Publication day is close, and the farm shop has only scattered notes for its seasonal produce bundle. The work must make sense to nearby families on a phone and lead naturally toward the choice to order a collection. The producer organizes the assignment around fact checking and uses a simple side-by-side choice as the first point of recognition.

The planning sheet translates that search language into its underlying job: to prepare a small or soft source image for several placements. The acceptance test is natural edges at the required size; speed alone is not an outcome. It prevents a broad technology discussion from replacing the work needed for the seasonal produce bundle.

The working brief fits on one page: the farm shop is promoting the seasonal produce bundle to nearby families, and the single communication goal is to order a collection. Tone is fresh and straightforward. Required inputs are approved wording, timing, cost, eligibility, delivery method, and contact path; unsupported results and false urgency are prohibited. Production fields cover exact spelling and case, included and avoided words, sensitive claims, source permissions, a 1:1 master crop, mobile margins, a 28-second limit, formats, owners, and publishing time. Confirmed facts, open questions, and illustrative examples occupy separate columns.

The first writing pass produces four message routes, not a pile of final captions. Four routes expose a simple side-by-side choice, show a three-step method, compare responsible choices, or document one decision at the farm shop. Selection follows the fact checking goal and rejects any route that needs a fabricated customer result.

One route branches into the explanatory post, short caption, hook, carousel sequence, spoken script, and title choices. Each card has one job: identify friction for nearby families, list inputs, demonstrate a hypothetical decision, review it, then invite people to order a collection. A missing price, date, or eligibility detail becomes a question for the owner, never a confident sentence invented by the model. The copy earns space by helping the reader act, judge, compare, or avoid a known failure.

The schedule also reserves enough time for correction, source checks, and another careful final reading before any finished asset is queued. For the seasonal produce bundle, the producer first makes one complete chain: approved message, adaptable composition, connected captions, and a short demonstration. The owner reviews that chain before variants are made, because a weak assumption copied into twelve exports becomes harder to locate and more expensive to repair. The production board records the asset, audience purpose, supporting fact, destination format, approval state, reviewer, and how it helps people order a collection. The learning note defines one observable response, records date range and exposure, and avoids crediting a creative change for movement that the available evidence cannot explain. The result informs the next fact checking decision but is not presented as universal proof, a guaranteed lift, or a customer testimonial. The operator uses descriptive file names, records approval beside the source, and separates superseded exports from current work; this reduces the chance that a polished rejection reaches a scheduled post. Before scheduling, the producer checks the actual derivatives in their destination previews for compression, crop, caption placement, links, and any change to the intended reading order.

The image has one communication task: make the deadline legible without generated type for the seasonal produce bundle. Its prompt names the seasonal produce bundle, places the decision in the foreground, keeps support behind it, limits color to terracotta, sand, and dark brown, and requests 1:1. No essential detail enters the mobile crop or caption-safe margins; real copy is added later as an editable layer. The sample labels 'Made Nearby' and 'Weekend Window' are illustrative examples only; no testing or performance is implied.

A vertical script capped at 28 seconds shows one decision, from loose notes about the seasonal produce bundle to a reviewed asset. The opener uses three seconds: spoken context about a simple side-by-side choice, a short caption, and a tight shot of source notes. Shot two collects inputs for four seconds; shot three shows the illustrative choice for six; shot four pauses on a human correction for five; shot five uses the remaining time for the approved next step. Each shot row keeps narration, screen copy, camera instruction, timing, evidence status, and reviewer distinct.

The assignment's failure watchlist includes invented texture, doubled lines, and oversharpening. A system may also change the subject between frames, deform text or interfaces, miss cultural meaning, imitate a familiar mark, or describe an example as a real result. A second reviewer compares each output with the source, inspects artwork at full size, and reads the script without prompt history.

Core facts stay fixed, while opening, order, pace, composition, and interaction are rebuilt for each destination. On a text feed, the producer explains the choice behind the seasonal produce bundle. An image feed foregrounds the visual task, while the carousel distributes the reasoning. The vertical clip puts a simple side-by-side choice in the first two seconds with large captions; a longer video retains the example and source notes; a community post asks one focused question before inviting people to order a collection.

Human approval starts by matching every claim to confirmed inputs, checking current external rules against reliable sources, recording the access date, and testing audience fit and voice. At full size and on a phone, visual review covers spelling, repeated letters, icons, hands, object count, edges, shadows, interface logic, stray marks, contrast, reading order, safe zones, and every crop from 1:1. Motion review compares shots for continuity, verifies captions and spelling, checks pronunciation and levels, tests pace and safe areas, inspects opening and closing frames, and repeats the viewing muted. Sign-off follows when post, ai image upscaler, and clip agree on evidence, keep the fresh and straightforward manner, and direct nearby families honestly toward the choice to order a collection.

Handoff keeps approved inputs, copy route, working text, layered design, captions, shot table, source notes, and checked exports under distinct names. That record lets a solo operator revise one fact without regenerating every asset or guessing what was approved. For the farm shop, the value is a smaller, reviewable set that gives nearby families one consistent reason to consider the seasonal produce bundle.