Text Generation

Generate text continuations using NovelAI's language models.

Basic Usage

import { NovelAI } from 'novelai-sdk-unofficial';

const client = new NovelAI({ apiKey: 'your-api-key' });

const text = await client.text.generate({
  input: 'Once upon a time, in a kingdom far away,',
  model: 'llama-3-erato-v1',
  maxLength: 80,
});

console.log(text);

Parameters

Parameter	Type	Default	Description
input	string	required	Input text to continue from
model	TextModel	'llama-3-erato-v1'	Model to use
maxLength	number	40	Maximum tokens to generate (1-2048)
minLength	number	1	Minimum tokens to generate (1-2048)
temperature	number	1.0	Randomness (0.1-100)

See Parameters for the complete list.

Models

Model	Constant	Status	Description
llama-3-erato-v1	ERATO	✅ Recommended	Latest model, based on Llama 3 (default)
kayra-v1	KAYRA	✅ Available	Kayra model
clio-v1	CLIO	⚠️ Legacy	Clio model (planned for deprecation)

Note: According to the official API documentation, llama-3-erato-v1 and kayra-v1 are recommended. clio-v1 is a legacy model planned for deprecation. Trial users can only use kayra-v1.

import { NovelAI, ERATO, KAYRA, CLIO } from 'novelai-sdk-unofficial';

// Using string
const text = await client.text.generate({
  input: 'Hello',
  model: 'llama-3-erato-v1',
});

// Using constant
const text = await client.text.generate({
  input: 'Hello',
  model: ERATO,
});

Temperature

Controls randomness in generation:

• 0.1-1.0: More focused, deterministic
• 1.0-2.0: Balanced
• 2.0+: More creative, varied

// More predictable
const text = await client.text.generate({
  input: 'The answer is',
  temperature: 0.5,
});

// More creative
const text = await client.text.generate({
  input: 'The answer is',
  temperature: 1.5,
});

Generation Length

const text = await client.text.generate({
  input: 'Once upon a time',
  minLength: 20,   // At least 20 tokens
  maxLength: 100,  // At most 100 tokens
});

Stop Sequences

Stop generation when specific text is encountered:

const text = await client.text.generate({
  input: 'Character: Hello!\nNarrator:',
  stopSequences: ['\nCharacter:', '\n\n'],
});

Note: stopSequences is tokenized via the token-count endpoint before the request.

Ban Sequences

Prevent specific text from appearing:

const text = await client.text.generate({
  input: 'Write a story',
  banSequences: ['violence', 'explicit'],
});

Note: banSequences also triggers token-count tokenization.

Example: Story Continuation

const story = `The old lighthouse keeper climbed the spiral stairs, 
his lantern casting dancing shadows on the stone walls.`;

const continuation = await client.text.generate({
  input: story,
  model: 'llama-3-erato-v1',
  temperature: 1.0,
  maxLength: 100,
});

console.log(story + continuation);

Example: Dialogue

const dialogue = `Alice: What do you think about the weather?
Bob:`;

const response = await client.text.generate({
  input: dialogue,
  stopSequences: ['\nAlice:', '\n\n'],
  maxLength: 50,
});

console.log(dialogue + response);