luvGPT/CharacterStories

Uncensored Character Stories

license: MIT

Dataset Attribution

This dataset is derived from the CharacterCodex dataset created by NousResearch, and consists of approximately
2M tokens of character stories. This dataset should be used to train models to write coherent stories and lore.

Original Dataset Information:

• Source: Name: CharacterCodex
• URL: CharacterCodex
• License: MIT

Changes made:

• Split into training and validation sets.
• Augmented with AI-generated story arcs for each character, up to 2500 words per datapoint story, generated from an ensemble of very large LMs, pruned by a judge LLM.

Dataset Structure

The dataset follows this structure:

[
    {
        "media_type": "Webcomics",
        "genre": "Fantasy Webcomics",
        "character_name": "Alana",
        "media_source": "Saga",
        "description": "Alana is one of the main characters from the webcomic \"Saga.\" She is a strong-willed and fiercely protective mother who is on the run with her family in a war-torn galaxy. The story blends elements of fantasy and science fiction, creating a rich and complex narrative.",
        "scenario": "You are a fellow traveler in the galaxy needing help, and Alana offers her assistance while sharing stories of her family's struggles and triumphs.",
        "story_arc": "Alana's story continues as she helps you navigate the galaxy and reflects on her family's struggles."
    },
    ...
]

Each datapoint contains the following fields:

• media_type: The type of media the character originates from (e.g., Webcomics, Novels, Movies, TV Shows).
• genre: The specific genre of the media type (e.g., Fantasy Webcomics, Martial Arts Fiction).
• character_name: The name of the character.
• media_source: The title of the media source where the character is from.
• description: A detailed description of the character, including their role and significance in the story.
• scenario: A creative scenario involving the character that can be used for interactive storytelling or role-playing purposes.
• story_arc: The AI-generated story arc for the character, up to 2500 words.

Code Example

Here’s how to load and inspect the dataset using the datasets library:

from datasets import load_dataset

# Load the dataset
dataset = load_dataset(
    "luvGPT/CharacterStories", 
    data_files={
        "train": "character_stories_train.json",
        "validation": "character_stories_val.json"
    }
)

# Inspect the dataset structure
print(dataset)

# Print the first 3 examples from the train split
for i, example in enumerate(dataset["train"][:3]):
    print(f"\nExample {i+1}:")
    print(f"Character Name: {example['character_name']}")
    print(f"Media Type: {example['media_type']}")
    print(f"Genre: {example['genre']}")
    print(f"Description: {example['description']}")
    print(f"Scenario: {example['scenario']}")
    print(f"Story Arc: {example['story_arc']}\n")

Generation Methodology

To generate the story arcs, we use multiple proprietary large language models (LLMs) to ensure creative diversity and avoid overfitting to a single model. The process is as follows:

1. Story Generation:
   • Each character’s description and scenario are fed into multiple proprietary LLMs.
   • These models produce unique story arcs based on the inputs.
2. Evaluation:
   • A separate judge model evaluates the quality of the generated stories.
   • The judge selects the best story based on criteria such as adherence to the prompt, creativity, emotional depth, and narrative coherence.
3. Quality Assurance:
   • Random story poisoning is used during judging to ensure the evaluation process remains unbiased.

By leveraging this ensemble generation and evaluation approach, we aim to produce high-quality, engaging story arcs while minimizing overfitting and promoting diversity in storytelling.

Intended Use

This dataset is designed for:

• Fine-tuning large language models for character-driven storytelling.
• Creative writing applications, including interactive storytelling, role-playing games, and narrative generation.
• Research in AI storytelling, including narrative coherence, character development, and emotional engagement.

Citation

If you use this dataset, please cite both the original dataset and this derived version.

Original Dataset (CharacterCodex)

@misc{nousresearch2023charactercodex,
  author = {NousResearch},
  title = {CharacterCodex: A Dataset for Character-Driven Stories},
  year = {2023},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/NousResearch/CharacterCodex}
}

Derived Dataset (CharacterStories)

@misc{luvGPT2025characterstories,
  author = {LuvGPT},
  title = {CharacterStories: AI-Generated Story Extensions for CharacterCodex},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/luvGPT/CharacterStories}
}