Unfiltered AI chat

Short answer: for fiction the problem is rarely subject matter, it is voice: filtered models break character to explain themselves. Uncensored and roleplay models stay in the scene, and SillyTavern can point at them here with an OpenAI-compatible key on the Pro plan ($45 a month).

Anyone who writes with these models knows the failure. The scene is going well and then a paragraph starts “Note that…” and the model steps out of the story to brief you on what it just did. For a summary that is fine. For a draft it is a hole you have to patch.

What actually breaks fiction

Not the big taboos most “uncensored” lists fixate on, in practice:

Abliterated and fine-tuned roleplay models hold the register. That is the whole reason the category exists.

Roleplay models, and the rest

The catalogue has a Roleplay category, and a wider uncensored set behind it:

Roleplay tunes are specifically trained on long-form dialogue, so they keep a voice across a whole scene instead of resetting to assistant each turn. Uncensored general models are better when the scene needs facts in it, and worse at staying in voice.

SillyTavern

This is the setup most writers use, and it needs one thing from us: an OpenAI-compatible endpoint and a key.

  1. Create an account, keep the 16-digit number, add credit and switch to Pro.
  2. Copy the API key from the account page (sk-obs1_…).
  3. In SillyTavern, choose the OpenAI-compatible source, base URL https://obscurify.ai/v1, paste the key, and pick a model from the list.
  4. Set the reply length you want in the preset; the model follows it.

The full walkthrough, including stops and temperature, is in the SillyTavern tutorial.

In the browser instead

If you don’t want a second app, the web chat carries the same models with a roleplay filter in the model picker. Chats stay in your browser rather than on the server, which for a draft you’d rather nobody read is the right default.

A guest can try the interface first with Llama 3.3 70b Instruct and 30 messages an hour, no account and no cookies, but guest chat is a standard model and will caveat.

The costs, honestly

Uncensored and roleplay models are on Pro, $45 a month with $35 of usage included. Long scenes are token-hungry: a 4,000-token context turn costs roughly four times a short reply, so a heavy session can pass the included usage. Prices per model are on the model pages, and usage past the credit is billed from the balance at list price.

There is no image generation here, so if a character needs a portrait, that happens somewhere else.

Common questions

Which models should I use for fiction?

Roleplay-tuned models for voice, uncensored general models when the scene needs accuracy. Both are on Pro.

Does SillyTavern work with this?

Yes, over the OpenAI-compatible API with a Pro account key. The setup is in the tutorial.

Do I need an email?

No. The account is a random 16-digit number, and the API key comes from it.

Are my drafts stored?

Not on the server. Chats live in your browser; with SillyTavern they live in SillyTavern.

Can I try before paying?

Guest chat is free with no account. It uses Llama 3.3 70b Instruct, a standard model, so expect it to break character.

Is there image generation for characters?

No. Text only.

Last updated 2026-11-02. Plan prices and model lists on this page are read live from the site.

Try it now: guest chat needs no account, no email and no cookies. When you want more models, an account is just a random number.

Get an account number See pricing
Get an account number