AI video shot list template: plan prompts, assets and review

Use the template below to plan an actual sequence, not a collection of unrelated prompts. The filled example is an illustrative 15-second vertical bottle reveal. Its prompts, timing and work allowances are proposed production decisions; no generation results, acceptance rates or completion times are claimed. Product documentation was checked on October 3, 2026.

Plan the sequence before the prompts

Write one audience, one message and one next action. For this example, a store wants viewers to recognize a particular bottle and open its product page. The video shows the approved item; it does not invent a performance claim, ingredient benefit or customer result. A shot earns its place by doing a different job: establish, reveal a detail, show a wider composition, then provide the next step.

Choose the intended delivery crop and total runtime before preparing assets. Here the planning target is a 9:16, 15-second sequence. Those values describe this brief, not universal social-platform requirements. Put still thumbnails on a rough timeline first. If the order does not make sense without motion, revise the story before spending generation credits. Read any spoken line aloud and time it; words and pictures must fit the same sequence.

Copyable per-shot planning template

Duplicate this record for every shot. The fields belong together even if your team stores them in a spreadsheet or project board. Keep one row per actual shot ID; alternate attempts belong under that ID instead of becoming new story beats.

The edit allocation is not the generation request. You might generate a supported five-second clip and use three seconds from it. Record both values so an operator does not request an unsupported duration merely because the timeline needs a short insert. Choose the usable segment after reviewing its movement; trimming cannot rescue a defect that remains visible in the final cut.

Completed example: a 15-second bottle reveal

Approved inputs: a front packshot named bottle-front-v1.png, a close detail named bottle-detail-v1.png, and a wider tabletop image named bottle-wide-v1.png. These are illustrative filenames, not downloadable assets. All three must show the same actual product, color, cap and label. Make an end card from the approved packshot, brand artwork and verified product-page copy. If you lack these inputs, preparing them is the first task.

The allocations sum to 15 seconds: 3 + 4 + 4 + 4. Generate S01–S03 separately, requesting a duration supported by the chosen model. Select the intended three- or four-second segments in the edit. S04 uses a still held for four seconds, so it needs no generation request. Keep the final four-second hold in the rough timeline rather than using every second for motion and squeezing the CTA into the last instant.

S01: identify the product

Composition: a front-facing bottle with breathing room above the cap and below the base. Motion brief: the bottle stays still while the camera moves gently toward it. Prompt: “Slow, gentle camera push-in toward the bottle. The bottle stays upright in the same place on the table. Preserve its silhouette, cap, label and background.” These are instructions to evaluate, not promises that the generated details will remain exact.

Acceptance: exactly one bottle; no bending cap or changing label; no added objects; sufficient crop for the intended overlay. Review the entire selected segment. If the camera movement changes essential artwork, use the approved still with a simple editor-controlled scale move. That fallback preserves the shot’s recognition purpose without making the next two shots depend on a lucky generated frame.

S02: show a detail without inventing a feature

Composition: the approved cap and upper label at a similar angle and lighting direction. Prompt: “A small, slow camera slide across the cap and upper bottle. The bottle remains still. Finish with the cap and neck fully visible.” Do not ask for a 360-degree orbit if you only have one view and exact hidden details matter. The shot shows visible design; it does not prove how the cap works.

Acceptance: cap remains attached; neck proportions stay recognizable; label does not become a new design. Choose a segment whose motion can be read in four seconds. If it reveals an invented side of the product, reduce the move or use an approved detail still. Record that change in the shot list; do not silently replace the product detail with an unrelated attractive close-up.

S03: return to the whole item

Composition: the same bottle on the wider tabletop image, with lighting compatible with S01. Prompt: “Slow camera push-in toward the single bottle on the table. The bottle and tabletop remain stationary. Keep the composition simple.” Returning to a wider view lets the editor finish the detail beat and prepare the end card without adding another product action.

Acceptance: one unchanged bottle; tabletop lines remain stable; no moving props; the selected ending supports a clean cut to the end card. If the wider input differs in label or color, repair that source before generation. If the result is acceptable alone but clashes with the first shot, inspect exposure and crop on the assembled timeline rather than regenerating every shot.

S04: make the action readable

Use the approved packshot and exact text “Explore the product,” or the real approved CTA for your brief. Build this card in your editor, keep it static for the planned four seconds and check it at delivery size. Do not generate the wording inside a moving scene when exact text is the deliverable. Verify the associated post or landing link separately; a readable card does not establish that the destination is correct.

Set a practical work allowance

For this illustrative brief, reserve 15 active minutes for asset checks and the rough timeline, 20 for preparing and reviewing the first motion attempts, 20 for focused repairs, 15 for assembly and 10 for final inspection. That totals 80 active minutes. It is a planning allowance, not a measured production time. Render and queue waits are additional and can overlap with other preparation; do not represent them as guaranteed turnaround.

If repairs consume their allowance, choose the planned still-image fallback, simplify the motion or adjust the deadline. Do not keep generating until the schedule disappears. Record actual active work and attempt counts afterward so the next plan can use your own evidence. For monetary budgeting, record the selected settings and the current quote per attempt; keep that separate from the editor’s seconds used.

Produce one hard shot first

Begin with S02 if the close-up detail is the likely failure point. It tests whether the product can survive the intended camera move. Prepare that asset, enter the prompt, choose supported settings, and inspect the complete output. If the detail fails, improve its source or simplify its motion before generating all three shots. This is a proposed sequencing choice for the example, not evidence that any model passed the shot.

For the approved packshot shots, open Magic Hour Image-to-Video, upload the exact row’s image and paste that row’s motion prompt. Record the selected model, requested duration, resolution and audio decision. The current creator exposes an image upload and motion prompt; available settings vary. Generate and download individual clips, then assemble the sequence in your editor. Do not expect the generator to build this four-shot edit automatically.

Magic Hour’s product-video recipe also documents generating one clip from a product photo and downloading the completed result. It warns that labels, logos and geometry can change. Use its linked instructions if your production route is the API, but keep the same shot IDs and acceptance criteria. This article’s completed example uses a browser handoff and does not require API setup.

Assemble with the shot list beside the timeline

Import the accepted files using shot IDs, then place them in order at the planned ranges. Choose each segment for its intended action and usable detail; do not automatically take the first few seconds. Keep rejected attempts outside the active timeline. Start with clean cuts so you can judge the sequence before adding transitions that may hide a jump or consume an important beat.

Add the approved copy, end card and any authorized music or voiceover in assembly. This example needs no spoken dialogue, and generation audio can be omitted when the selected tool supports that choice. If you add narration, time the actual recording and revise the allocation rather than speeding speech to rescue an overcrowded plan. Inspect labels and overlays at the exported crop and size, not only in an enlarged preview.

Watch the complete 15-second sequence with sound and again without it. Confirm the product stays recognizable across cuts, each beat has a different purpose, and the CTA has its planned reading time. Then inspect the exported file; a timeline preview is not the delivered artifact. Put the accepted file, chosen source versions, prompt settings and review note back into the shot record.

When the plan breaks, change the responsible row

A source is missing: mark the shot blocked and prepare the asset or choose a truthful substitute. Motion damages a product detail: simplify that shot or use its still fallback. An insert feels redundant: remove its beat and redistribute the edit time. The CTA is unreadable: revise the end card or allocate more time. None of these requires rewriting every prompt or moving acceptance criteria after seeing an attractive result.

If changing one shot alters continuity, inspect its neighbors. A new camera angle or brighter background may need a different cut even when that shot passes on its own. Keep the shot ID stable, name the changed asset or setting, and review the revised sequence. More than one production route is acceptable when each serves a defined purpose and the edit remains coherent.

Keep this template focused

Use the shot list to coordinate story, assets, motion and assembly. A prompt library helps write individual instructions; an output-QA checklist helps inspect delivered media; a cost worksheet prices attempts; an endpoint guide designs a transition between two images. They answer related questions, but none replaces the per-shot record and timed sequence here. Start with the small plan, fill every row that affects production, and omit fields your actual project does not use.

Official sources and product references

https://docs.magichour.ai/integration/evaluating-media-apis

https://docs.magichour.ai/get-started/starter-recipes