Inspiration
Generative AI makes it incredibly easy to create images, videos, and social content. But creating a consistent AI persona over time is still surprisingly difficult.
As a persona evolves, its identity becomes scattered across character descriptions, prompts, reference images, model settings, campaign rules, and previously approved assets. When multiple AI agents and generation models are involved, it becomes even harder to answer simple questions:
Which persona specification produced this asset? Which prompt and model were used? What content was approved before? And what context should an agent use for the next generation?
We built Persona Foundry to solve this context problem.
Our idea was to treat an AI persona not as a single prompt, but as an evolving collection of structured metadata, relationships, history, and creative decisions. DataHub provides the context and lineage layer that makes this possible.
What it does
Persona Foundry is a context-aware, multi-agent studio for creating and managing consistent AI personas.
Users define a persona with its identity, visual characteristics, personality, brand rules, and creative direction. Specialized AI agents then collaborate to plan and generate content while using the persona’s existing context.
DataHub acts as the underlying context graph connecting:
- Persona and character specifications
- Prompt versions
- AI models
- Generated images and videos
- Campaigns
- Approved assets
- Generation history and provenance
Before generating new content, agents can retrieve the relevant persona context instead of relying on an isolated prompt.
After generation, Persona Foundry records what was used and what was produced, creating lineage such as:
Persona → Character Spec → Prompt → Model → Generated Asset → Campaign
This allows creators to understand where every asset came from while helping AI agents maintain consistency across future generations.
How we built it
Persona Foundry is designed around a multi-agent workflow.
Different agents take responsibility for creative direction, persona consistency, content generation, and quality review.
DataHub provides the shared context layer between these agents. Persona metadata, prompt versions, model information, generated assets, and their relationships can be represented and retrieved through DataHub.
When a content request arrives, the workflow:
- Identifies the target persona.
- Retrieves relevant persona and campaign context.
- Builds a context-aware generation request.
- Sends the request to the appropriate image or video generation model.
- Reviews the result for persona consistency.
- Records the generation metadata and lineage.
- Makes the resulting history available for future agents and generations.
Instead of every agent maintaining its own fragmented memory, they operate against a shared source of context.
Challenges we ran into
One of the biggest challenges was deciding what should actually define an AI persona.
A persona is much more than a character prompt. Its identity can be distributed across visual references, personality descriptions, generation parameters, previous successful prompts, campaigns, and human approvals.
Another challenge was designing relationships between these pieces of information without making the workflow overly complicated.
We also had to balance creative flexibility and consistency. If agents follow previous generations too strictly, the persona becomes repetitive. If they ignore history, character identity quickly drifts.
This made us realize that the goal isn’t to reproduce the same prompt repeatedly. The goal is to provide agents with enough structured context to make better creative decisions.
Accomplishments that we’re proud of
We’re especially proud of shifting Persona Foundry from a simple AI content-generation workflow into a context-driven system for managing AI personas over time.
Instead of treating generated images as isolated outputs, Persona Foundry connects them back to the persona, prompt, model, and creative context that produced them.
We also designed the system around multiple specialized agents rather than one large generation prompt. This creates a foundation where creative direction, generation, consistency checking, and governance can evolve independently while sharing the same context.
Most importantly, Persona Foundry demonstrates a new use case for metadata and lineage:
not only understanding where data came from, but understanding how an AI identity evolves.
What we learned
The biggest lesson was that AI personas don’t primarily have a generation problem — they have a context problem.
Modern models can already generate impressive images, videos, and text. The difficult part is maintaining identity and creative intent across hundreds or thousands of generations produced by different models and agents.
We learned that concepts traditionally associated with data platforms — metadata, ownership, lineage, relationships, and governance — become extremely valuable in generative AI workflows.
Data lineage can become creative lineage.
Metadata can become agent memory.
And a metadata graph can become the shared context layer that allows many specialized AI agents to collaborate around the same evolving identity.
What’s next for Persona Foundry
Our next step is to expand Persona Foundry into a complete operating system for AI-native characters and digital creators.
We plan to add deeper integrations with image and video generation providers, automated visual consistency evaluation, richer human approval workflows, and campaign-level analytics.
We also want agents to learn from approval history. When creators approve, reject, or modify generated content, those decisions can become additional context for future generations.
Longer term, Persona Foundry could manage entire virtual talent agencies containing many personas, campaigns, agents, models, and millions of generated assets.
Our vision is for every AI persona to have a traceable creative history — and for every agent working with that persona to understand the context behind it.
Built With
- api
- datahub
- docker
- fastapi
- gemini
- mcp
- openai
- postgresql
- python
- react
- typescript
Log in or sign up for Devpost to join the conversation.