Skill: Model Allocation (Anchor-Chain Economics)

What

Use expensive models at anchor points, cheap models for everything between. Quality propagates through the chain.

The Pattern

PRO → flash → flash → flash → PRO (complex event) → flash → flash → PRO (payoff)

When to use PRO (expensive)

Object references — ALWAYS. These are visual DNA referenced across the entire sequence. Spend here.
Start frame (panel 01) — sets the world. Lighting, room, perspective, style. Everything chains from this.
Complex compositions — multiple subjects interacting, 3+ characters in frame, technical difficulty.
Emotional reveals — close-ups that need to land. Baby's face. A character's expression.
Technical difficulty — hands holding objects, phone screens, reflections, selfies.
New environment — first time seeing a new space. Establish it properly.

When to use FLASH (cheap)

Transitional movement — character walks one step forward, turns head, reaches for something.
Object tracking — following a character's movement across sequential frames.
Hand tracking — simple hand movements (reaching, pulling, gripping) that chain from previous.
Any "Same scene" delta — the model sees the previous frame's pixels. It just needs to know what moved.

Why It Works

Expensive frames produce high-quality visual anchors
Frame chaining feeds those pixels to the next frame
The cheap model INHERITS quality — it's not generating from scratch, it's modifying what it sees
Cost follows complexity, not sequence position

Cost Math

Pro: ~$0.10–$0.15/image
Flash: ~$0.04/image
14-panel scene: 6 Pro + 8 Flash = ~$1.16 (vs ~$1.96 all-Pro)
Savings: ~40% with minimal quality loss on transitions

Scene Plan Annotation

Mark every panel in scene-plan.md with PRO or FLASH and a one-line "Why":

| 07 | PRO | 3 characters composed together |
| 08 | FLASH | Transition |

The Pattern

PRO → flash → flash → flash → PRO (complex event) → flash → flash → PRO (payoff)

When to use PRO (expensive)

Object references — ALWAYS. These are visual DNA referenced across the entire sequence. Spend here.

Start frame (panel 01) — sets the world. Lighting, room, perspective, style. Everything chains from this.

Complex compositions — multiple subjects interacting, 3+ characters in frame, technical difficulty.

Emotional reveals — close-ups that need to land. Baby's face. A character's expression.

Technical difficulty — hands holding objects, phone screens, reflections, selfies.

New environment — first time seeing a new space. Establish it properly.

When to use FLASH (cheap)

Transitional movement — character walks one step forward, turns head, reaches for something.

Object tracking — following a character's movement across sequential frames.

Hand tracking — simple hand movements (reaching, pulling, gripping) that chain from previous.

Any "Same scene" delta — the model sees the previous frame's pixels. It just needs to know what moved.

Skill: Model Allocation (Anchor-Chain Economics)

Additional Files (20)

Skill: Model Allocation (Anchor-Chain Economics)

What

The Pattern

When to use PRO (expensive)

When to use FLASH (cheap)

Why It Works

Cost Math

Scene Plan Annotation

Related Skills

<h1 align="center">

Frontend Typescript Linting.mdc

2. Apply Deepthink Protocol (reason about dependencies

Additional Files (20)

Skill: Model Allocation (Anchor-Chain Economics)

What

The Pattern

When to use PRO (expensive)

When to use FLASH (cheap)

Why It Works

Cost Math

Scene Plan Annotation