A prompt becomes a narrated, animated course.

A wizard tuned for instructional design — not "AI vibes."

• Audience + tone + learning-objectives schema enforced by the model — outputs validate against Zod, never freestyle JSON.
• Instructional-design principle picker (Bloom's, Gagné, Merrill, 5E) shapes the script style.
• Every step is editable — undo any AI choice with no penalty.

Ground the AI in YOUR materials.

• Upload PDFs, DOCX, or Markdown source docs.
• "Strict mode" forces the AI to answer only from your materials — no training-data invention.
• Cited inline in the script so reviewers can fact-check.

Narration that knows what it's saying.

• OpenAI + ElevenLabs voices, plan-gated.
• WhisperX forced-alignment marks every word's appearance time, down to the millisecond.
• Cue annotations (`[show: X]`, `[chart: revenue]`) feed the layout AI for synced visuals.

A starting layout — not a blank canvas.

• Layout AI picks a template per scene and assigns each cue to a slot.
• "Redesign scene" gives you an instant fresh look without losing content.
• Manual edits are sticky — re-running AI never overwrites your moves.

Batch the whole course in one click.

• The orchestrator walks every scene through narration → alignment → layout in sequence.
• Live progress UI shows per-scene + per-step status (you saw the moving card on the homepage).
• Per-scene retry — if one step fails, the rest of the course is unaffected.

AI cue annotation pass.

• Once the script is final, an annotation pass inserts visual cues at the right moments.
• Heuristic + LLM hybrid — cheap, deterministic where possible, smart where not.
• You can edit every cue in the WYSIWYG script editor.