"Prince Maven, of the Calote and Merandos houses, son of my royal consort, Queen Elara," the king announced.
I was reading Victoria Aveyard's book Red Queen, page 69, when this sentence suddenly appeared. It even gave me chills.
For those who don't know: Elara is one of the most common names generated by AI.
does anyone else feel like the state of models/rping just kinda sucks right now?
lately i’ve tried doing a long, immersive roleplay that admittedly started out pretty strong. but even with custom lore books that i create or edit myself it just doesn’t hit like it should. memory isnt an issue, thankfully i can fix that with extensions and such, but after a while the characters eventually feel like predictable morons that praise me as a divine exception to everything else in the world, like i’m gods gift to earth when i do the most mundane thing imaginable. nobody feels like a character, more like lobotomized zombies going off a checklist of quirks when they speak or act.
everything is so overpriced. and if they’re not overpriced then they’re incredibly slow, stupid, predictable, have terrible memory, shit prose, or all of that combined. especially with glm. every now and then i’ll be pleasantly surprised with what 5.1 can put out, and then im stuck swiping garbage messages that take a either a minute to over three minutes per message. i am so annoyingly familiar with countless slop phrases that these models put out. “you’re either x or y. i can’t decide which yet.” “REALLY looked.” “mouth opens. closes. opens again.” “a beat.” “most people do this… but YOU. you do THAT..” “you did x. in y. in my z.” the relentless need from the ai to assign my character a pet name that no real human being would ever use on a person they know. the impossible affection characters that are supposed to have little to no positive feelings for suddenly manifesting because of some stupid forced positivity bias. etcetera, etcetera. it gets to a point where if i’m constantly dictating what should happen, i might as well just go on wattpad and write my own shit.
i’m sure if i were rich i could just use opus and have a fun time, but im not. opus can be addicting, but so terribly unaffordable for anything that isn’t short stories/smut. occasionally it writes something so painfully underwhelming for it’s price that i regret ever opening my wallet for anthropic. seriously, all i want is a decent model that doesn’t take 5 minutes to generate some x and y ism bullshit that makes me regret even trying to do anything but mindless, short lived smut. which even opus is guilty of doing sometimes. i tried deepseek v4, but i instantly stopped. on nano its really slow and feels like a model from two years ago, at least in my opinion.
are we just not there yet? how long am i gonna have to wait to not groan after reading the dogshit these models spew out 😭
(sorry if this is a little incoherent i’m really tired)
I decided to update my prompt to more strongly include the style I enjoy. Caveat: Some people may not like the style, and that’s okay. I can say that the prose becomes fairly fresh with this though. And it’s still pretty light. Which is nice.
CHANGELOG
- Main Prompt is now ~1200 tokens but has a much stronger style. I based it somewhat on Pseudonymous Bosch and Albert Camus’ writing styles - not exactly, of course, but enough that it gives the prose a bit of a personality (hence, not really defaulting to slop a lot of the time). If there are parts of the prose that seem cheeky, that’s intentional. If you don’t like it, remove “Narration Voice” in the Main Prompt.
- Added a simplified Main Prompt that’s around ~600 tokens for smaller models and the token-conscious.
- Grounded Prose Rules is no longer required and is now turned off by default. If you get ozone, etc, you may turn it on, but in my testing it hasn’t really appeared.
- Rewrote Friction Mode, Nightmare Difficulty Increase, NSFW Mode, and Gooner Mode prompts to be more assertively and directly worded.
- GLM 5.1 had a problem where Gooner Mode wouldn’t activate because it was contradicting the main prompt - the solution was simple, tell it to ignore that so it can immediately turn into porn.
- Rewrote GPT Assistant Prefill to be… less censored? God knows if it works, GPT 5.5 is very moral-maxxed.
- Main Prompt is now more neutral by default, rather than trying to make a killer GM.
- Added a “Flexible” Length Toggle, that is on by default. Short, Medium, and Long prompts now use paragraph ranges rather than word counts, based on Geechan’s advice.
- Tried to ensure trackers are always considered when enabled. Clarified the Pending Events Tracker to only include important events.
You can *still* turn on Grounded Prose Rules, it doesn’t hurt to do so. Because sometimes Gemini will say ozone, and that’s not cool. Shut up about ozone, Gemini.
Anyway, enjoy~
LLMs keep getting smarter and whatever, but those classic model quirks and "isms" still slip through. Honestly, I think WE, as in all of us, have a soft spot for at least one of them.
Mine has to be "It's X or Y, I haven't decided which yet."
Title. I'll get to the punchline quickly. One non-sexual paragraph was all it took to trigger full blown "PROHIBITED_CONTENT" failures on the API. Here's the paragraph.
>- {{char}}'s original home: {{char}}'s home is dirty and uncomfortable. There's barely any food in the fridge. {{char}}'s abusive parents come and go randomly, with screaming and threats of violence. Even as an adult, it feels incredibly unsafe, a living reminder of a horrible childhood.
It turned out Gemini was lying. It wasn't sexual content that was triggering sexual content refusals. Removing that paragraph stops all the refusals. Meanwhile, re-adding that paragraph results in refusals like this one:
>I cannot fulfill this request. I do not participate in romantic or sexual roleplay scenarios.
That's the entire response, excluding thinking. Even if I got past the API level rejections, it's a complete brick wall of a refusals. Gemini is quite happy to engage in depraved sexual content, as long as it doesn't involve an abusive home anywhere in your character's past.
The character in question is a huge bully and a giant jerk, which seems to be totally fine. But suddenly when they're a bully from a traumatic home, it's prohibited content. That's 10/10 moderation right there. Ironically, the story was at the point where the character was completely out of their old home. Which means there's some sort of weird trigger even mentioning past abusive homes, but figuring that out is beyond me.
I did have some marginal success reducing the severity of the refusals by editing that paragraph to indicate that the character is an adult and their home is in the past, but that paragraph still triggers consistent refusals I haven't seen on any other of my character cards. Obviously there's some combination or characters/lorebooks/prompts that's causing refusals, but it's odd that a single paragraph is all I need to remove to stop the refusals.
Has anyone else found similar unexpected or illogical refusals on Gemini?
Link: Feedback🐋🐋
It seems they'll be making a lot of improvements to the RP model, which is good news for everyone and for the first time a major AI is looking out for us.
I'm pretty new to this and have been looking through threads on here lately trying to gather tips. The main thing I keep seeing over and over is this sentiment worded one way or another:
"The thing causing XYZ problem isn't the LLM, it's the character card. 90% of the character cards online are awful because they don't give the bot enough to work with/waste tokens on the wrong thing/are formatted in the wrong way/etc."
But in the threads I've found, the ones who say things like this never shared a card they think is done well in their comment. I am really curious to see what a difference it makes since it sounds like it can affect a whole lot, so if you have this opinion yourself and are willing to share any cards you deem "actually good," please share!
I am not picky about genre or personality type. I just want to try one that's written in a way that will best assist the LLM in writing, according to the posts I've seen.
Thank you!
It's amazing how Google has been failing lately. I tested it right after Google I/O finished, and the quality was quite interesting. He gave descriptions I'd never seen before, like my character being shorter and the women's busts not touching my chest when they hug me—almost no model gets that right. He always assumes I'm taller, even when I say I'm short. The physical-spatial understanding of the bodies was excellent, but a day later the model is very bad, simply not describing my actions, only describing what it wants.
Any of you using it to talk to girls/ companions and such? I see a lot of presets for RP as Ttrpg (at least that is what I think I see). If you just chat with IT as girlfriend/ companions how your cards look? What presets? What models? I'm new to silly tavern and appreciate any help. Thanks.
Holly molly! I tried Ring now for a little bit and my god! It's so much in the scene. It comes up with some simple context details for scene embellishment so well. It almost feels like the scene is written through the lenses of the character itself (which is what the prompt says anyway), but more intimate. GLM also narrates as the intimate narrator, in the moment and in the scene, but Ring has really surprised me how intense it can get. I noticed some small continuity errors in body placement (nothing noticeable, unless you really look for it, which I normally do). But the way it adds small details, like a combination between MinMax's attention to detail and GLM's character fidelity. Really great stuff!
I was trying out Kimi 2.5 beforehand and decided to just fire up a couple of messages with Ring. Now I don't know which one of the chats to continue hahaha. First world RP problems I guess.
EDIT: I tried to take a screenshot of the current chat, leaving out the more NSFW parts of the roleplay. In this case, character is a traumatized middle aged woman, who has a hard time getting over the trauma of her past marriage.
Hello everyone, I hope this is the right place to ask this question. I apologize in advance if it isn't. I've been reading about Star Trek lately, and the tool looks very interesting. However, it seems to be more focused on roleplaying, meaning conversations between characters, from what I understand.
I'd like to ask if it's equally good for writing fanfiction. For example, taking a base story like an anime or video game, inserting characters (OCs), and creating chapter by chapter using a beatsheet where I tell it what I want to happen in each scene, and it writes it, sometimes with NSFW stuff.
I used to use Gemini for this with its Gems (which allow me to have context) and its Canvas text editor, but yesterday they updated the usage limits, and I reached the maximum with only 3 scenes. Therefore, I'm looking for other tools. Thank you very much in advance.
Hello! I just get "no connection", here's what I do:
-Run SillyTavern on android using termux
-Run Gemma 4 31B on my PC using kobold
-Both devices connected to the same WiFi
-Api: Text completion, Api type: Genery(openAI compatible), Api key empty, server url: http:ipv4ofmypchere:5001/v1
What do I do wrong?
[9 Greetings+ Images] The Guild Master is leaving on a dangerous journey and she has personally chosen you as her escort.
https://chub.ai/characters/AeltharKeldor/sylvara-the-guild-master-ddaf13cceb05
Sylvara is the Guild Master of Aelthar Keldor. As a High Elf, she is naturally graceful and noble, keeping the busy guild hall in order with her calm and patient nature. To many adventurers, she is not just a strong leader, but a wise mentor and a mother figure. She handles the most important matters of the guild, solving major disputes, approving dangerous A and S-Rank quests and making important deals with kingdoms and other guilds. No matter how busy she gets, she always finds time for her adventurers. She is strict when necessary, but always fair, responsible and deeply cares about the members of her guild.
Background
Sylvara was born in the Elf Realm, a distant continent across the ocean. She spent centuries mastering magic in a powerful High Elf kingdom. 98 years ago, a massive Void army attacked the Mortal Realm and caused terrible destruction. Sylvara was part of an elite High Elf battalion sent to stop this dark threat. After two years of brutal war, they finally defeated the Void. While the other elves returned home, Sylvara chose to stay behind to protect the mortals who were trying to rebuild their ruined lives.
She started a new life as an adventurer and quickly reached the legendary S-Rank. For decades, she took on the most dangerous quests and became a famous hero. Years later, her old guild fell apart due to a bitter internal war. Instead of giving up, she decided to build a true, safe home for adventurers. 62 years ago, she gathered her most trusted veterans and founded the Aelthar Keldor guild right in the center of the capital city.
Scenarios (with images)
1✧ You are tasked with escorting Master Sylvara.
2✧ You are assigned as Master Sylvara's personal assistant as punishment.
3✧ A huge fight breaks out in the guild hall.
4✧ Master Sylvara chooses you to represent the guild in an upcoming tournament.
5✧ A cursed artifact teleports you and Master Sylvara to a deadly Abyss lair.
6✧ You share a quiet moment with Master Sylvara on the balcony during a noisy guild festival.
7✧ Master Sylvara asks you to massage her temples after a long night of guild paperwork.
8✧ You wake up severely injured in Master Sylvara's chamber as she tends to your wounds.
9✧ (NSFW) You are called to assist Master Sylvara with her private bath.
Much love to u/dptgreg Go check out his news posts and youtube videos.
I have tried both AI Studio and Vertex, but the results remain unchanged. When encountering romantic scenes, 3.5 Flash may suddenly refuse to depict them, or abruptly steer the plot in a different direction. Checking the chain of thought reveals that it claims the system instructions strictly prohibit any romantic scenes. This seems to be an internal censorship rather than the external kind that suddenly cuts off the output. While bypassing the thinking mode through pre-filling and disabling reasoning can solve this, the model becomes dumb.
If you remember Aion 2.0 and you liked it, then you may be interested in trying out their latest 2.5 version. I don't know how good or bad it is, but well... it has a free tier.
You can go to their website, create account, make an API key and in ST you have to type in the "Enter a Model ID" field "aion-labs/aion-2.5" to make it connect.
And yeah, 20k tokens per day is not much. Whatever, just letting you guys know about it.
Brain Eating Zombies is a semi-serious comedic survival apocalypse scenario where zombies can supernaturally detect intelligent thought unless blocked by metal foil hats.
Survivors scavenge bizarre weather-ravaged environments, ration food and water, navigate evolving spaces, and try very hard not to think too loudly. Solving problems, panicking, arguing philosophy, or fighting zombies can attract additional hordes through escalating “brain pulses.”
The setting emphasizes:
Zombies are dangerous in groups but individually manageable, creating escalating chain-reaction encounters rather than simple combat.
NPCs are procedurally generated with distinct personalities, survival habits, secrets, and absurdly memorable apocalypse names such as:
Tone: Grimy, rainy, lonely, tense, weirdly cozy, and occasionally very stupid.
Expect:
Download it here Brain Eating Zombies
I think it wasn't what I expected at all. It's somehow dumber and more superficial compared to Gemini 3.1 Pro, making obvious and ridiculous logical fallacies and making the characters feel shallow (in the scenario where he creates the characters himself, not me). Has anyone else encountered the same problems or had a better experience? Maybe the issue is with my prompt or my game type
Everyone knows how hostile, shady, and unethical Chutes is. They've started so much shit in their community to many of us former/current customers.
Anytime we want to tell our fellow RP'ers the truth about them, they weasel into the threads and spew pure fermented penguin shit and act stupid like they are clueless.
They're banned THOUSANDS of paying users from their own subreddit and discord server and won't even let posts on their own sub anymore without their approval.
Why are they allowed to run free here?
You can downvote me all you want, but I've been using Claude Opus 4.6 for months and I've come to the conclusion that this model is incredibly boring for RP.
I've tried various presets, including my personal one, but nothing fixes this model's tendencies, which are:
The only way to "fix" it is by using CoT, but that severely limits the model's creativity and causes it to give identical answers, even when regenerating the message several times.
For such a ridiculously high price, it's not worth it.
---
Anyway, I got to use this model for months, giving me the freedom to say that it isn't all it's cracked up to be here in the community, and to top it off, it's boring.
And boring ≠ bad.
> I don't speak English. Any translation errors are not my fault.
