TLDR: Constella is an AI-powered multi-agent tool designed to help storywriters create and develop interconnected characters. It features FRIENDS DISCOVERY for generating related characters, JOURNALS for exploring characters’ inner thoughts on shared themes, and COMMENTS for manifesting character relationships through interactions. A study showed it helps writers expand character casts, deepen understanding of character dynamics, and maintain authorial control by providing adaptable, intermediary creative resources.
Creating compelling characters and their intricate relationships is a cornerstone of long-form storytelling. However, many writers face significant challenges in this process, from envisioning new characters that influence existing ones to balancing similarities and differences, and intricately fleshing out their dynamics. A new tool called Constella, developed by researchers at KAIST, aims to address these very struggles by leveraging the power of Large Language Model-based Multi-Agents (LLM-MA).
Constella is designed as an LLM-based multi-agent tool that supports storywriters in creating interconnected character casts. It draws inspiration from social media interactions to provide intuitive features. The core idea is to help writers build expansive communities of related characters, compare their inner thoughts and emotions, and deepen their understanding of character relationships.
Key Features of Constella
The tool offers three main features:
FRIENDS DISCOVERY: This feature helps writers expand their character ensemble by suggesting new characters related to a given one. For instance, if you have a protagonist, you can prompt Constella to suggest their “greatest enemy” or “long-lost friend.” It generates mini-profiles for three new characters, complete with names, introductions, backstories, and bidirectional relationship descriptions. Writers found this feature helpful for discovering diverse characters in desired contexts and even uncovering unexpected yet meaningful connections. It also allowed them to create new storylines based on the suggested characters, or even build entire communities of interconnected figures early in their ideation process.
JOURNALS: This feature allows writers to delve into the inner worlds of their characters. By providing a shared theme, Constella generates diary entries from multiple selected characters, revealing their internal emotions, thoughts, and behaviors. This helps writers compare how different characters interpret the same situation, aiding in distinguishing their personalities and creating nuanced similarities. It was particularly useful for developing supporting characters and immersing writers in multiple perspectives simultaneously, helping to distribute their creative attention more evenly across the cast. Writers also drew inspiration from the visual and spatial descriptions within the journal entries, using them as symbolic reflections of characters’ internal states.
COMMENTS: Building on the Journals feature, Comments enables characters to respond to each other’s journal entries. This feature surfaces deeper layers of character relationships through simulated interactions. Writers used it to explore the psychological interplay between characters and elaborate on relational subtleties, such as hidden animosities or reinforcing existing dynamics. While some found it less detailed than Journals, it was effective for observing brief, open-ended character reactions.
Impact on Writers and the Creative Process
A deployment study with 11 storywriters showed that Constella integrated well into their practices and even introduced notable shifts. Many writers found that FRIENDS DISCOVERY mirrored their usual approach of conceptualizing new characters in relation to existing ones. JOURNALS was also highly aligned with how writers consider different character reactions to shared situations. The tool encouraged writers to start with character development, even if they typically began with plot conflicts. It also shifted their focus towards deeper engagement with a broader range of characters, including those they might have otherwise overlooked.
Crucially, Constella was perceived as a “complementary” tool or an “assistant writer,” rather than an AI that takes over the creative process. This was achieved by design choices that constrained AI involvement, such as requiring writers to create initial main characters themselves, and by providing “intermediary materials” like journal entries and comments instead of ready-made story text. These formats encouraged writers to actively interpret and extract meaning from the AI-generated content, preserving their authorial control and fostering a sense of ownership over their work.
Also Read:
- HAWK: A New Framework for Seamless Multi-Agent AI Collaboration
- Bridging Generations: How AI and Ancient Characters Empower Elderly Migrants to Share Their Stories
Future Directions
Writers expressed interest in using Constella for large-scale stories requiring expansive worldbuilding and complex relationships, such as serialized works. They suggested improvements like generating characters associated with multiple individuals, including character relationship maps, and organizing characters into groups. There’s also a desire for features that allow characters to recall past entries, version control for character profiles, and the ability to reorganize journal entries into new categories or timelines. While the tool is not intended for direct narrative progression, future iterations could explore how to provide more action sequences or in-the-moment conversational interactions without compromising authorial agency.
Constella represents a significant step in supporting storytellers by helping them build rich, interconnected character casts. For more technical details on the system’s implementation and prompt design, you can refer to the full research paper available here.
