After the First Good Frame

We propose a revision ledger for AI-assisted creative production. It records what each attempt should change, what it must preserve, and why the result was kept or sent back.

A promising frame still has to move, hold together across a cut, and survive revision. At CULTESC, we want those decisions to leave useful evidence about creative control and the effort behind finished work.

Define the change and what must stay fixed.

Write the request and its preserved constraints: “Hold the pause longer; keep the silhouette, camera, and action.” This gives unintended changes a clear reference.

Explore a direction, then record the chosen references. A reference lock gives later revisions a defined starting point. Whether locking earlier saves effort remains untested.

Set acceptance criteria for the approval at hand: a look, a usable shot, or a finished sequence. Motion, timing, sound, and the surrounding edit need their own review.

Separate stage from intent.

Stage identifies the work: reference preparation, exploration, performance, assembly, edit, sound, captions, or delivery. Intent records why:

Explore

Find a direction or test an option before the answer is settled.

Repair

Resolve an identified mismatch against the agreed brief.

Polish

Refine a result that already meets its current criteria.

Assign intent before judging the result. An unused exploration may have done its job; a repair may introduce another problem.

Record each reviewed attempt.

Use one row per reviewed candidate or manual revision. Link related attempts, keep the counting rule fixed, and allocate shared preparation, batch costs, and labor once.

What an attempt needs to tell us

Record	Make it specific
Identity	Piece, shot, attempt, operator, and reviewer IDs.
Purpose	Requested change, preserved constraints, stage, and intent.
Setup	Tool, visible model/version, and input or reference IDs.
Effort	Active human minutes, with waiting recorded separately.
Consumption	Quantity and unit; attributable cash spend if known.
Decision	Reject, partial, or accept; approval scope and reason.
History	Parent attempt, defect codes, preview milestone, and any reopening.

Active minutes record hands-on effort. Rendering can overlap with editing another shot; allocate that person’s time once. Keep currencies and different vendors’ credits separate, and label subscription allocations.

Method and sources

This is a proposed logging method, not a validated efficiency measure. It establishes no finishing-cost share, failure frequency, or tool ranking. Audience recognition and response require separate evidence.

A preprint on professional editors’ evaluation of generated video ads informs the emphasis on explicit review criteria. Its editing-quality dimensions do not validate this ledger or measure production effort.