Inspiration
In professional creative environments, the biggest challenge isn’t generating visuals — it’s translating human intent into something machines can reliably execute. Creative briefs are often intentionally abstract, strategically rich, and constrained by legal, budget, and brand considerations. Traditional prompt-based workflows struggle to handle this complexity at scale. This project was inspired by real production workflows inside creative agencies, where ambiguity is not a flaw — it’s part of the process. The goal was to explore how Bria FIBO’s JSON-native control could be used not just for better images, but for better process.
What it does
Creative Intent Agent is an agentic, JSON-native workflow built on top of Bria FIBO. It translates messy, real-world creative intent into production-safe, deterministic visual parameters. Instead of outputting a fragile text prompt, the system: Asks clarifying questions Separates human intent from execution constraints Resolves contradictions and risks Outputs structured JSON defining camera angle, lighting, field of view, color palette, composition, and emotional intent The resulting JSON is directly compatible with Bria FIBO’s professional, deterministic generation controls.
How we built it
The project was built as a lightweight Python-based agent system: A messy creative brief is ingested The agent asks targeted clarification questions, similar to a producer or strategist Human responses are captured in an intent layer Ambiguities, contradictions, and production risks are resolved A FIBO-ready JSON configuration is generated The system was designed to mirror real enterprise workflows, where such an agent would run inside a private environment and feed structured data into Bria FIBO’s generation layer.
Challenges we ran into
One of the main challenges was balancing creative ambiguity with production safety. Creative intent is often subjective and emotional, while generation systems require precision and determinism. Another challenge was navigating real-world constraints such as API access, gated models, and local hardware limitations during the hackathon. This reinforced the importance of designing workflows that are modular, abstracted, and production-aware rather than tightly coupled to a single execution environment.
Accomplishments that we're proud of
Designing an agentic workflow that mirrors real creative production processes Demonstrating JSON-native visual control instead of prompt engineering Creating a system that prioritizes safety, scalability, and repeatability Aligning Bria FIBO’s technical strengths with real-world creative needs
What we learned
This project highlighted that the real power of generative AI in professional environments lies not in better prompts, but in better systems. By separating intent, constraints, and execution, AI generation becomes more reliable, explainable, and scalable. Bria FIBO’s structured controls make this approach especially powerful for enterprise and production use cases.
What's next for Creative Intent Agent
Future work includes: Connecting the agent directly to live Bria FIBO API or local model execution Expanding safety and compliance rules for regulated industries Adding a UI layer for non-technical creative teams Supporting versioned intent and creative iteration over time Creative Intent Agent is a step toward making AI visual generation a reliable production tool — not just a creative experiment.
Log in or sign up for Devpost to join the conversation.