Phase 2 — Style & Identity
Visual aesthetics, FACT and NARRATIVE rules, character sheets, and asset management.
Phase 2 locks how your video looks and who appears in it. By establishing stable character reference sheets and visual style rules before video rendering, your project maintains visual consistency from the first second to the last.
1. Top Controls and Generation Models
The phase toolbar provides key generation controls:
- Character Status Counter: Tracks defined character count against detected characters.
- Image Model Selector: Select Nano Banana 2 Lite (fast and economical), Nano Banana 2.1 (balanced), or Nano Banana Pro (maximum definition).
- GENERATE CHARACTERS: Automatically renders 1:1 canonical reference images for all characters lacking portraits.
- CUSTOMIZE Mode: Upload custom character photos from your local computer or pick existing files from Google Flow's library (PNG, JPEG, WebP).
2. Canonical Character Sheets and Visual References
Each character identified during script analysis receives a dedicated identity card:
- Prompt / Baseline Description: Sourced from the
FICHAscript tag, switchable between AUTO and CUSTOM editing modes. - 1:1 Canonical Portrait: Frontal or three-quarter view, clear facial features, neutral lighting, rendered in the project's NARRATIVE style.
- VISUAL REFERENCE: Attach a photograph, sketch, or illustration to preserve exact likeness across generations without overwriting the character prompt.
- Downstream Anchor: Character portraits automatically anchor appearance in Phase 3 storyboards and Phase 4 video scenes.
3. Style Engine: FACT and NARRATIVE Rules
Visual direction is divided into two distinct style fields:
| Field | Target Scenes | Aesthetic Purpose |
|---|---|---|
STYLE_FATO | B-Roll / Documentary | Establishes technique, texture, and light for scenes without visible character faces. |
STYLE_NARRATIVA | A-Roll / Characters | Establishes identical technique adapted for portrait rendering and human anatomy. |
4. Default Visual Style and AI Suggestions
Projects ship with a validated production style and an automated suggestion engine:
- Default Aesthetic (Graphite Navy Gold): Graphite pencil on off-white paper with navy blue inks and a single gold accent (
#fcc414), full-bleed frame without vignettes or artificial framing. - SAVE STYLE: Compiles updated prompts across all scenes. Existing generated media is preserved and not invalidated.
- SUGGEST STYLE (AI): An intelligent Gemini assistant reads the parsed script to suggest a coordinated FACT and NARRATIVE style pair.
- Enforced Style Rules: Prompts must start with
VISUAL STYLE: <name>., first two sentences cannot exceed 200 characters, both styles must share the same hex accent, and framing divisions or HUD graphic requests are rejected.
5. Style Testing Sandbox
The TEST STYLE tab allows you to enter any text prompt and generate a single 1:1 test image using the current style fields:
- Validates how lighting, line work, and colors render before committing to generating entire scenes.
- Consumes a single image generation credit.
- Test images are automatically saved to your Google Flow library.
6. Assets, Props, and Visual References
The ASSETS & PROPS tray registers reusable reference graphics such as company logos, product shots, specific vehicles, or recurring props:
- Key Generation: Entering an asset name generates a normalized identifier (for example, "Company Logo" becomes
company_logo). - Direct Script Directive: Use
ASSET: keywithin any scene block to bind up to 3 visual assets per scene. - Reference Routing: Assets enter storyboard generations as direct reference images. In video generation, they anchor scenes without storyboards.
Ready to build and launch your SaaS?
Get 100% full source code ownership, no proprietary wrappers, with architecture ready for millions of requests on Cloudflare.