Story to Screen: The AI Film Production Framework

You have IP and a verified stack, and every instinct says start generating. This module stops you on purpose. You'll build the discipline layer instead: the three Day-1 canon documents (Vision Lock, Discipline Brief, Deviations Log) and the session laws that keep them alive, learned the expensive way on a real production. You walk away holding a complete, citation-audited Vision Lock for your own IP. That document is the creative constitution every later module gates against.

Module 1 gave you the documents that say what your film IS; nothing yet stops a paid generation from contradicting them. In this module you install the enforcement layer: the five-field pre-shot gate, the six-lock pre-scene gate (Rule 14), and the three-agent QC system with its pixel-evidence standard and red-team pass. You walk away holding your Day-1 gate stack. A Production Discipline Brief, an initialized Deviations Log, and your first two completed gate sheets are the artifacts every fire from Module 3 onward is graded against. The gate takes 60 seconds; the production this course is built on lost full days every time it skipped one.

This is the module where you finally spend credits. It is also exactly where an ungated production bleeds them. You'll learn how a still fires as a gated parallel pair with every parameter explicit, why everything that must match ships as pinned pixels while prose carries only what is new, and how the catalog becomes your production's source of truth. No working doc ever earns that status. You walk away holding your own BASELINE/_CATALOG/ with five gated, approved stills inside it: the artifact every later module pins from.

Everything you built in Modules 1-3 was preparation for motion. Motion is where the reference production paid its single most expensive lesson: nine failed attempts, across two tool architectures, to make a video model perform a state change by describing it. In this module you declare your film's structural path before anything fires, learn why a wash is interpolation succeeding at its actual job, and master the two techniques that execute state changes correctly. One is a masked grade. The other is authored keyframes off one shared base. You will walk away holding a complete per-scene brief with a logged 4-pass gauntlet trace for every motion prompt in it.

Every module before this one taught you to make the right thing. This one teaches you to control when money leaves the account and to tell the truth about that spend afterward. You'll study the ~17.5-credit unauthorized fire that wrote the GO-token and DRAFT-materials laws, then install both in your own pipeline. You walk away holding the two documents that govern every paid fire from here to the capstone: a completion ladder of individually priced, individually authorized rungs, and a cost ledger where every number carries RECORD, DERIVED, or UNKNOWN.

Audio begins with architecture, not generation. Break down a locked 60-second sequence into three separately owned lanes for VO, score, and SFX. Then prove that each cue has one job, Digital_One remains system sound rather than a speaking character, and silence is written onto the timeline as deliberately as an impact. The finished artifact is the contract every later audio module will execute.

A convincing voice track begins with the cut, not the voice tool. In this module you lock a short sequence, slate every line to a named clip, write one versioned voice-canon entry, and plan connected emotional-arc renders that split into one dry file per line. You also resolve a real source conflict: a broader registry contains a Digital_One voice ID, but Chapter 1 canon O-032 routes that presence to system sound and never to spoken VO. The result is a VO timing and canon pack that an editor can place without lip sync, baked effects, or guesswork.

Build score one cue at a time. Define the picture function, carry forward the score's shared system, freeze exact style and exclude fields, and plan a comparable 3–4-take audition that rejects palette drift before preference. The production's exact THE COST and RUN / TITLE prompts show the authored standard; their pending phase status teaches equal discipline about evidence.

Sound effects are built separately, then meet score and voice only in the edit. In this module you author a two-layer D1-style effect, plan a varied world bed from four short takes, reverse a generated dissolve so sound rhymes with picture, and place one VO line on its named clip in the four-track V1/A1/A2/A3 map. You also protect the film's loudest decision: the blast-door hit that kills every bed and turns silence into an edit event. The result is an SFX mix map an editor can execute without guessing.

You arrive with approved stills, approved clips, and a finished audio pack. None of it is a film yet. Assembly is where locally-good assets go to die globally, unless you own the seams. In this module you gate every scene boundary with the four carry fields, run a post-cleanup log that fixes blemishes deterministically with zero re-rolls, install the two meta-laws that govern you when a step keeps failing, and retire superseded assets in writing. You walk away holding a shipped cut and a living failure ledger that rides into every future project.

Ten modules ago you wrote a Vision Lock; now you own a complete canon stack, a gated pipeline, a failure ledger, and a shipped cut. This capstone proves those things belong to you and not to the project you built them on: you sort the framework into what carries and what resets, bootstrap a complete second IP in one sitting, and learn how a local agentic pipeline turns the human disciplines of Modules 1-10 into machine-enforced preflight checks that keep your hands on the gates. You walk away holding NEW_IP_BOOTSTRAP/, a graded second-IP framework folder you can re-run on any project you will ever touch.