Test the idea.
Learn before scaling.

1. Give the test one decision to resolve.

A promising AI clip can make a campaign feel possible. It does not necessarily tell a producer whether the next ten shots can be delivered, whether the product stays accurate or how much finishing remains. A useful shot test has a narrower job: resolve one uncertainty before more production depends on it.

Begin with a question whose answer changes the plan. “Can we make something cinematic?” is too broad. “Can this approved image support a small fabric movement while the product detail stays readable?” gives the team a concrete decision. If it works, the shot can move forward. If it does not, the team can revise the source, simplify the action or choose another method.

The acceptance question belongs beside the creative brief. A technically stable clip can still feel lifeless; an expressive clip can still misrepresent the jacket. Keep both considerations visible rather than allowing the most attractive preview to settle every question at once.

Ordinary Folk, a Vancouver motion studio, describes a process that moves from message through design and storyboards into animation. Its published process is not an AI testing method. The relevant lesson we draw from it is to establish the visual foundation before asking motion to solve the whole project. Source: Ordinary Folk’s process.

2. Freeze the starting point and the review conditions.

Choose one source image and give it a version name. Confirm that it already contains the approved jacket shape, pose, crop and lighting. Record the intended output dimensions and duration. If the source is still changing between attempts, it becomes difficult to tell whether a different result came from the prompt or from the image.

Runway’s image-to-video guide describes the input image as the source of composition, subject, lighting and style, while the prompt describes motion and temporal progression. It recommends starting with the most important motion and adding detail as needed. This is vendor guidance for its documented workflow, not proof that the fictional shot will succeed. Source: Runway’s image-to-video guide.

For Merefield, the proposed baseline is a single approved still, a five-second duration where supported, one aspect ratio and one recorded model version. Keep those choices fixed during the first comparison. If the tool does not support a requested duration or setting, record the actual value instead of presenting the intended one as what was used.

Agree the viewing conditions too. Use the same playback size, speed and crop for every candidate. Review the picture without an added soundtrack initially, so a musical accent does not make one candidate seem more successful for unrelated reasons. Later, test the selected candidate in the intended edit with sound.

Check the source for motion cues. A jacket already stretched by a strong gust may conflict with the idea of barely perceptible movement. Runway’s guide illustrates how implied motion in an image can conflict with a requested action. That example is a vendor demonstration; it supports inspecting the starting image, not assuming that any source can be made to behave identically.

3. Design three passes with a clear purpose.

The following passes are proposed briefing variants, not tested prompts or a recommended spend. Agree the number of attempts and the review point before generating anything. Three passes mean three questions; each pass may need more than one candidate to understand whether an attractive result is repeatable enough for the project.

Start with A. If the jacket’s construction fails there, do not automatically proceed to a more complex camera move. The baseline has already exposed an issue worth resolving. The next useful action might be changing the source asset or reducing the motion requirement rather than generating all three planned variants.

B changes the intended fabric movement while retaining the other brief elements. C is a separate creative decision after the fabric direction has been chosen. This sequence is a practical production exploration, not a laboratory experiment claiming perfect isolation of every variable. Generated outputs can differ in ways the prompt did not request.

Do not combine a new source, stronger wind, a different model and a camera orbit in one comparison and then attribute the result to a single prompt phrase. If production constraints force several changes at once, describe the result as a new setup. Honest labelling makes the findings more useful to the next person.

4. Record usable evidence, including failures.

Keep a simple candidate record. Capture the source version, exact prompt, model name and version when exposed, duration, aspect ratio and settings actually available. Record a seed only if the tool provides one. An absent setting should be marked unavailable, not reconstructed from memory or invented for the sake of a complete-looking spreadsheet.

Give each output an identifier and retain unsuccessful candidates where practical. If a review deck contains only the strongest clip, it can show the creative possibility but cannot show how reliably that possibility was reached. Those are different claims, and a production decision may need both.

In the fictional example, a useful observation would be “the jacket hem moves as intended, but the fastening changes shape near the end.” That is more actionable than “almost there.” It identifies a trade-off and a time range to inspect. It does not yet establish whether a repair is viable.

Report the actual number of candidates reviewed if the test is run. Do not infer a success rate for a whole tool or future campaign from a handful of outputs. A small production test can reveal a workable route or a recurring problem in this setup; it cannot establish universal model performance.

If an output needs cleanup, retain both the raw and finished versions and document what changed. Only then would a before-and-after comparison be supported. This morning’s retouching decision guide explains how to separate a local repair from a change that requires a new base image.

5. Review against requirements before preferences.

Separate mandatory requirements from creative preferences. In the Merefield plan, a recognisable jacket silhouette and correct fastening are requirements. The amount of fabric movement and the energy of a camera approach are preferences within an acceptable range. An expressive candidate should not override an essential product constraint.

Watch each candidate at normal speed first. Record the overall impression without immediately scrubbing for isolated faults. Then inspect the relevant details over time. A jacket can look correct in the opening frame and drift later; a frozen screenshot cannot establish temporal stability.

Check whether the usable material is long enough for the actual edit. A five-second output with only a short acceptable section may still be valuable for a quick cut, but it does not satisfy a five-second continuous-shot requirement. Record that distinction rather than labelling the entire candidate a pass.

Have the creative and product reviewers look at the same candidate versions. Ask each to explain the reason for their judgement. If one prioritises atmosphere and the other accuracy, the disagreement may be about the brief rather than the quality of the work. Resolving that before full production is part of the test’s value.

The source-readiness guide helps decide whether the starting image is suitable. The camera movement guide helps define a move. This test plan adds a different layer: a record of what changed, what happened and what decision the evidence supports.

6. Finish with a decision and a fallback.

End the review with one of three outcomes: proceed with the tested setup, revise a specific part of the setup, or use another production route. “Keep trying” is only useful when it includes a new hypothesis and an agreed limit. Otherwise the test becomes open-ended production without a clear definition of completion.

Before scaling the approach, identify what the test did not cover. A single jacket shot does not establish consistent performance across different garments, faces, backgrounds or delivery crops. Choose the next difficult case deliberately rather than assuming the easiest successful frame represents the entire campaign.

A useful handoff includes the approved source, selected output, exact settings that were known, review notes, remaining finishing and the fallback. This gives the producer a basis for estimating the next stage and gives the client a clear account of what has actually been demonstrated.

The purpose of a shot test is to reduce uncertainty. A beautiful result is welcome, but a well-documented reason to change direction can be equally valuable. Both help the team commit to a production route with a clearer understanding of its limits.

Sources checked 22 October 2026. Merefield is imaginary and the three-pass plan has not been run. Runway guidance is attributed to the vendor; Ordinary Folk’s process is an external studio reference. No Nightwolf tool usage, generated comparison, client commission or performance result is claimed. Explore imagery is used for independent visual annotations.