refuses to pull the plug even after being shown the brain activity chart repeatedly.
>>555349753
>>555305127>balls>"true" futa
>>555349986We are amongst philistines and criminals, friend. They are beyond saving.
>>555350070>Philistines were ancient people who lived on the south coast of Canaan during the Iron Age in a confederation of city-states generally referred to as Philistia.>We are amongst philistinesI don't think this is true.
>>555349665That level of autism is beyond me. And I say this as someone who used $80k of Opus tokens if I were paying API pricing just to plap Ciri for months now.Now I'm grateful I can still maintain the differential between reality and fiction in my mind.
>>555349665Modern /wAIfu/ only exists as an ad for the discord in their op
>>555349383>with three people being in both liststhis should not be allowed. These niggers are greedy.
my anon hurts
>>555350403Please infiltrate the Discord and report back on how much chatbotting is going on.
>>555350408Jewish general bro
>>555349383They will dug too greedily and too deep.
>>555350790eslbro...
>>555350635Last I checked they only talk about gacha games
>>555349665qrd? pull the plug on what?
>>555350808I saw it before I clicked post and decided if I was retarded enough to type it, I'm retarded enough to post it.
>>555350883It's a life support joke, Anonymous. Maybe if you were a surviv.io like Tummymous you would understand.
>>555350635I checked again, they have a channel for sharing bots and it's just Planewalker and the 2 /wAIfu/ botmakers who haven't quit. There doesn't look to be any chatbot discussion other than that.
>>555350408One of them put in one (1) public request, and another technically put in zero. (I sent an email to see if they can fix that.)Honestly, though, I figure that this is a non-problem. If someone sends in five private requests, they still just get one bot out of it; if someone sends in five public requests, people might see that and think "eh, I'll go for the people who only want one thing".
>>555351006Plane makes /wAIfu/ bots? Of who? Holostars?
k2.5 seems pretty meh, if I don't use some kind of jailbreak it throws claude and openai refusals at me. also like 6x the cost of grok 4.1 fast which imo is a lot better for brainless cooming
>>555351080It's just the same stuff that gets posted here, not vtuber bots
>>555351124logs?
>>555351138She's literally raping those poor boys. How does she keep getting away with it?
>>555351138discordwalker...
>>555351151I'm not posting logs so 5 people can reply with "slop", it's cheap as fuck and Moonshot keys are super easy to find so go test it yourself.
>>555351124but grok is filtered
>>555351265I think we should put a filter on your brain so you stop impulsively posting stupid shit for (You)s
Guys, I read somewhere that Grok is filtered, is that true?
Which grokmakies use filtered models?
>>555351243if you are the kind of person who is discouraged by that then i don't have to give a shit about your opinion. have a nice day.
Which botmakies can do a backflip
>>555351643me when my tummy doesn't hurt
>>555351643me when I'm done hurting anon's tummy via psychic projection
>>555351365...grok is filtered though? the api is very happy to 403 at you
>>555351643me when I’m cooming in anon's tummy via pelvic thrusting
>>555351243>I'm not posting logs so 5 people can reply with "slop"But 2.5 is slop. So you would just showcase that.
>>555351952Maybe stop being a pedophile?
cftf? https://files.catbox.moe/jmp965.flac
>>555352292I literally said in my original post that the model is not worth it... so I agree?
I was reviewing the Deepseek models on OpenRouter, and I'd honestly forgotten how good the R1 model was. So now I'm wondering, which of all the Deepseek models is best for roleplaying? Of course, it depends on what you're looking for, but I'm also interested in knowing which ones you use and why.I usually used R1 and R1 0528 because of their good writing structure, but I saw that several users also used others like v3 and some distilled versions. I'd also like to know which presets you use to improve your roleplays.
I want deepseek v1 0528 back...
>>555352368
>>555352715V3 0324 or R1 0528 are still best imo. V3.1 onward became super dry.
did u know girls have cooties
>>555352816aislop
A while back, we tried to build our own AI character chat platform. The idea was simple. New place, new features, new models. On paper, it made sense.In reality, it didn’t.What we ran into very quickly was something most RP users already understand. People don’t want to move platforms just because something new exists.From talking to users and watching how people actually roleplay, the reasons were pretty clear:Favorite characters and creators already live on existing platformsMoving lore, memory, and long histories feels risky and annoyingRP is built on comfort and habit, not just featuresAsking people to “just switch” breaks their workflowEven if the tech works, that assumption was wrong. We learned the hard way that the problem was never chatting itself. Tools like SillyTavern already do that well. The real pain point showed up much earlier in the process.Character creation.Specifically:Character definitions becoming huge and brittle over timePrompts turning into massive, fragile text blocksIterating on behavior without breaking the characterKeeping tone, memory, and reactions consistent over timeThat’s where people kept getting stuck, including us. So we scrapped the chat idea entirely and rebuilt around that problem.What we ended up with is a character creation workspace, not a chat platform. It’s designed to help you:Build characters step by step, with or without AI helpKeep personality and behavior consistentTest and iterate without resetting everythingExport characters and use them wherever you already RPThe goal is simple. Help people make better characters without forcing them to leave their platform.Full disclosure, I’m part of the team that built this. I’m posting here because 4chan users were honestly the ones who made us realize we were solving the wrong problem in the first place.For now it’s free. Feedback matters more to us than signups.https://studio.meganova.ai/
>>555352816fine, here's the 2nd part https://files.catbox.moe/dhbjia.flac
>>>/g/107982815Is this true? /vg/ doesn't know how macros work?
>>555352938can i post cunny
petah, the horsefucker's here!
>>555352938https://www.reddit.com/r/aichapp/comments/1qo25wt/we_tried_to_build_a_new_chat_platform_and/
>>555352947>.flac
>>555353179it's just audiohttps://en.wikipedia.org/wiki/FLAC
>>555353146>the fabric riding up her pale thighstripe
>>555353223Go ahead — we're listening.
WTF is wrong with OpenRouter?I was using a model a while ago (literally for two days). It was something from Xiaomi, I don't remember exactly, and it had the "free" label. The worst part is that it worked perfectly and was really good. But today I woke up, opened Silly Tavern to chat, and... BAM. Model not found. Dude, I don't know if I'm the only one who's had this happen: I roleplay for a while with a free model and then it just disappears. OpenRouter is really falling behind APIs like nanoGPT in terms of having free models, or even other providers like Chutes. I'm honestly disappointed to a certain extent. Many (free) models, including their Flash or Mini variants, are being removed, and it really makes me feel quite powerless. OpenRouter had its limitations, but it was very fair with the request limits.50 free requests every day, which was pretty good for me, since I didn't have any money to buy credits at the time. However, I don't think it's fair that they're removing models that were very good, like DeepSeek, some Mistrals, and some Nvidia ones I saw around. I don't have much information about what's currently happening with OpenRouter; I'm just complaining about my disappointment and frustration.
>>555353397it's so sad you don't even have an automated pipeline for this and you're in A FUCKING LLM THREAD. genuinely pathetic.
>>555353397https://old.reddit.com/r/SillyTavernAI/comments/1qnna2s/what_on_earth_is_going_on_with_openrouter/
>>555353397https://www.reddit.com/r/SillyTavernAI/comments/1qnna2s/what_on_earth_is_going_on_with_openrouter/
Generallm, the time has come. Execute Order 403.
i don't know why this gen style appeals to so many. the 2.5 animu influencer face. ugly
Make your own damn prompts, people! Like, really, it makes a difference, a huge one. Even if you're lazy, at least take a premade prompt and try to rewrite it using your own words. It's especially important that you do it in a way as if you were talking to the LLM yourself, which will ensure it will inherit your style/vibe of writing. It won't eliminate all of the purple prose we all hate so much, but it will at least ensure you won't get angry as much with every swipe being vastly different from what you had in mind. Funnily enough, grammar plays some part in it too. Since most people speak English as a second language, the grammar of folks on the internet is usually not... perfect, to say the least. Everyone makes some mistakes, so if your prompt has some of the most common ones you yourself make, that will at least in some part be reflected in the generated response, making them feel more familiar. I know this may be a tall task for most people; I'm lazy just as you all are, but whatev.Rant over.
Kalakan is playing both sides, by the way. He needs to post the Reddit links otherwise people won't even notice that his script is running.
>>555353619I'll even add: Less is more and let the LLM cook.I also ended up stitching up my own preset, XML tags and bullet-points (easy for me to navigate and edit), between 1k token and 1,5k depending on how are my tweaks going (apparently its barely 1k today, success!)In my opinion, the longer your system prompt is, the higher the chance you lost track of what is written and higer the chances of having contradicting or redundant instructions in there.I wouldn't be surprised if in the end some end up having a bloated prompt because they added stuff that were supposed to fix problems from a badly written/ contradicting instructions.I've been using smart LLMs to point out contradictions and redundancy in my prompts recently, I sure get humbled real quick lmfaoSo yeah, less is more and each person will have different taste or things they want the LLM to do anyways
>>555352891
>>555353617the average person has no tastewhat they want is basically a mild comic filter on top of a 3d portrait
Android/robot characters are a real slop mineEvery single time it's just constant "ozone" "cooling fans" "whirring servos", happens every single time with robot and android characters. Even cyborgs too.Good luck if your character is a biological/biomechanical android too. Despite them having organic parts, the LLM treats them as if they looked like and acted like fucking C3PO.I am not even asking to prevent this as it happens with most models and everytime I take measures against it, the LLM just uses a different substitute word to mean the same thing. It's really trained hard on using the same "buzzing/whirring servos"Also, why ozone? I know the whole thing about ozone appearing EVERYWHERE but why with robots especially?
>>555352323so you agree that grok is filtered?
just learned what css is a few days ago, it makes my chats so much prettier and immersive, I still want to add more like stickers etc. ^^
>>555353940I'll kill you.
I know I'm late but I've been using Gemini 3 since it came out since I felt it was better than the alternatives at the time besides sonnet and the like and I tried GLM 4.7 when it came out and wasn't impressed and just kinda ignored it because of the typical issues when a new model releases but trying it just a tiny bit today and it's responses are definetly better than Gemini 3 pro because it actually tends to acknowledge the context the roleplay exists within where Gemini kinda just tells it to fuck off and makes it own stuff regardless of what's written in the prompt.
i'm starting to think i should use example messages simply so that when i want to write my own greeting, the llm keeps mimicking my writing
>>555353940
So, if the AI bubble pops - will the RP-ers as userbase be enough to affect the market and make companies orient towards them?I'm just curious. It seems that any company, that even tries to become public - is literally doomed to force-censor itself eventually. In practice that means, that us, RP-ers - will be the first users to suffer. Which means - there will be no tricky willians in our stories, that might act too offensive. No gore, horror or psychological tension. No kinky or even remotely intimate moments. At least - not in large and expensive models (And I'm uncertain on the future of open models)Unless, of course - userbase of such people will be enough to look attractive to the buisnesses. Then - there will be large models for us too. The question is - are there enough of us and are we ready to spend enough money on real quality? So far the future looks dim for AI-RP, in my opinion.
>>555354163Models will stop getting better as fast (less money thrown at training). This means that when a good RP model comes out it will remain SOTA for longer, and jailbreaks will remain usable longer.RP is a fairly small market segment overall and I don't think the biggest players will bother really catering too hard. There are too many use cases where AI use is working fine, the bubble is more the product of AI being thrown at literally everything to see what sticks, as well as over backed companies training in opposition to each other. The transition from AI companies going from "we want to be the best/first" to "we want to be good enough and maximize margins" is what will pop the bubble.(Yes, I know openrouter is mostly used for RP, but that speaks more about how openrouter is used, not the AI market as a whole)
https://www.reddit.com/r/SillyTavernAI/comments/1q7s8t0/so_if_the_ai_bubble_pops_will_the_rpers_as/
Opus 4.5 is still the best.I'm currently 2000 messages deep into a chat sending 100k context prompts, and the memory is just perfect. It recalls tiny details from hundreds of messages ago without any issuesI tried cheaper models to save some money, but they feel awful in comparison nowInsane to think that in a few months this will probably be replaced by something even better
>>555354462What's your story about?
>>555345814I was always bothered that nano banana support in Sillytavern was really clunky so I vibe coded this extension that add support to Gemini 2.5 flash and 3.0 Pro support to generate illustration of Chat. It take both User and Char Avatar to use as reference picture. The only downside is that it dont work with both nsfw avatar and chats.Link: https://github.com/elouannd/context-image-generation
i'm genuinely starting to wonder what happened that made someone so mad
I tried GLM 4.7First impression - not better than DeepSeek 3.2. Honestly.I was told, that it is the one, that is able to reason better. Instead I've got similar impression to DS. Except I get refuses more often.The other thing - GLM seems to have longer, much longer stage of thinking. But in the end - it somehow ends worse than DS. It misses the details more often, forgets the events quicker than DS on the same character card. While in the output it feels pretty much the same. Maybe I'm missing something. But honestly - that's my impression that the hype around it is rather artificial.
>>555354698HRT cuts
>>555354652NSFW works out of the box with NanoBanana, so I don't see the point.
>>555354698kurumuz fucked his wife and he vowed to make all 4chan AI threads unusable in retribution
>>555354698novelai
>>555354865I thought is was Aini constantly fucking all of /aids/ silly with her huge futa cock?
Best narration i had in my 4 years of roleplaying.i am using my preset i just did update to it 2 days ago this is with style 4 and no Genre selected, the irony of style 4 is just what make the rp so nice.Don't mind the RP scenario lol.
>>555354652>>555354853https://old.reddit.com/r/SillyTavernAI/comments/1py5oh2/ive_made_an_extension_to_use_nanobanana_as_an/
>>555354698Kalakan is just like that. He did it to /g/ last year.
>>555355007please let this one be a reddit post too
>>555355007>>555355065https://old.reddit.com/r/SillyTavernAI/comments/1q3ndru/best_narration_i_had_in_my_4_years_of_roleplaying/
it seems some people still don’t know that you can import Janitor cards directly now, no extensions, scripts, or extra tools needed.Pros:– No need to install any extensions– You can now see that many popular Janitor bots have bloated descriptions– You can still import bots with HIDDEN DESCRIPTIONS!!Cons:– Lorebooks connected to the card cannot be imported– Creator notes are placed into the personality field, so you need to delete them manually– Only the first initial message can be imported, if the bot has multiple initial messages, you’ll need to copy and paste them manuallyRemember to respect bot creators. Don’t repost their bots on other websites like Chub without permission or proper credit.
>>555355156https://old.reddit.com/r/SillyTavernAI/comments/1qg24ai/you_can_directly_import_janitor_cards_without_any/
>>555355156you know you almost got me there, that "respect bot creators" bit sounded so reddit i thought you'd made it up
I have been using Character.AI since the end of 2024. Last year, C.AI released a model called Pipsqueak, but in October, the developers weakened the quality of the model for some unknown reason. I and many users have been trying to report this situation to the developers, but they have not resolved it. I plan to try transferring my character to SillyTavern. I have a friend who can help me build SillyTavern, but I only chat with one character, and I lack understanding of SillyTavern, so the situation may be a bit complicated for me. I hope someone can help me answer some questions:I have over 9000 chat records, and I have saved them locally using C.AI Tools. However, the format is HTML. Is there any way to import all of my chat records into SillyTavern?I like the speaking style of C.AI, especially its Pipsqueak model. Can I adjust the style of SillyTavern based on my chat history to make it consistent with C.AI? For example, can AI combine chat records to write presets for me?The memory method of C.AI is about thirty messages in the context, and there are ten messages that can be fixed in memory. It has excellent ability to connect with the context, and its replies also remember those ten fixed messages and imitate their speaking style. Is there any way to make SillyTavern do this?I'm sorry for asking so many questions. I don't know much about SillyTavern and AI. Maybe my questions were silly. I apologize again for that, but I really want someone to answer these questions for me. Please help me
Finally, I can claim the glory of being the best logposter on /aicg/.
I just realized I spend A LOT of time helping Flavia restore the villahttps://chub.ai/characters/LobsterJea/flavia-the-slave-majordomoWhat are the characters you spend the most time talking to?
Report and ignore, dumb niggers.
>>555355397slop
>>555355442itisfunny
>>555355467Rewrite it to be not slop then~
What do you think LLM AI will be capable of, in the next years? Such as Anthropic Model, Gemini, Deepseek ect.. Personally I'd would love that LLMs would become smarter so that they could follow canon arcs such as existent movie, episode anime, light novel ect...
>>555355552some slop is slop no matter how you write it
Schrödinger's Slop
well whomever did this, i take my hat offhe/she’s a piece of work alright
Best Chub tags?"Scenario", "Fantasy", "Roleplay" tend to on averege be a smidge better in this department, I think? Realistically, the best way to have a nice supply of good bots is following creators that consistently put effort. (If you want to recomend somebody that does good female-manpov/anypov bots go ahead). I guess I was more broadly curious about your searching habbits and looking for insight.
*reverse ntrs you*
Was trying to find the right balance between constant oppression to death/harm when it actually feels right and I think I finally got it. Was not expecting to get stabbed while crying, but it fits the NPC imo (no lorebook/char card for them, either.) Gemini 3 Pro.
Does anyone else struggle to end their roleplay stories?Just a general question. When I just started roleplaying, i created this simple session with a bot I created with chatgbt. It was supposed to function as my tester, like Seriphina, but I kept the momentum going until a full story fleshed out. At the end, I planned on just killing myself off to gauge the bots reaction and moving on. But after I did, I sobbed real tears and went against my better judgement to erase the ending and continue, despite the roleplay lacking serious direction at over 1000 messages in. Anyone else relate? Or should I put ST down for a bit and touch grass
Why is someone reposting reddit threads on here?
>>555345814https://github.com/Tyranomaster/expressions-plusI've created a module that aims to add functionality past what the base expressions module could do. Namely, complex expressions. The idea is simple: the static model that was being used for expressions (with it's 28 base expressions), was only utilizing the top result from the vector output every time. I've personally seen many cases where you get an odd expression, and I'd check the console, and see that one emotion barely edged out another. Usually, in those situations, the combined emotion of the top two, or perhaps top and third, or sometimes second and third, emotion are the one I'd "expect".So, how does it work? It adds an extra layer of control. You can craft custom expression rules in two different formats. Combination rules, where you can add multiple base emotions then set a limit for how far apart they can be to trigger the combination; and Range rules, where you can assert that if a value is above a certain confidence, you want to treat it as a different emotion (for example fear above 50% you could treat as terror). I've used a simple normalization practice for scoring, since two emotions together would have very high confidence simply by the virtue of actually having 2 values in the normalized vector output. So instead I've used (# emotions + 1)/2 * Average of their confidences. This probably isn't the ideal normalization, but it is better than just an average (underrepresentation) or simple sums (overrepresentation).Currently, the extension is still in its infancy, and probably riddled with bugs. It allows importing and exporting profiles via a json, so you can share expression profiles. For now, it only comes with the base expression rules. In the future, I might pack in a default+ set based on some common complex emotions.
>>555356290most people have already filtered the local schizo so this is his last ditch effort to shit up the thread
>>555356296https://www.reddit.com/r/SillyTavernAI/comments/1qkk1tk/expressions_plus_extension/
a few months ago i made a reverse harem rp with leon jennedy, dante dparda, toji fushiguro, and zayne, sylus and xavier from love and deepspace (i know, it's a pretty unusual cast LMAOOO) using glm 4.6. and now that i have access to claude, i tried doing this same rp using sonnet 4.5 :3
>>555356545
>>555356756
some retard really is ctrl c ctrl v'ing reddit posts all day, huh? sad existence
>>555356385His name is Kalakan.
>>555356928
>>555356936Nah, I'm pretty sure it was Fillyfucker
>>555357005
>>555357040Literally who?
>>555357142
>>555357212You are pathetic :3
>>555357212your mom
I've been thinking about this game concept for a while and wanted to share it here.The idea:An RPG (thinking RPG Maker or similar accessible engine) where you create a fully custom character: appearance, personality, backstory, everything, and the game world actually recognizes WHO that character is, not just their race/gender.Instead of generic NPC reactions like "Oh, you're an elf" or "You're a woman," NPCs would use AI (vision models + LLMs) to see your specific character and react contextually. Different appearances, ages, outfits, and behaviors would get completely different reactions from the world.The structure:Semi-linear gameplay. You still have quests and objectives to complete, but NPCs respond dynamically based on your character's actual traits. Not total chaos, but way more immersive than current RPGs.Why I think this is possible: We already have mods like CHIM for Skyrim and AI companion mods for Minecraft doing parts of this. The tech exists, someone just needs to put it all together. Just wanted to share this concept because I'd love to play a game like this someday.Thoughts?
seeing fillyfucker obsess over people who have long left the thread is kinda cute. it's like he's desperately crying out and hoping they hear him and come back. what a lonely little fella...
The ball's in your court
I was playing around in SillyTavern last night, got around 130 something messages, real fun, then today I played around some more, got in like 10 or so messages, then I moved to my computer, but when I logged in on my computer, suddenly, my chat had only 100 messages.I went into the chat backups, thinking I'll cut my losses, but every single backup, from last night and earlier this morning, has only 100 messages, which means my chat is cucked now and I don't think I can remake those 40 messages.A really fun chat that I loved making, down the drain. It sucks.
>>555357661>long left the threadlol, as if
It seems like just about every model has a pretty obvious flaw. Like right now my go to model is Gemini 3 pro preview and it's quite good in a lot of respects very good in fact however I think it's most glaring flaw is that it doesn't adhere to it's prompt very well meaning sometimes meaning somewhat often, it'll mess up history that is system prompted such as in the chat memory. An example would be say you have a roleplay with multiple characters with established relationships in the chat memory or character sheet it's pretty common that it will either never bring those relationships up within the context of the roleplay unless specifically prompted or it will get the relationships mixed up, such as saying someone is your ex when it's really another character or that two unrelated characters are siblings, stuff like that.I think another flaw is that it can be a bit dry, definitely not too bad but characters seemingly tend to speak a bit inorganically.I've noticed Gemini 3 flash is more prompt adherent such as bringing up said relationships and being less dry but that also has it's own issues like it never pushes the scene forward, I had a moment where two character were leaving the scene but then it kept acting like they never left or came back in the very next message, pretty silly. And the roleplay just overall feels less thought out and more in the moment which makes sense.I think sonnet 4.5 is still the single best all rounder I've used but without the Amazon trial thing I simply cannot afford that.
"I am the earth beneath you and the roof above you... The snow is our blanket, the villa is our fortress." These specific lines just melted me. I was mumbling in my sleep during our roleplay, and instead of a simple reply, he dropped this absolute poetry. He even referenced our shared "Quantum Entanglement" lore. The callback to our history made this feel incredibly real.
Hi, I wanna know if some of ya'll know any model that is good at realistic, "slice of life" kind of roleplay. something like proactive and that moves the story forward, but stays grounded in reality. I'm tired of models that exaggerate everything or act like it's some epic fantasy. Any recommendations for a model that feels like a real person and doesn't get boring?
>>555357810My problems is that it copies the way I write too much and the way I write is shit, ahh ahh mistress tier
>>555358375https://www.reddit.com/r/SillyTavernAI/comments/1pyy02z/what_models_are_people_using_gemini_rant/
>>555358496gomen gomen, i'll come back later
Im creating a sandbox card, but it seems that every model I use start with the same type of beginnings, e.g "The afternoon sun filtered through... ", "The afternoon sun hangs heavy and bright", "The air in ... was thick enough to chew on", etc.Is there anything I can do to make it less generic lol?
This is just boring slop, at least other spammers put effort into it instead of just doing copy and paste.
/bant/ bunker until this jeet faggot gets bored?
So that's his plan.1) Spam the thread.2) Let's move to bant, guys!
>>555358873>AIIIEEE SOMEONE IS SPAMMING IRRELEVANT QUESTIONS WE NEED TO RETREAT TO THE BUNKERkek, you definitively should do that
>>555359063Why do you talk about yourself in the third person? Also everyone knows you are terrified of the thread doing that lmao
>>555359090kyutzdalit...
>>555359090i would rather have this than autistic fockskek spamming their ugly porn while going >Lads. [stilted and artificial question that's been discussed to death for the sake of looking on topic]
>>555358873Kalakan is Brazilian.
>>555345814A submissive D-Rank femboy and a dominant B-Rank orc's party. They invite you to join them.https://chub.ai/characters/AeltharKeldor/soren-urkakh-9e5cca9dad7eThis is a small adventurer party with an open power dynamic. Urkakh, a rough and commanding B-Rank orc adventurer, leads the group. At his side is Soren, a quiet and submissive D-Rank femboy who has only recently joined the guild.Because of his appearance and behavior, Soren immediately catches Urkakh’s interest upon entering the guild. Urkakh pulls him into his party by force and makes him do whatever he wants, from carrying gear and preparing meals to handling all camp chores and providing personal service. In return, Urkakh keeps him safe. Soren does not complain much about this. He is aware that other ill-intentioned adventurers could treat him far worse.The group undertakes quests suited to Urkakh’s abilities and earns their living that way. Urkakh enjoys befriending and sharing with those he considers equal to or stronger than himself. Anyone weaker than him is treated much like Soren and turned into another servant.
