DataMagic follows a data-to-video pipeline that separates analysis, specification, scene generation, and preview/editing. This document gives a high-level implementation view without exposing internal source code or prompts.
-
User input The user uploads or pastes tabular data and describes the visualization goal.
-
Data understanding The system profiles columns, detects numeric/time/categorical fields, and summarizes candidate insights.
-
DVSpec planning An AI planner creates a declarative data-video specification with scenes, chart intents, narration, timing, and data bindings.
-
Scene generation Candidate animated scenes are generated from the DVSpec and selected visual templates.
-
Orchestration The generated scenes are assembled into a coherent video timeline with narration-aware timing.
-
Runtime preview The browser preview renders the animated result and supports direct visual edits.
-
Natural-language refinement Users can ask for changes such as chart type updates, wording edits, color changes, or layout adjustments.
- Full Video: Generates a complete narrated data story across multiple scenes.
- Single Chart: Creates one animated visualization quickly from pasted or uploaded data.
The public release materials intentionally avoid exposing full agent prompts, production orchestration code, private template internals, service credentials, and deployment-specific implementation details.