CastingBox - AI Video Creation Workspace

Project Overview

CastingBox is a domain-specific AI product for video and short-form content creation. It is designed as a production workspace rather than a single image generation tool, bringing creative communication, script structure, asset management, storyboard generation, and future video production into one continuous workflow.

The product is moving toward a chat-first creation model: users describe an idea naturally, the agent interprets intent, and the workspace turns that idea into reviewable scripts, reusable assets, storyboard boards, and generation records.

Figure 1. Asset Studio workspace for storyboard assets and generated boards

Product Positioning

CastingBox is built for creators who need repeatable production quality across episodes, boards, characters, scenes, and props. Instead of leaving prompts, files, and generated results scattered across tools, it keeps the creative structure and generated media inside a shared workspace.

• Use chat as the entry point for story ideas and production intent
• Structure content into Season, Episode, and Storyboard Board layers
• Extract reusable character, scene, and prop asset requirements
• Bind generated or uploaded assets back to specific storyboards
• Preserve generation history, parameters, scripts, shots, and review state

Core Experience

The current product experience centers on Asset Studio and Storyboard. Asset Studio collects generated visual assets and storyboard outputs for a workspace, while Storyboard manages the episode script, structured shots, asset requirements, board generation, and review flow.

Figure 2. Episode-level storyboard workspace with script, board generation, and review state

Figure 3. Generated assets and storyboard outputs retained in the workspace

Workflow Model

CastingBox is organized around a structured creative production loop:

• A creator describes an idea in natural language
• The system turns the idea into a script, shots, and asset requirements
• Characters, scenes, and props are generated, uploaded, or bound from existing assets
• Storyboard boards are generated from confirmed structure and assets
• Users review, adjust, replace, and approve outputs before moving forward
• The workspace keeps assets and generation history available for future episodes

This workflow keeps human creative control in the loop while giving agents a structured place to write back results.

Technical Foundation

The current product is implemented as a real web workspace using Next.js App Router, React, tRPC, Drizzle ORM, PostgreSQL, shadcn/ui, Radix UI, Tailwind CSS, object storage, and an AI image generation pipeline.

The main product modules include:

• Asset Studio for workspace-level generated assets, storyboard outputs, and creative history
• Storyboard for Season, Episode, Storyboard Board, scripts, shots, requirements, and board generation
• Digital Asset Management for uploaded assets, generated media, metadata, and future search or reuse flows
• Generation Records for preserving scripts, asset bindings, storyboard boards, and generation runs

The next direction is deeper chat-first authoring, external agent and MCP tool access, video generation, richer asset relationship graphs, team collaboration, and version comparison.