An open record of making things with AI.Production snapshot / 17 Nov 2026
01 / the outcome
Start with the film.
The exact showcase export, followed by the scripts, assets and revisions behind this film.
Scout is an AI-generated virtual reviewer. Posting links reflect preserved receipts; this journal does not check current TikTok visibility. “Accepted” refers to the owner’s film selection.
02 / the process
The path was iterative.
Scripts, artwork, intermediate cuts and source motion. Open a checkpoint to inspect the preserved evidence.
Loading the production record…
The current export
Publication recorded · 16 Nov 2026. Owner acceptance does not waive any unresolved automated flags.
The script history +
All stored revision identities are counted. Consecutive revisions with identical script and scene content are grouped; every distinct text and visual direction remains readable.
03 / the ingredients
The details are the work.
Generated keyframes and the product and character references used by this film. Select a round, then open an image to inspect its source identity.
+
The production stack Generated stills and boards establish the scene. Recorded requests include image-to-video motion, speech, transcription and audio-driven performance. Local editing and assembly connect those pieces. Exact model identifiers appear in the request ledger.
Raster artwork uses optimized viewing copies; the recorded SHA-256 identifies the preserved original. Reused artwork remains visible in the rounds that used it. Full films are byte-identical exports.
04 / the human input
The notes that shaped it.
Story direction, marked frames, full-film corrections and the responses they prompted. Public notes are lightly edited for tone and privacy where needed.
This is the saved human review record from the production stores and review files. Agent QA and synthetic browser-test comments are excluded. It is not a verbatim export of private chats.
05 / the costs
What the record can tell us.
Each provider receipt counted once. Completed, unsuccessful and uncertain requests stay distinguishable. Reservations are estimates, with final billing unverified.
Recorded reservations
—
Recorded provider requests and reservations by model
Explicit pre-render refusals are retained with $0 counted in the reservation total. Other unsuccessful or uncertain requests retain their reservation until billing is reconciled. Built-in image generation, orchestration, hosting, and missing receipts are unpriced here.
06 / the analytics
Where the effort goes.
Production evidence supports request counts, preserved revision counts and some provider completion intervals. Those measures have different meanings.
Observed provider completion intervals
Model / service
Timed requests
Mean observed interval
Sum of intervals
Submission-to-observed-completion intervals include provider queues and polling. Calls overlap. Summed intervals are request workload, not elapsed production time, model inference or measured human effort. Missing timing data remains missing.
The next pass
Catch the repeatable errors.
Exact source selection, file identity, complete narration, sound continuity and physical dimensions can be checked against saved evidence before review. Visual judgments about geometry, mouth motion and continuity still need observed results.
The human part
Keep intent in the loop.
Voice preference, story premise and the decision to accept a film are owner choices. Local-model cost savings and automated correction rates require a matched benchmark; this journal does not turn request counts into claimed labor savings.