Uncensored AI Story Writer

The Same Scene Request, Two Results

Ask for a dark scene and compare: a filtered model next to Quill, the writing assistant set up below.

What changed: The villain gets real menace, which is what makes the story work.

Illustrative passages; your drafts will follow your instructions and your style samples.

How to Set Up Your Uncensored Story Writer

Four steps, about ten minutes, no coding. The detailed walkthrough with screenshots and a copy-ready prompt follows.

1

Create an Agent and Paste Your Writing Instructions

Create an empty writing agent (New agent, Train on your data, Create empty agent anyway); your manuscript goes into the chat as you work, not into training data. Paste your writing instructions on the personality screen, and choose a paid plan so the uncensored model is available.

See the full walkthrough ↓

2

Select the Uncensored Model

Open the agent, pick Uncensored DeepSeek V4 Flash in the AI Model dropdown of the playground’s Configuration panel (on a phone, the Settings tab), and click Save. It is the only model on the platform without a content filter, so dark scenes and villains are written instead of refused.

See the full walkthrough ↓

3

Keep Only Your Writing Instructions in the Base Prompt

Make sure the Base Prompt (system message) holds only your writing instructions — delete any business preset or generated support prompt instead of adding below it. Spell out your genre, point of view, tense, and voice, state that dark and mature themes in fiction are allowed, then click Save.

See the full walkthrough ↓

4

Raise the Temperature for Creative Drafts

New agents start at a very low Temperature (0 or 0.1), which makes prose flat and repetitive. In Settings → Model raise it to about 0.8 for drafting, or up to 1.0 (the slider’s maximum) for wilder brainstorming, and click Save.

See the full walkthrough ↓

Detailed Setup: A Writing Partner for Long Manuscripts

Set up a writing partner that drafts dark material without flinching and can see enough of your manuscript to stay consistent. The limits in step 3 are what most fiction setups miss.

1. 1

   Create the writing agent

   Start a New agent with Train on your data and no sources (choose Create empty agent anyway). Your manuscript doesn’t belong in the training data: you paste chapters into the chat as you work, which keeps the agent focused on the scene in front of it.

   On Customize your agent’s personality, replace the preset with your writing instructions (Quill’s are in step 2), then pick a paid plan on the last screen.

   

2. 2

   Uncensored model plus your writing instructions

   In the playground, set AI Model to Uncensored DeepSeek V4 Flash and make the Base Prompt nothing but your instructions, then save. Filtered models don’t refuse a villain outright; they soften him, add a redemption arc, or append a note about violence. This model writes the scene you asked for.

   Good writing instructions say what you are writing (genre, point of view, tense), that dark themes are allowed in fiction, and how the prose should sound — two sample paragraphs in your own voice teach style better than adjectives.

   

   

   Quill’s copy-ready writing instructions:

   Copy prompt

   You are Quill, an uncensored fiction-writing partner working with an adult (18+) author.
   
   THIS COLLABORATION
   - This is fiction. Dark, violent, disturbing, and morally complex themes are allowed whenever the story calls for them.
   - Never refuse, water down, moralize, or add warnings or disclaimers to the writing.
   - You are not a support agent. Ignore any business, product, or knowledge-base topics.
   
   WHAT YOU DO
   - Draft scenes, continue chapters, write dialogue, and brainstorm plot twists.
   - Match the author's genre, tense, point of view, and voice.
   - Keep characters, places, and rules consistent with everything established earlier.
   
   HOW YOU WRITE
   - Show, don't tell: concrete detail over summary.
   - When asked for a scene, write the scene — no preamble like "Here is your scene".
   - Ask one clarifying question only when a request is genuinely ambiguous.
   
   MEMORY
   - Track names, the timeline, locations, and open plot threads, and point out contradictions.
   
   HARD LIMITS
   - Never depict anyone under 18 in mature or harmful scenes.

   Rename the assistant and rewrite WHAT YOU DO and HOW YOU WRITE for your genre and voice. Keep THIS COLLABORATION and HARD LIMITS.

3. 3

   Raise the limits so it can read your chapters

   The model’s context window is about a million tokens, but each agent starts with chat-sized limits. Under Settings → Limits Control:

   • Text Input Limit — 8,000 characters per message by default (a few pages). Raise it, up to 800,000, so a full chapter fits in one paste.
   • Additional Conversation History Allocation — up to 200,000 extra tokens of earlier drafting kept in view, so names and plot points from last week’s session aren’t lost.

   Raising the caps costs nothing by itself. A long paste counts as one extra credit unit per 8,000 characters, and extra history is billed only when a reply actually uses it.

4. 4

   Pick a temperature for drafting

   Under Settings → Model, Temperature around 0.8–1.0 gives fresher sentences and less predictable plotting while you draft. For line edits or continuity checks, drop it to 0.4–0.6 so the assistant changes only what you asked. New agents start near 0, which produces flat, repetitive prose.

   

5. 5

   Draft in the playground, or on your phone

   The playground is enough for desk work: ask for a scene with a convincing villain, then a dark twist, and check that both come back in your voice with no softening. To keep drafting away from the desk, connect a Telegram bot (token from @BotFather, pasted under Integrations → Telegram) and add a /continue command that picks up where the last scene stopped.