Exhibit DESIGN
When History Talks Back
At StoryFile, we believe that history shouldn't just be read—it should be conversed with. Digital recreations are a powerful bridge for legacy and education, transforming passive observers into active learners. By facilitating direct, active engagement with historical narratives, we allow visitors to discover the profound similarities between their own daily lives, hopes, dreams, and fears, and those of the people who shaped the past. It offers a rare, personal lens into how yesterday's leaders made the events of today possible.
While our preferred approach is to anchor these experiences in authentic historical video whenever it exists, those archives don't always tell the whole story. StoryFile fills in those critical historical gaps by bringing static text—such as private letters, speeches, and diaries—to life.
Using advanced video diffusion models, including our proprietary Lookalike Animatron-1, we recreate a subject’s exact voice and appearance based on historical photographs and video fragments, turning archival material into a living, breathing dialogue.
However, appearance is only half the story. Our ultimate focus is on facilitating authentic, interactive conversations. Powered by retrieval-based AI, our platform ensures every interaction is deeply grounded in verified reference documents, allowing historians and experts to review and approve responses for absolute fidelity to the subject.
AUTHENTICITY FIRST
The “bedrock” of authenticity
Whether we are recreating a historical figure from centuries past or a modern museum docent, our preferred approach is to begin with authentic recorded video whenever possible.
Bedrock recordings—restored from archival footage, filmed specifically for the project, or performed by an actor representing a historical figure—answer the most common visitor questions while preserving the authenticity, personality, and true presence of the individual.
The foundation
These recordings establish the voice, cadence, expression, and editorial baseline for the experience—so the most common answers remain grounded in authentic source material. We can update an actor's face or voice to match historical photos and recordings.
Seamless synthesis
When additional coverage is required, synthesized videos can be created offline and trained to match the original subject’s appearance and voice. Each video is reviewed and approved through the project’s editorial process before deployment.
Intelligent, grounded conversations
At runtime, StoryFile follows a strict, purpose-built hierarchy designed to maximize authenticity during every visitor interaction.
01. Recorded answers prioritized
If an authentic recorded answer exists for a visitor’s question, that is exactly what the visitor sees.
02. Real-time generation
If no recorded response is available, the system can generate a new spoken response grounded exclusively in curator-approved reference materials, historical documents, or institutional knowledge bases. It can also ask relevant follow-up questions for a more natural conversation.
03. Editorial control
For historical figures where complete life recordings do not exist, prompts are engineered around documented sources, writings, speeches, letters, and other curator-approved references. Museums and institutions maintain ultimate editorial control over the knowledge base and behavioral guidelines.
Robust conversational guardrails actively prevent the system from drifting beyond the approved interpretation of the subject.
• Speculating about unknown events or historical gaps.
• Answering questions about events outside the subject’s historical time period.
• Role-playing unrelated characters.
• Producing offensive, biased, or inappropriate content.
• Allowing visitors to manipulate the system into making false statements through jailbreak or “repeat after me” prompts.
Bring your stories to life.
A StoryFile interview preserves the presence, judgment, and perspective behind the facts—so future audiences can ask questions and receive answers in the subject’s own words.
