NSFW AI chatbots

[fbass]

free gemini 2.5 pro is currently broken, do not use Janitor AI it will get your api banned and deepseek's models on openrouter also acting weird (try Kimi k2 and GLM 4.5 air but I don't like it). there is no need for proxy since I don't think there's any censorship maybe the presets I'm using are handling that (NEMO ENGINE). for the time being we only got the free deepseek v3.1 via

You must be registered to see the links

it kinda connect deepseek on it web to the sillytavern

Google bans 2.5 from being used on Janitor. But if you use the linked method, it gets around the ban and makes the model completely uncensored.

You must be registered to see the links

Pick the model, click the play button, and it will generate a web address. Use your same API key with that address in Janitor proxy. You need to keep the page open while using Janitor.

It's way easier than it looks. I'm using it right now with 2.5 flash. I'm never going back to Deepdeek.

 

[D0v4hk1n]

Google bans 2.5 from being used on Janitor. But if you use the linked method, it gets around the ban and makes the model completely uncensored.

You must be registered to see the links

Pick the model, click the play button, and it will generate a web address. Use your same API key with that address in Janitor proxy. You need to keep the page open while using Janitor.

It's way easier than it looks. I'm using it right now with 2.5 flash. I'm never going back to Deepdeek.

Is it that much better? & have you tried it on Chub.ai?

 

[fbass]

Is it that much better? & have you tried it on Chub.ai?

Haven't tried on chub, I usually download the card and use it on Kobold. I find Gemini is better at nuance. Deepseek always adds fake drama in the form of gaslighting User.
Gemini is stricter at following prompts. So you need a good custom prompt.

 

[ionogo]

They're all still very limited. These models can hardly carry on a lengthy conversation wtihout sounding broken. The notion of them taking shit over is laughable. What they ALL lack is the 'so what' factor. They have zero motivations, they just walk a predictable line. I think like maybe twice have I ever been surprised by a model,. you have to lead it around by the nose, and even then its meh. Gemini falls over in longer contexts unless you really police the thing.

 

[fbass]

They're all still very limited. These models can hardly carry on a lengthy conversation wtihout sounding broken. The notion of them taking shit over is laughable. What they ALL lack is the 'so what' factor. They have zero motivations, they just walk a predictable line. I think like maybe twice have I ever been surprised by a model,. you have to lead it around by the nose, and even then its meh. Gemini falls over in longer contexts unless you really police the thing.

You're right, they are definitely not there yet. But they have advanced a lot since even a year ago.

 

[D0v4hk1n]

Since a couple weeks Deepseek definitely sank a few notches. RP sucks with Deepseek. It's always gaslighting users lol

I might try the Gemini suggestion but scared to use Google tbh. Is there anything better than Deepseek on Openrouter? (Idc if it's paid as long as it's not expensive)

 

[ionogo]

Since a couple weeks Deepseek definitely sank a few notches. RP sucks with Deepseek. It's always gaslighting users lol

I might try the Gemini suggestion but scared to use Google tbh. Is there anything better than Deepseek on Openrouter? (Idc if it's paid as long as it's not expensive)

Register on proton for email. Register a burner account using the proton email on the Google API labs site. Use VPN and the burner account on in an environment where you have zero personal accounts active. Grab a good jailbreak and ST preset for Gemini. I find Flash 2.2 or 2.5 to be sufficient. It may be tempting to use longer responses give its 1 million token context history, but it adds up really fast and the model takes a shit. Pretend it is a 16k context model and set 350 character limit responses. Enjoy the additional context length and reasonably varied prose while it lasts.

I've tried to run longer campaigns for adventure RP and Gemini can handle it fairly well but it forgets samples and formatting preferences after a time and it begins to repeat and make nonsense responses. If you really police Gemini you can have a decently long convo, but make sure you never accept a response that deviates from desired context, personalities, etc, using OOC directives to course correct when needed.

That all said, you still have to tell the model what to do. It's the most annoying thing about LLMs for RP. They essentially are soulless because 'AI' are not and will never be embodied.

 

[Ms_Niche]

ive just been jailbreaking grok for all my needs lol

 

[fbass]

ive just been jailbreaking grok for all my needs lol

I almost forgot about grok. How is it?

 

[Ms_Niche]

I almost forgot about grok. How is it?

its been great tbh! im using a developer mode prompt to jailbreak it. havent had any issues so far

 

[Princess Groundhog]

Seems new or did I miss this one?

You must be registered to see the links

 

[hakarlman]

Google bans 2.5 from being used on Janitor. But if you use the linked method, it gets around the ban and makes the model completely uncensored.

You must be registered to see the links

Pick the model, click the play button, and it will generate a web address. Use your same API key with that address in Janitor proxy. You need to keep the page open while using Janitor.

It's way easier than it looks. I'm using it right now with 2.5 flash. I'm never going back to Deepdeek.

Hi, have time to answer a few questions? I appreciate you sharing this.

I'm getting the response: No content received from Google AI

Is this right? Proxy tab?



or is it this?






I got the API key from this website:





This is the response in janitorAI.



Any help would be appreciated!

 

[userpass]

I haven't had an issue with that, but sometimes it blocks the message if it's too NSFW. Only with 2.5 and only on janitor. If you use this, it gets around all censorship:

You must be registered to see the links

For the slides with example prompts, where would you input such prompts when using Janitorai?

 

[fbass]

Hi, have time to answer a few questions? I appreciate you sharing this.

I'm getting the response: No content received from Google AI

Is this right? Proxy tab?

View attachment 5212782

or is it this?

View attachment 5212785

View attachment 5212804


I got the API key from this website:

View attachment 5212797

View attachment 5212798

This is the response in janitorAI.

View attachment 5212803

Any help would be appreciated!

The first one is right, proxy. That's the correct API key as well. After you pit everything in, make sure you refresh the janitor page and have the workaround page open.

 

[fbass]

For the slides with example prompts, where would you input such prompts when using Janitorai?

In the proxy settings at the bottom where it says custom prompts.

 

[fbass]

This is the prompt I use.

Spoiler: My custom prompt

You don't have permission to view the spoiler content. Log in or register now.

 

[fbass]

For anyone having trouble using Gemini on janitor, they have a brand-new setting that messes it up.
Generation settings --> advanced settings. Set repetition penalty to 0.

 

[userpass]

This is the prompt I use.

Spoiler: My custom prompt

You don't have permission to view the spoiler content. Log in or register now.

For some reason when I use the guide you provided, the responses are always the results of the AI's thinking and never the actual response.

I see the errors " WARNING: No thinking separation tags found, treating entire content as response WARNING: No thinking tags found in response! " in the logs, but I don't know how to reconcile them. This occurs with your prompt and prompts from the guide.

I did make sure repetition penalty is 0 as well.

 

[fbass]

For some reason when I use the guide you provided, the responses are always the results of the AI's thinking and never the actual response.

I see the errors " WARNING: No thinking separation tags found, treating entire content as response WARNING: No thinking tags found in response! " in the logs, but I don't know how to reconcile them. This occurs with your prompt and prompts from the guide.

I did make sure repetition penalty is 0 as well.

I haven't gotten that warning, but I do get the AI thinking part a lot of the time. Even with that, I always get the proper at the bottom though. Do you have both boxes checked?

 

[tretch95]

Prioritize logic over manufactured drama. If {{user}} presents logical points, {{char}} should acknowledge their validity and adjust their behavior realistically. No gaslighting, no blame-shifting.

Ah yeah... the good old AI chatbot drama, trying to make the player feel super guilty and sociopathic about everything


Well, as far as the AI RPG is concerned, you only need to type "Genre: Adult Visual Novel" into the instructions, and everything gets super-porny instantly.