/shot-specifier

name: shot-specifier
description: >
  Per-shot production specification workflow: takes a completed scene inventory (from
  scene-inventory-extractor-v2) and decomposes every scene into numbered shots with full
  directorial direction — actor position and movement, camera mount and motion, lens,
  lighting setup, practical effects, timing, and clip boundaries. Generates storyboard
  keyframe images via nanobanana, assembles video generation prompts with model routing,
  and maintains an asset pipeline with consistent file naming and a generation manifest.
  Use when a scene inventory exists and the workflow must move from scene descriptions to
  individual, generation-ready clips. Also trigger when the user mentions "shot list",
  "shot breakdown", "storyboard", "video prompt", "model routing", "clip generation",
  or "per-shot direction".

Shot Specifier

Takes a completed scene inventory and reference image set as input and produces generation-ready shot specifications: numbered shots with full directorial direction, storyboard keyframes, video prompts, model routing, and an asset pipeline.

This skill is explicitly **expensive**. Generating storyboard images for every shot in every scene requires many image generation calls. Per-shot video generation is more expensive still. This cost is the price of consistency and production quality — do not skip shots or reduce keyframe coverage to save cost unless the user explicitly requests it.

Input Requirements

Before beginning, verify the following inputs are available:

| Input | Location | Required? | |-------|----------|-----------| | Scene inventory document | `{project}/scene-pack/{project}_scene_inventory.md` | Required | | Reference images | `{project}/image_out/` or `{project}/refs/` | Required | | Prompt keyword library | `{project}/{project}_prompt_keywords.md` | Required | | Continuity inventory | `{project}/{project}_continuity_inventory.md` | Strongly recommended | | Video role manifest | Section in scene inventory | Required |

If the prompt keyword library or video role manifest are absent, run `scene-inventory-extractor-v2` phases 2.4 and 11.6 first before proceeding.

Execution Context

This skill requires:

• **nanobanana image generation Model Context Protocol (MCP)** (see

`references/storyboard-generation.md`), with every storyboard image call explicitly using `model: gemini-3-pro-image-preview`

• **Vision capabilities** (for storyboard consistency verification)
• **File system access** (structured output directories)

Generation runs **silently** — no user confirmation gates during storyboard or prompt phases. Halt only on consistency failures that require human judgement. If `gemini-3-pro-image-preview` is unavailable through nanobanana, or if it cannot accept the required reference images or character-consistency images for the current storyboard operation, **STOP** and report the blocker. Do not continue with a fallback image model.

Workflow Overview

| Phase | Name | Output | |-------|------|--------| | 1 | Input Audit | Verified inputs; flagged gaps | | 2 | Shot Decomposition | Numbered shot list per scene with duration budgets | | 3 | Frame Role Assignment | Video role mapping per shot from reference images | | 4 | Shot Direction | Full directorial spec per shot | | 5 | Storyboard Generation | Keyframe images via nanobanana | | 6 | Storyboard Consistency Check | Vision QA; BLOCK/WARN report | | 7 | Video Prompt Assembly | Full prompts with model routing | | 8 | Asset Pipeline | File naming, generation manifest, tracking |

> **Read order:** Before Phase 4, read `references/shot-direction.md`. Before Phase 5, > read `references/storyboard-generation.md`. Before Phase 7, read > `references/model-routing.md`. If Phase 7 routes any shot to `seedance_2_0`, also > load `seedance-2-deep-dive` before finalizing the prompt, reference plan, duration, > aspect ratio, and generation notes. If Phase 7 routes any shot to `kling3_0`, also > load `kling-3-0-deep-dive` before finalizing shot structure, camera motion, Elements > or Motion Control references, duration, aspect ratio, audio notes, and generation > notes. The asset pipeline conventions are in `references/asset-pipeline.md`.

---

Phase 1: Input Audit

1. Locate and confirm the scene inventory document. 2. Inventory all available reference images and note their categories (characters, locations, props, style anchors). 3. Load the video role manifest. If absent, halt and instruct the user to run `scene-inventory-extractor-v2` Phase 11.6 before proceeding; do not derive a replacement manifest in this skill. 4. Load the prompt keyword library. If absent, halt and instruct the user to run `scene-inventory-extractor-v2` Phase 2.4 before proceeding. 5. Load the continuity inventory if present. Flag if absent — continuity gaps will be carried forward as WARN items. 6. Record the list of scenes to be processed and their target clip counts.

---

Phase 2: Shot Decomposition

For each scene in the scene inventory, decompose it into numbered shots.

Duration Budget

Before writing any shot rows, establish a **duration budget**:

* **Scene ID:** SC-{XX}
* **Scene description:** {brief}
* **Total duration target:** {N} seconds
* **Clip count:** {N} shots
* **Per-clip allocation:** SH001: {N}s, SH002: {N}s, ...

**Budget rules:**

• Clips must be 4, 6, or 8 seconds.
• One clip = one action in one location. If the action requires more than 8 seconds,

split it.

• Multi-beat scenes need multiple clips. A scene with setup + action + reaction is three

clips minimum.

• Establishing shots: 6–8 seconds. Insert/detail shots: 4 seconds. Dialogue beats: 6

seconds. Complex action: 8 seconds.

Shot Row Format

For each shot, record:

#### {S{XX}_SH{XXX}}

| Field | Value |
|-------|-------|
| Shot ID | S{XX}_SH{XXX} |
| Frame size | XW / W / M / CU / ECU / POV / INS / OTS |
| Lens | {focal length} mm {spherical/ana