Cine-Ops Pro

The Problem

Most AI image and video prompt resources are collections — a thousand example prompts in a Google Sheet, a Notion of vibes, a Twitter thread of "secrets." They scale linearly. Add a prompt, get a prompt. Nothing compounds.

And nothing in them protects the subject. Drift the prompt by two adjectives and the face you wanted shifts into someone else. The industry calls this "AI garbage." It's actually unconstrained generation. Different problem, different fix.

The Insight

Real cinematography is already codified. An Arri Alexa 35 with a Cooke S4 produces a particular look — 17 stops of dynamic range meeting the warm Cooke rendering. A RED V-Raptor with Arri Signature Primes produces a different one. A Sony Venice 2 with Panavision C-Series anamorphics produces a third. The physics is the protocol.

If you encode that physics as JSON, the prompt formula becomes a pure function. Same inputs, same output. Every prompt is reconstructable from a six-field payload. That's not a collection. That's a protocol.

The Formula

{shot_type} of {face_core_positive},
{theme_prompt}
{camera_token} with {lens_token},
{lighting_token},
{composition_token},
{color_grading_token},
{aspect_ratio} aspect ratio.

Eight inputs, one deterministic string. Round-trippable. Signed with ed25519. Verifiable in the browser. The same builder powers the web dashboard and the MCP server.

The Reusable Asset: Subject Fidelity Pattern

The negative prompt block is not a styling trick. It is a governance pattern: a small JSON schema that pins identity attributes the generative model should not drift. We call it the Subject Fidelity Pattern.

C2PA protects authentic media downstream. Subject Fidelity protects identity upstream, at the generative layer. Different parts of the same provenance stack. Cine-Ops Pro is a reference implementation of the upstream piece.

The schema is open. It lives under the Kinetic Gain Protocol Suite at cineprompt.spec.json. Anyone can implement it. We just shipped the first one.

The Stack

The Numbers

89 themes. 6 camera bodies. 6 lenses. 30 lighting presets. 19 composition rules. 13 shot types. 8 color grades. 4 aspect ratios. The working permutation space is approximately 89 × 6 × 6 × 13 × 30 × 19 × 8 × 4 ≈ 76 million unique deterministic prompts before any user-supplied Face Core. With a user-supplied Face Core, the space is unbounded.