Generation Skills

The generation category contains six sub-skills that handle the creative planning and intelligence side of video production — from analyzing reference videos to designing shots, building storyboards, managing characters, and orchestrating full production pipelines.

Cinematography

Design shots, scenes, aesthetics, transitions, and video prompts using cinematic language. The core entry point is a 10-dimension interactive shot design protocol.

Key Features

• Shot Design — Interactive 10-dimension protocol for designing individual shots (framing, camera movement, lighting, composition, style, and more)
• Scene Design — Three-layer scene construction for atmosphere and environment
• Aesthetic Control — Lighting, color grading, composition, and visual style refinement
• Transition Design — Shot-to-shot transitions and first/last frame techniques
• Prompt Generation — Convert creative intent into executable AI video prompts using the Subject + Action + Scene + Camera + Style formula
• Advanced Techniques — VFX shots, axis crossing, shot-countershot, complex multi-scene narratives, animation workflows

Foundation Knowledge

Cinematography loads shared L1 principles (cinematic language, prompt architecture, video motion model) and L2 taxonomies (shot types, camera movements, lighting, composition, style catalogs) on demand.

Downstream

Designed shots flow to the API providers for generation, or into storyboards for multi-shot sequences.

Storyboard

Convert stories, scripts, and narratives into structured shot lists with character identity keywords for multi-shot consistency.

Key Features

• Story-to-Shot-List — Break any narrative into a numbered shot list with timing, framing, and transitions
• Character Design — Build visual character definitions and extract 15-25 word consistency keywords that lock identity across shots
• Identity Locking — Embed consistency keywords into every shot prompt so characters look the same across all generated media

Workflow

1. Define characters with visual specs and consistency keywords
2. Break the story into a numbered shot list
3. Call cinematography for detailed shot design per entry
4. Generate prompts with identity keywords embedded
5. Lock character identity across the entire sequence

Downstream

Shot lists feed into the production line for end-to-end execution. Character definitions can be stored in the character bank. Per-shot prompts go to API providers for generation.

Character Bank

Manage a character asset library for consistent AI-generated characters across productions.

Key Features

• Search and Recommend — Find existing characters matching a concept using semantic search (search_bank.py)
• Create Characters — Build structured JSON character definitions with identity anchors (face, hair, body, wardrobe, setting, lighting, constraints)
• Repair and Normalize — Convert raw prompt strings into the canonical structured format
• Multi-Pane Assets — Generate photobooth grids, webcam sheets, turnaround layouts, and cinematic contact sheets
• PromptRef Integration — Append community characters from the PromptRef API with automatic deduplication
• Batch Generation — Generate Daydream multiview images and storyboard video batches via Google Flow pipeline

Character Format

Each character entry is a structured JSON object with fields for subject, hair, body, pose, clothing, accessories, photography, background, vibe, constraints, and negative prompt.

Content Ops

Build and run repeatable AI content production systems. This is the strategy and operations layer — the “what” and “why” of content production.

Key Features

• Brief Template — 7-section framework for complete creative briefs
• Quality Checklist — AI smell detection with 5 QC passes to catch AI-sounding output
• Model Decision Matrix — Task-to-model/tool routing for choosing the right AI tool
• UGC Scripts — 5 UGC script formats with humanization rules
• System Prompts — Reusable Claude role prompts for consistent output
• Content Prompts — Templates for ad copy, email, social, and sales pages
• Learning Path — 14-day zero-to-production system for content creation

Workflow

1. Build the quality filter (agree on checklist)
2. Match model to task
3. Write a complete brief
4. Generate + edit like a human (run QC passes)
5. Set up system prompts for consistency

Downstream

Briefs inform storyboard shot planning. Quality checklists serve as QA gates for all other sub-skills.

Production Line

Pipeline protocol library defining 5 end-to-end production workflows. This is not an independent orchestrator — the top-level creative-engine dispatches chains that reference these detailed protocol files.

Pipelines

Each protocol file defines detailed step logic, quality gates, artifact formats, and examples. The orchestrator manages step sequencing, artifact passing, and session context.

Video Intel

Turn videos into actionable intelligence — either analytical reports or replication templates.

Modes

• Analysis Mode — Fetch a video, extract/generate transcript, run native video understanding (Gemini 3.1 preferred), then produce a structured viral factor report explaining why the video performs
• Reverse-Engineering Mode — Break a video into reproducible components: character design, shot-by-shot construction, pacing mechanics, narrative architecture. Produces a replication template with meta-prompts for the production line
• Video Fetch — Download videos from any platform: Instagram, TikTok, YouTube, Bilibili, and more. Uses fetch-video.py with uv run

Downstream

Analysis reports inform content strategy (content-ops). Reverse-engineering templates feed directly into the video-replication production pipeline.