Novelcrafter custom API: an uncensored model for your manuscript

Add an OpenAI Compatible Connection in Novelcrafter, pick an uncensored prose build with a 262K+ token window and see what each chapter costs.

Steps checked October 24, 2026 · prices as of November 16, 2026

Through a custom API connection, Novelcrafter can draft with models it never lists by name. This guide sets one up for uncensored prose builds, chosen for context windows big enough to carry a lot of the manuscript in one prompt, and works out what a chapter costs on each from the live catalog. Setting names come from Novelcrafter's help page as read on 24 October 2026.

New to builds that had their refusals removed? Read AI with no restrictions first, then return here to connect one.

What you need

  • A key from the console. It begins with sk-unr- and is displayed a single time. Credit is prepaid from $10 by card, WeChat Pay or crypto. A key with no credit on it still connects; the first generation then returns 402.
  • A Novelcrafter account. Novelcrafter runs in the browser and, per its Help pages, has no built-in credits: every model comes from a vendor you connect with your own key.
  • The ids of the models you want, copied from the uncensored model list or from the block in the steps below.

To try a build before you set anything up, open /chat: the same models, in the browser, with a few free turns a day and no account.

Add the connection in Novelcrafter

Novelcrafter names a handful of vendors directly and reaches every other one through a single connection type, the OpenAI Compatible Connection. Set one up like this.

  1. Open the AI connection settings in Novelcrafter.
  2. Scroll to the end of the vendor list, where OpenAI Compatible Connection sits, and press add.
  3. Give it a Name, put https://api.unrestrictable.ai/v1 in the base URL field and paste your key.
  4. Give the model list a moment. Novelcrafter usually fetches the models by itself.
  5. If the list stays empty, turn on Use custom model list and add the ids by hand, one per line or comma-separated. Novelcrafter does no spell check on them, so paste instead of typing. The four used in this guide:
text
1qwen3.8-27b-fable
2qwen3.8-27b-queen
3qwen3.8-27b-novelist
4gemma-4-31b-novelist

One more note from the help page: when a vendor's model names do not match the names in Novelcrafter's default collections, a custom model collection has to be created. Ids such as qwen3.8-27b-fable are specific to this API, so plan on doing that once.

The three connection values

Only three values are involved. The labels below are those on Novelcrafter's OpenAI API Compatible help page, which was last updated on 19 August 2026 and which we read on 24 October 2026.

Name
Unrestrictable
Free text; how the connection appears in your list
Base URL
https://api.unrestrictable.ai/v1
Finishes with /v1, which the help page asks for
API key
sk-unr-live-…
From the console; displayed a single time

Base URL and key are all the help page requires up front; models come after. It remarks that compatible vendors tend to have a base URL ending in /v1, and this one does.

Uncensored builds with room for a manuscript

Each scene prompt may bundle codex entries and earlier text with the beat itself, and the context window decides how much of that the model can read in a single call. That is the filter behind the four builds below: every one is tagged roleplay in the catalog and has a window of at least 262,144 tokens.

Uncensored models and their prices per million tokens
Wire formats
Qwen3.8 27B Fableqwen3.8-27b-fableRoleplay$0.875$5.25524Kopenai
Qwen3.8 27B Queenqwen3.8-27b-queenRoleplay$0.875$5.25524Kopenai
Qwen3.8 27B Novelistqwen3.8-27b-novelistRoleplay$0.35$2.10262Kopenai
Gemma 4 31B Novelistgemma-4-31b-novelistRoleplay$0.35$1.575262Kopenai

Prices per million tokens, read from the live catalog on 2026-11-16.

Model idWindowOutput ceiling per call
qwen3.8-27b-fable524,288 tokens32,768 tokens
qwen3.8-27b-queen524,288 tokens32,768 tokens
qwen3.8-27b-novelist262,144 tokens32,768 tokens
gemma-4-31b-novelist262,144 tokens32,768 tokens

Windows, output ceilings and prices from the live catalog, 2026-11-16.

Three are tunes of Qwen 3.8 27B, whose relatives are on the uncensored Qwen page. The fourth is a Gemma 4 31B tune, listed next to its siblings on the uncensored Gemma page. Run one beat through two of them and keep the voice you like. The catalog can tell you windows and prices; it cannot tell you taste.

The output ceiling bounds a single generation. A scene of 1,000 to 2,000 tokens is far below it on all four.

Windows in words

Windows are measured in tokens, and a token is less than a word. For English prose a common estimate is about 0.75 words per token. Using that figure, the table converts each window into an approximate word count, so you can picture how much codex and back-story one call can hold.

Model idWindowApproximate words
qwen3.8-27b-fable524,288 tokensabout 393,000
qwen3.8-27b-queen524,288 tokensabout 393,000
qwen3.8-27b-novelist262,144 tokensabout 197,000
gemma-4-31b-novelist262,144 tokensabout 197,000

Window × 0.75, to the nearest thousand. Names, dialogue and punctuation shift the true ratio.

Sending the entire draft is seldom the goal. Each prompt token is charged as input on each call, so a prompt holding the codex notes that matter plus the last scene or two costs a small share of one holding the whole book. Where the big window earns its place is the call that needs a wide view, say a rewrite that must agree with several chapters written earlier.

Give Novelcrafter its own key and a cap

Any key works in any client, but a dedicated key for Novelcrafter is simpler to look after. In the console you can hold several keys, put a monthly limit on each and read the usage per key. Should a key leak, you revoke that one and the rest carry on.

The limit also puts a ceiling on a long night of drafting. Set it at what the book may cost you this month. Once the key hits it, every request returns 402 with the code monthly_limit_reached until you raise the limit.

The price of a chapter

Suppose a chapter takes 12 generations. Each one sends around 30,000 tokens of codex and earlier text and receives around 1,000 tokens of prose. Those are assumptions, not measurements. How much context each beat drags along is up to you.

ModelOne chapterA 25-chapter draft
Qwen3.8 27B Fable$0.38$9.45
Qwen3.8 27B Queen$0.38$9.45
Qwen3.8 27B Novelist$0.15$3.78
Gemma 4 31B Novelist$0.14$3.62

Worked out at render from catalog prices on 2026-11-16: generations × prompt tokens × input rate, plus generations × prose tokens × output rate.

Input is nearly the whole bill. 12 × 30,000 comes to 360,000 input tokens, against only 12,000 tokens of output. A wide window lets more of the book in, and you pay for all of it, every call. Choosing which codex notes a beat carries saves more than changing models. All per-token rates are on the pricing page.

Test the key from a terminal

This free request returns every model id your key can use. Replace the placeholder with your key and run it.

curl
1curl https://api.unrestrictable.ai/v1/models \
2 -H "Authorization: Bearer sk-unr-live-…"

A 200 with ids in it confirms the key, and those ids are exactly what goes in a custom model list. A 401 points at the key, and the body says how:

http
1HTTP 401
2{"error":{"message":"No API key was sent.","type":"authentication_error","param":null,"code":"invalid_api_key"}}
http
1HTTP 401
2{"error":{"message":"The API key is not valid.","type":"authentication_error","param":null,"code":"invalid_api_key"}}

Response bodies captured from the API on 24 October 2026.

When the connection fails

401: key

"No API key was sent." says the key field is blank. "The API key is not valid." says it is mistyped, truncated or revoked. Make a new key in the console and paste all of it.

404 model_not_found: model id

One id in your custom model list is missing from the catalog: a typo, a display name in place of an id, or a retired build. Rerun the request above and copy the id out of the response.

402: balance

Your prepaid credit has run out. Add funds in the console.

Browser issues: CORS, Firefox, Safari

Novelcrafter talks to the API straight from the browser. The API accepts requests from every origin, Authorization header included (preflight verified 24 October 2026), so no proxy is involved.

The help page warns that a few vendors follow the standard loosely, which can break things in Firefox and Safari, and recommends a Chromium browser like Chrome, Brave or Vivaldi. We have not run this connection inside Novelcrafter on Firefox or Safari ourselves. If the models load in Chrome and not elsewhere, stay in Chrome.

A fixed prompt goes to each model every day in the daily real-model check, which is how we know the model behind an id matches its label. Base URL, key and id details for other apps are on the docs page.

Questions

Any vendor that follows the standard works through its OpenAI Compatible Connection: a base URL, normally ending in /v1, plus a key. The help page also names vendors that cannot be reached from the browser this way.

No. Novelcrafter’s Help says there is no built-in credits system; you bring a vendor and your own key, and that vendor bills the usage. Here it is per token, prepaid from $10. See per-token prices.

Every id on the uncensored list works through the connection. For a long manuscript, begin with one whose window holds your codex plus the recent chapters; the four in the table above all have 262K tokens or more. See the uncensored model list.

The key or the base URL is wrong, or detection never ran. Test the key with the terminal request, then enable Use custom model list and paste the ids in.

Novelcrafter’s help page reports trouble with some vendors in those browsers and points users to Chromium. The API itself accepts any origin, but we have not tested Novelcrafter in Firefox or Safari.

No. These builds are abliterated or uncensored fine-tunes, so the refusals were removed from the weights. There is no prompt text to add in Novelcrafter. See AI with no restrictions, explained.

Every model above is on the uncensored model list with its live price per million tokens. Per-token pricing sets the whole catalog side by side, and the family directory groups the builds by base model.

Get a keyPrepaid from $10, no subscription.

New models and price changes, once a month.

New uncensored builds with their prices, price moves, and what the daily model check found.

Join on Telegram