GliaDirectors addresses the fragmentation in video production tools by unifying scripting, storyboarding, and rendering into a single, chat-based interface. As the Product Designer and Frontend Developer, I led the creation of 'Director' personas, AI agents that don't just execute commands but collaborates creatively with various workflows.
Video production is fragmented across specialized tools: script editors, storyboard software, and complex video editors. This creates a high barrier to entry for non-experts who struggle to translate abstract ideas into visual narratives. The goal was to transform this intimidating process into a conversational workflow where the AI acts as a creative partner, guiding the user from a simple prompt to a finalized video.
Developed within the GliaCloud Lab Team, this project started out as a prototype exploring the potential for editing video content with natural language. Later targeting SMB marketers and e-commerce owners who need professional video content but lack the technical skills or budget for traditional production teams.

The fragmented nature of current generative content creation landscape can often leave newcomers disoriented.
Balancing Automation & Control
The primary challenge was balancing automation with control. Early prototypes that split tasks too finely (Script Agent -> Timing Agent -> Art Direction Agent...) resulted in a loss of coherency. We needed a system that felt like a cohesive 'Director' capable of making big creative swings while still allowing the user to step in and say, 'Make it more energetic' or 'Change the mood'. My primary design goal was to translate technical backend agents into a human-centric experience that felt authoritative yet collaborative. The interaction model mirrors the backend's LangGraph architecture but presents it through a user-friendly lens.
Generative UI
We moved away from a pure chat interface because we've found this a difficult medium to convey the service's capabilities, the black input box often introduces more confusion than freedom. Instead, we adopted a 'Generative UI' approach. Introducing skills to generate structured data like storyboard and art direction. Presented in interactive React components directly within the chat stream. This transforms the conversation into a dynamic workspace where users can visualize the AI's proposals and refine them in real-time.

Introduces storyboards as part of the chat artifact, aligning the context between what user sees and the evolving prompts agent sees.

Agents expands the chat into interactive storyboards, allowing granular control over the narrative.
Specialized Director Personas
The framework design of each agents meets somewhere between the generative models' capabilities and the creative tasks we assigned. We want to give each agent a distinct 'taste' and domain expertise.

Specialized agents like 'Demy' and 'Avery' handle specific creative domains, preserving stylistic coherence.
Earlier Prototype
We developed an earlier prototype with a more granular human-machine collaboration on context setting. This involved a dedicated art direction loop and feedback cycle, allowing users to define the visual language before any video generation began. This is not included in the current version since the result often struggles to adapt to the evolving context, but we are continuously working on this direction as the models are improving rapidly and we believe this is a critical piece to achieve more consistent and controllable creative outputs.

Visualizing the diverge-and-converge workflow for context engineering.