NSFW AI chatbots

[desmosome]

Tbh, we are so close. Gooning with Gemini (directly or using it as proxy on JAI/Chub) is some quality smut that you can direct exactly as you desire. It's not even comparable to the 8k context window crap and retarded LLMs that was the early days of NSFW chat bots.

The only thing we really need right now is image integration. I know various diffusion models have come a long way, and I'm sure some mad lads using SillyTavern to maximum effect probably already can illustrate their gooning chats pretty well. But for ease of use, a truly multimodal input/output model is what we are waiting for.

Something that can write your long form slow-burn corruption arcs, and then do individual voice acting for each character, and then illustrate the scenes with consistent style and characters whenever you ask. Like google basically has all the parts, the LLM, the voice model, NanoBanana.

I'm pretty sure this kind of generalized approach is what all companies are aiming for, as it's the path towards "AGI". But man, western companies won't ever loosen the Christian filters on image/video generation. Gotta hope China can step up.

 

[Pretentious Goblin]

Tbh, we are so close. Gooning with Gemini (directly or using it as proxy on JAI/Chub) is some quality smut that you can direct exactly as you desire. It's not even comparable to the 8k context window crap and retarded LLMs that was the early days of NSFW chat bots.

So is Gemini good for that? I've been eyeing that context window with envy, but ChatGPT has kept me by playing ball with my weird requests. Plus it's Google, so I don't know about privacy. Does Gemini have the same issue where chats get super-laggy if they get long?

 

[SunshadeAlpha]

Wonder if anyone has any suggestions regarding TTS or Image Generation than can be used with SillyTavern that are NOT run locally and arent expensive af like ElevenLabs? Im mainly using it on my phone via Termux, so its not really possible to run things locally.

Im fairly new to using SillyTavern tbf, only really started because I noticed SpicyChat was becoming way tamer and way more forgetful. My setup is focused more on using an extension to either guide a story solely from the character card settings and me just directing how things go, or to type a basic outline of my reply and have it generate a proper response for me which i can edit if needed.

If anyone wants to copy my setup, its fairly simple.

1. Get SillyTavern up and running either on PC or on Android via Termux.
2. Get an OpenRouter account and add $10 worth of credits. This gives you a really high daily free limit for chats with Deepseek without actually using any the credits themselves. If you dont want to add any credits, you can still do this but the amount of messages will probs be smaller.
3. Connect your OpenRouter API and use the "Deepseek v3.1 (Free)" model.
4. Make sure the "Enable Web Search" box is NOT checked in SillyTavern settings.
5. Download "Chatstream v2.1 Cold" preset from Reddit and import it into SillyTavern. This makes it so you dont have to mess around with the settings or prompts.
6. If you are using Deepseek, then I HIGHLY reccomend you go to the Prompt Post Processing dropdown at the bottom of the Model connection tab and change it to "Single user message (no tools)". Before I changed that i kept getting some random wiki dumps about random softwares, or random Weibo links to articles about Huawei when I tried using "Impersonate" or "Continue".
7. If you want the interactive story mode extension, or just want to write something basic for the AI to go into detail with for your response, then add "Guided Generations" as a SillyTavern extension. It makes it so you can either just type how you wish for the story to go and press the dog button and it will carry the story on. Or you can type in a basic response and press the impersonate button and it will use your {{user}} card/persona as a template and generate a detailed chat message for you that you can send.

Edit: Depending on the card, sometimes the Impersonate button in Guided Generations was just generating the exact same message for me and ignoring my input. If that happens to you as well I found switching to a different SillyTavern preset helped. I used "Marinara's Spaghetti Recipe" for that (current version is v6.0). Can get a link to it from Reddit. It made responses a little bit slower for me, but also included a "Thinking" section you could click on when it was generating a message which had the AI tell you what it was taking into account and focusing on, which is very useful for editing the card details to better match your preferences.

Then just either create your own cards or download some from online. I've read that either Chub or JannyAI are probably the best sources rn for Cards. But if anyone has any other good sites for cards, please share.

Hopefully this is helpful for some people who want to try things out but dont want to spend ages tinkering away on settings/prompts.

 

[desmosome]

So is Gemini good for that? I've been eyeing that context window with envy, but ChatGPT has kept me by playing ball with my weird requests. Plus it's Google, so I don't know about privacy. Does Gemini have the same issue where chats get super-laggy if they get long?

Well, google is free for 100 message per day or something like that (I think they nerfed it recently though) and it does get overloaded at peak hours lately. But I'm using that. You can't get anything better for free. And even if you consider paid models, Gemini would be among the top considerations anyways.

Gemini has a "serious" writing style. Like a novelist. And it's very angsty. It vastly prefers to do dark stories full of anger, hurt, and trauma rather than fluffy romance or comedy. Like if you create a scenario where a character is kind of an asshole on the outside but very caring on the inside, Gemini will tend to make them pretty insufferable in general. But this darker tendency works just fine for most of my use case since I do corruption, NTR and shit like that.

 

[tretch95]

Oh man, lol.

Anyone tried to use that AI RPG since the update yesterday?

It's not even "bad", but just total nonsense. Not only that it permanently disrupts the story with extreme events, but it's like the storytelling eventually does a sudden switch to a different movie with different characters.

 

[PsychicStress]

Gohiai is quite good. I've been using it for a while.

 

[Saerwen]

MuselandAI and ArousedAI is shutting down.

 

[D0v4hk1n]

Hey guys just thought I'd let you know I'm using ChatGPT 5 thanks to a preset on chub and it's a masterpiece. Also isn't that costly. Spends like $0.01-0.05 per message.

 

[Orc Wizard]

I started with janitor 1 month before chutes went paid, the absolute horror.

Moved on to OR, the 50 messages a day becomes 3 because each attempt consumes 1 message, just tried setting up gemini but kept getting the "no content received from google ai" thingy and gave up

I am willing to pay for OR's 1000 messages a day but there's no reliable method to pay internationally for me since I don't own a credit card (I'm from Brazil), and even if I paid... it would still be a mess with all the errors I get when using it.


Now I'm using JLLM because it's all I got left, It's not that bad.... But still miles behind DeepSeek, it manages to keep itself in character for like... 15-20 messages before it becomes its own thing that you can easily gaslight into whatever you want.

Like the classic "I walk away..." and then the character you're interacting with immediately grows feelings for you and begs for you to stay, Lol.

Oh, and sometimes it gets into a loop of repeating itself but using different wordings, and its hard to get it unstuck.

 

[D0v4hk1n]

I started with janitor 1 month before chutes went paid, the absolute horror.

Moved on to OR, the 50 messages a day becomes 3 because each attempt consumes 1 message, just tried setting up gemini but kept getting the "no content received from google ai" thingy and gave up

I am willing to pay for OR's 1000 messages a day but there's no reliable method to pay internationally for me since I don't own a credit card (I'm from Brazil), and even if I paid... it would still be a mess with all the errors I get when using it.


Now I'm using JLLM because it's all I got left, It's not that bad.... But still miles behind DeepSeek, it manages to keep itself in character for like... 15-20 messages before it becomes its own thing that you can easily gaslight into whatever you want.

Like the classic "I walk away..." and then the character you're interacting with immediately grows feelings for you and begs for you to stay, Lol.

Oh, and sometimes it gets into a loop of repeating itself but using different wordings, and its hard to get it unstuck.

If it helps, Deepseek since about a couple of weeks has become truly horrible to work with. Like genuinely, JAI LLM is better in some cases.

The only thing that's been actually good since Deepseek for me is ChatGPT 5 or, if you're rich, Claude 4.1 or 3.7 sonnet.

 

[ionogo]

If you're up for paying (but not too much) for API access, check out Arliai Hosted in SE Asia, they take crypto, and their fundamental value is privacy. There are plenty of models, some at 235B.

 

[D0v4hk1n]

If you're up for paying (but not too much) for API access, check out Arliai Hosted in SE Asia, they take crypto, and their fundamental value is privacy. There are plenty of models, some at 235B.

wdym about privacy though? Even things like Open Router are relatively private

 

[tretch95]

Would be helpful if people wrote what they use the bot for, and why they think it's good. E.g.

> kinky chat partner
> autonomous story writer
> interactive story with choices
> full open world sandbox RPG
> etc


You know, it's cool if ppl enjoy chatting with "Sonic the Hedgehog", but that's not really something to measure LLM utility. Same like you don't measure image generation by random "naked woman with big tits" prompts anymore.


Example with the latest model on perchance, it's so disturbing to use the AI-RPG right now. Every scene eventually turns violent and bloody, people panic or have pain for no reason, it's a total nightmare. Or characters clearly described as a human in a realistic contemporary setting suddenly have "moth-like antennae" and shit like that. And that's just before the AI writes about the "red monster eyes" glaring at your from the wardrobe.

It also has some kind of "Woke Mode" active, since every potential sex partner - no matter how interested or desperate - immediately freaks out when the protagonist kisses them without asking for permission. And the following paragraph makes sure to tell you that it's caused by the lack of verbal permission. Sometimes a gentle, reassuring stroke on the shoulder causes the NPCs to scream and run away, curling up in a fetal position and trembling (etc).

Maybe it sounds funny when described like that, but it really gets to you after a while. The new LLM there is totally subscribed to paranoia, insanity, violence, plot twists. Would recommend to stay away until they restore the old one, or find a better new model.

 

[desmosome]

If you are using anything other than Gemini, ChatGPT, Deepseek, Grok, or Claude, you are doing it wrong, frankly. Especially if you actually want to have a story instead of just going straight into a scene and calling it a day.

Unless your random AI site allows API proxies to these big LLMs, there is no point even suggesting it. Just use Janitor or Chub with proxy.

All those AI girlfriend mobile apps and the huge amounts of copy pasta AI chatbot sites are basically just banking on people's ignorance in this relatively new field. An average joe will be quite impressed by even a 8B model with like 8k context window. But anyone that actually dives into this will soon learn the enormous chasm between the flagship models and the random shitty models being used on chatbot sites.

 

[FrankyMcGuyver]

So uhm chatGPT has become completely useless for me, what do you guys use nowadays?

 

[D0v4hk1n]

So uhm chatGPT has become completely useless for me, what do you guys use nowadays?

It's been a while since I chatted. I suggest checking SillyTavern Reddit where they post updates on this stuff. Maybe people are back on Deepseek?

 

[saddlface]

chubai

 

[Pretentious Goblin]

So is ChatGPT just not willing to do anything remotely NSFW now since very recently, or is it just because I unsubbed?

 

[FrankyMcGuyver]

So is ChatGPT just not willing to do anything remotely NSFW now since very recently, or is it just because I unsubbed?

No I had 2 days left on my subscription and it was terrible. Everyone is complaining about it. I was kinda hoping you would have the golden alternative but I guess not. Why did you unsub?

 

[Pretentious Goblin]

No I had 2 days left on my subscription and it was terrible. Everyone is complaining about it. I was kinda hoping you would have the golden alternative but I guess not. Why did you unsub?

I was out of inspiration for scenarios, I suppose. it comes and goes. Plus its memory and multitasking aren't quite there yet for the indepth sims I'd like to be doing, plus long chats get unresponsive, so I have to copy everything over to a new chat but then it loses some context; plus some frustration with its habit of getting into loops and being biased towards whatever outcome the user wants, which means you can't really trust its judgment for things that aren't readily verifiable. I was working on a text-based mystery-NTR game, but when it constantly forgets the rules, and gets into loops of the same thing, and I don't trust it to play the NPCs in a realistic way, with resistance and emotional turmoil, instead of just ratcheting up the NTR... eh. Too much handholding required.

I think and hope that in a few years we'll have models we can run locally on affordable hardware, uncensored, tweakable and with decent reasoning and memory.