PersonaMuse: Cultural Intelligence for Character Creation

Inspiration

As writers and game designers, we often face the challenge of creating fictional characters that feel truly authentic and resonate with specific audiences. We wanted to go beyond generic AI character generators by grounding characters in cultural intelligence—the kind of insight that reflects what people genuinely enjoy across music, film, fashion, and more. This inspired PersonaMuse, a tool that uses Qloo’s Taste AI™ and a powerful LLM to craft richly detailed, human-like personas from audience tastes.

What it does

PersonaMuse takes a user’s cultural input (e.g., “likes Kendrick Lamar, Scorsese films, and surrealist art”) and generates a fully fleshed-out fictional character including:

Name, age, personality

Backstory and motivations

Fashion sense, music/food/book preferences

Quirks, emotional depth, and habits

It does this by combining Qloo’s cultural preference expansion with a large language model that synthesizes those insights into creative, believable personas.

How we built it

Frontend: Built using Streamlit for rapid development and a clean user interface.

Cultural Intelligence: Integrated Qloo’s Taste AI™ API to analyze and expand audience input into semantically connected tastes across multiple domains.

LLM Integration: Used Mistral-7B-Instruct via Hugging Face’s free inference API to generate rich character narratives from the cultural graph.

Prompt Engineering: Fine-tuned prompt templates to align the LLM output with the tone, structure, and depth needed for storytelling.

Challenges we ran into

API Rate Limits: Managing Hugging Face inference limits while maintaining quality response time.

Taste Representation: Converting free-text audience descriptions into formats Qloo could parse and extract meaningful recommendations from.

Prompt Balance: Striking the right balance between giving enough cultural detail and allowing creative freedom for the LLM.

Accomplishments that we're proud of

Created an end-to-end working prototype that generates high-quality, culturally grounded characters.

Seamlessly integrated two AI systems (Qloo + Hugging Face) for a novel creative use case.

Achieved high-quality, varied character outputs with no personal data required, keeping it privacy-first.

What we learned

How to creatively combine cultural data and generative AI to elevate storytelling.

The power of semantic cultural graphs for understanding audience taste across domains.

The importance of prompt structure and iterative refinement in controlling LLM outputs for creative tasks.

What's next for PersonaMuse

Character Gallery & Save Feature: Let users save and revisit their favorite characters.

Genre Customization: Allow users to specify the story genre (sci-fi, fantasy, romance) for tailored outputs.

Voice & Visual Expansion: Add speech synthesis and image generation (e.g., using SDXL or DALL·E) for multimodal storytelling.

Writer API Plugin: Integrate into tools like Notion, Scrivener, or Obsidian as a creative assistant.

Built With

• hugging
• llm
• qloo
• streamlit