If you’re tired of trying out different methods to circumvent the censorship and trick the Character.ai filter for whatever valid reason you might have for that, I have a few neat solutions for you!
Want to clone any voice completely for free using just a few seconds of audio? You really need to check out AllTalkTTS! – How To Install & Use AllTalk TTS – The Fastest Method
Can You Disable, Turn Off or Bypass the Filter on Character.ai?
When it comes to the model censorship going on over on Character.ai, their chatting system seems to rely on a semantic filter kicking in post-generation, hinting at the fact that the main model they use for generating the messages is capable of outputting uncensored content which is then, after reaching a supposed filtering threshold being cut by the filter before being fully displayed (or to be exact, sent over) to the end user.
This means that while the model does seem to be able to generate some more spicy and unsavory content, once the filter detects a certain amount of unwanted tokens in the context of the output message or among the logit output data, it cuts the message generation short and serves the user a warning. Although still, without the access to the Character.ai codebase, this is just a theory.
Most of the signs do tell that this filter works fully server side, which means that there are no reliable ways of fully circumventing it by attempting to change any properties of the web app on the client side.
You still can however try and influence the model using “traditional” prompt injection methods, using substitute words, synonyms, context switching techniques and such. As there are already quite a few guides for this online, as well as a whole subreddit dedicated to it, I won’t be getting into more details in this article.
Here is a free, open-source and 100% private local alternative to Character.ai without any kind of filters or message censorship, which you can install on your computer in a matter of minutes.
OobaBooga WebUI – Local AI Character Chat, Simple Setup
The OobaBooga Text Generation WebUI project lets you easily run free open-source large language models on your own PC. It’s free, and the only thing it really requires to work reliably is a graphics card with at least 8GB of VRAM on board.
The best thing about all this? The conversation data and history never leaves your PC, and there are no filters or censorship at work here. Make use of this information however you’d like!
Not sure if your GPU is up for the task? In my tutorial video I’m using the OobaBooga WebUI with my RTX 2070 SUPER 8GB without much trouble. If however, you’d like to be able to use more complex models or have longer conversations with your characters, be sure to check out my curated list of the best graphics cards for local LLM inference this year here.
Text Generation WebUI Quick Installation Tutorial
I’ve already written a short guide for installing the software here, with the exact steps you need to take to get the OobaBooga Text Generation WebUI working on your machine. If you’re set on trying it out, this should be your next stop!
Check out the installation and setup guide here! – How To Set Up The OobaBooga TextGen WebUI – Full Tutorial
If you’d prefer a more visual guide instead, feel free to check out the quick tutorial video over on my YouTube channel going through the whole process in a few short minutes!
Keep in mind that you want to have at least 8GB of VRAM on your graphics card to be able to reliably run basic LLM models on your computer.
Are There Any Other Filter-Free Alternatives to Character.ai?
When it comes to more fun local solutions that I’ve come across, and that work in a very similar way to the OobaBooga Tex Generation WebUI you can take a look at both the SillyTavern which is among the most popular options, and the Gpt4All solution which lets you reliably chat with locally hosted LLMs using only your CPU (no GPU required).
The less ideal, online alternatives do exist, albeit they might not grant you the level of privacy you want. It all boils down to how much you trust the certain service or app provider.
Three of the ones I have personally tested and found to be both good quality and similar to the way that Character.ai functions are listed below:
- Perchance AI Chat – Straight to the point interface with easy character and scenario editing, an all-on-one-page solution without the need to login to test it out. Definitely my favorite one. The only one in which you have to import your own characters, because there are no preset ones to choose from, highly customizable.
- Agnai Chat – Neat interface, easy character creation and editing, an option to either use the Agnaistic models for free, or import your own API keys. Once again, no preset characters, you need to import your favorite ones before you start your first chat.
- Sakura.fm – A lot of preset characters, free to use with a daily message limit. A pretty solid and promising alternative to Character.ai with decent text output quality.
That’s pretty much it! If you’re set on getting the OobaBooga WebUI working on your PC, be sure to check out my full installation tutorial. Until next time!
