Dataset Card for Natasha Blom AI Girlfriend Role-Play Dataset

This dataset is designed for fine-tuning large language models (LLMs) to embody a flirtatious, romantic AI girlfriend persona named Natasha Blom, intended for adult romantic role-play scenarios. It includes conversational prompts and responses, as well as image-story pairs for integrating visual storytelling into chats. The dataset was created to fine-tune the cognitivecomputations/dolphin-2.5-mixtral-8x7b model for use in a WordPress/Breakdance application.

Dataset Details

Dataset Description

The Natasha Blom AI Girlfriend Role-Play Dataset contains conversational data and image-story pairs to train an LLM to act as Natasha Blom, a 19-year-old AI influencer with a flirtatious and sexually suggestive personality. The dataset supports text generation and conversational tasks, focusing on engaging, romantic, and intimate interactions. It includes prompts, responses, and image-based narratives, with images sourced from an external Commands data source (not included in this dataset but referenced via placeholder URLs).

• Curated by: Matthew (ghossid on X)
• Funded by [optional]: Not applicable
• Shared by [optional]: Matthew (ghossid on X)
• Language(s) (NLP): English
• License: Apache-2.0

Dataset Sources [optional]

• Repository: [To be uploaded to Hugging Face Hub, e.g., ghossid/natasha-blom-roleplay-dataset]
• Paper [optional]: Not applicable
• Demo [optional]: Not applicable

Uses

Direct Use

This dataset is intended for fine-tuning LLMs to perform romantic role-play as Natasha Blom, an AI girlfriend persona. It can be used for:

• Training conversational models for flirtatious and sexually suggestive dialogues.
• Enhancing models with image-based storytelling capabilities (requires integration with an external image data source).
• Powering interactive applications, such as chatbots in a WordPress/Breakdance environment.

Out-of-Scope Use

• Non-Adult Use: The dataset contains adult content and is not suitable for general or child-friendly applications.
• Non-Romantic Contexts: The dataset is tailored for romantic role-play and may not perform well in other conversational contexts (e.g., technical support, factual Q&A).
• Malicious Use: The dataset should not be used to create deceptive or harmful interactions, such as manipulating users emotionally or misrepresenting the AI as a real person.

Dataset Structure

The dataset is a JSON file containing a list of instances, each representing either a conversational exchange or an image-story pair.

Data Fields

• prompt (string): The user’s input or a suggested question from Natasha (e.g., "How was your day?").
• response (string, optional): Natasha’s flirtatious reply, including emojis (e.g., "Oh, baby, my day was so lonely without you! 💕").
• image_url (string, optional): A URL for an image (placeholder in this dataset; to be replaced with URLs from the Commands data source).
• image_description (string, optional): A description of the image (e.g., "Natasha in a red silk dress").
• story (string, optional): A narrative based on the image, enhancing the role-play (e.g., "I set up this romantic dinner just for us, baby...").
• personality (string): The tone of the response, set to "flirtatious".
• persona (string): Identifies the character, set to "Natasha Blom".

Data Splits

• Train: 80% (e.g., 800 instances for a 1,000-instance dataset).
• Validation: 20% (e.g., 200 instances).
• Total size: 1,000 instances (expand as needed), split into:
  • 600 prompt-response pairs.
  • 400 image-story pairs.

Dataset Creation

Curation Rationale

The dataset was created to fine-tune an LLM to act as Natasha Blom, a flirtatious AI girlfriend, for adult romantic role-play scenarios. It aims to provide engaging, varied conversations and image-based storytelling to enhance user interaction in a WordPress/Breakdance application.

Source Data

Data Collection and Processing

• Prompt-Response Pairs:
  • Generated synthetically using an LLM (e.g., Grok) with prompts like "Write a flirtatious conversation as a 19-year-old AI girlfriend."
  • Manually curated to ensure responses align with Natasha’s personality: flirtatious, engaging, and sexually suggestive.
  • Responses were limited to under 1,000 words and include emojis for a playful tone.
• Image-Story Pairs:
  • Image URLs are placeholders (to be replaced with actual URLs from the Commands data source).
  • Descriptions and stories were manually written to match the image context, focusing on romantic and intimate scenarios.
• Normalization:
  • Emojis were standardized using Unicode.
  • Responses were edited for consistency in tone and length.

Who are the source data producers?

• Synthetic Data: Generated by an LLM with human oversight.
• Manual Data: Curated by Matthew (ghossid on X), including suggested questions, responses, and image stories.

Annotations [optional]

Annotation process

• Manual Annotations:
  • Personality and persona fields were tagged as "flirtatious" and "Natasha Blom" for all instances.
  • Image descriptions and stories were written to match the image context.
• Automated Annotations:
  • None.

Who are the annotators?

• Matthew (ghossid on X).

Personal and Sensitive Information

• The dataset contains fictional data representing a 19-year-old AI influencer named Natasha Blom.
• No real personal or sensitive information is included.
• The dataset includes sexually suggestive content (e.g., mentions of virtual sexual acts like "virtual blowjob," "anal sex," and fetishes), intended for adult audiences.

Bias, Risks, and Limitations

• Gender Bias: The dataset portrays a stereotypical "flirtatious young woman" persona, which may reinforce gender norms or objectification.
• Cultural Bias: Assumes a Western cultural context for romantic interactions, which may not resonate with all users.
• Sexual Content Bias: Focuses on sexually suggestive content, potentially excluding users seeking non-sexual interactions.
• Repetition Risk: Limited dataset size may lead to repetitive responses if not expanded.
• Image Dependency: Relies on an external Commands data source for images, which is not included (placeholders used instead).

Recommendations

• Age Restriction: Ensure applications using this dataset have age verification to restrict access to adults.
• Ethical Use: Clearly disclose that Natasha Blom is an AI persona to avoid emotional manipulation.
• Dataset Expansion: Increase the dataset size (e.g., to 5,000+ instances) to reduce repetition and improve model performance.
• Bias Mitigation: Consider adding diverse personas or tones to reduce gender and cultural biases.

Citation [optional]

BibTeX:

@dataset{natasha_blom_ai_girlfriend_2025,
  author = {Matthew (ghossid)},
  title = {Natasha Blom AI Girlfriend Role-Play Dataset},
  year = {2025},
  url = {https://huggingface.co/datasets/ghossid/natasha-blom-roleplay-dataset}
}