Phase 1 — Script

Phase 1 establishes the structural backbone of your video by parsing a written script into individual 8-second scenes, detecting characters, lines, and audio cues without consuming AI generation credits.

1. Top Controls and Targets

The upper toolbar configures pacing and scene constraints:

• Duration Target: Select AUTO or set a fixed duration of 1, 2, 3, 5, 10, 15, 20, or 30 minutes.
• Scene Target: Sets the scene count from 1 up to 225 scenes (225 scenes corresponds to a full 30-minute video).
• Speech Pacing: Calibrates the allowed word count per scene to guarantee dialogues fit natural timing (Standard, Moderate, Fast).

2. Script Editor and Deterministic Parsing

The SCRIPT tab provides a direct text editor:

• ANALYZE SCRIPT: Performs an instant, rule-based structural analysis without contacting Google Flow or spending credits.
• Detected Elements: Automatically extracts project title, video language, scenes, characters, character sheets, lines, locations, SFX, and assets.
• Built-in Starter Template: New blank projects include a validated 1-scene reference template in your interface language.

3. Script Syntax and Scene Structure

A valid script starts with a global header followed by numbered scene blocks:

4. Scene Styles: FACT vs NARRATIVE

The engine categorizes each scene into one of two visual modes based on the AMBIÊNCIA tag:

• FACT Mode (B-Roll / Documentary): For scenes with no visible character faces (landscapes, objects, documents, architecture, anonymous silhouettes, hands). Narration voice-over is standard; character lip-sync is disabled.
• NARRATIVE Mode (A-Roll / Character Scenes): For scenes centered on characters with visible faces, requiring character sheet anchors and natural lip-sync synchronization.

5. Speech Pacing and Word Boundaries

The speech pace setting governs the allowed word count per 8-second clip, ensuring that lines never cut off prematurely:

6. Non-Blocking Validation Warnings

When script lines deviate from production guidelines, the scene card highlights specific advisory warnings:

• Colors specified in HEX, render styles, or requests for on-screen graphic text in action lines.
• Missing camera shot definitions or multiple cuts requested within a single continuous scene.
• Transient emotions or temporary facial expressions mistakenly placed in character baseline sheets.
• Word count outside the selected speech pacing bracket or quotation marks in spoken text.

7. AI Story-to-Script Generation

If you only possess a raw article, story, lesson, or theme, open the ENTER STORY tab:

• Input Capacity: Paste up to 30,000 characters of raw reference text.
• Gemini 2.0 Flash Generator: The Flow text model divides the text across scenes in batches of 10 scenes per call.
• Automatic Word Auditing: If generated dialogue exceeds the target pacing limit, the system runs up to 2 automated repair rounds per batch.

8. Non-Destructive Reanalysis

Editing script text and clicking ANALYZE SCRIPT preserves already-rendered media for unchanged scenes. Only modified scenes receive the amber OUTDATED label.