Speak as the owner. Let the useful features and practical tradeoffs shape the story.Editorial summary of the saved direction
Elyse / behind the films
Made with AI.
Shaped by people.
Twelve films.
A very human creative process.
The first idea, the awkward takes, the small corrections. An open record of the work behind Elyse’s videos.
Follow the process ↓Loading the preserved films…
An open record of making things with AI.Production snapshot / 14 Nov 2026
01 / the outcome
Start with the films.
Six owner reviews and six Standout films. The exact selected exports, with the earlier creative work preserved below.
Elyse 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, stills, complete films, and the return to the original voice. Open a checkpoint to see what was preserved at that stage.
Loading the production record…
14 Nov / the six final owner reviews
“all good. ready for posting; give me prompt for the posting agent to find them”
The selected original-voice films remain unchanged. Voice experiments and superseded candidates remain part of the history.
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, pencil boards, and the reference portrait. Select a collection or 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.
A clamp, a bowl, a screen, a cable. A small mismatch can change the whole film.Editorial summary of the visual reviews
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
—
| Model / service | Requests | Completed | Reserved USD |
|---|
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.
| 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.