Plan an AI Video Storyboard Before Generating Clips

Give each shot one job

Plan the sequence before generating clips. Define what the viewer must understand, assign one purpose to each shot, and decide how accepted clips will be joined in an editor. A storyboard connects the script, source images and motion prompts; it does not guarantee a model can generate the entire sequence in one request.

This guide provides an editorial planning template, not a tested model comparison. In VidVid, use the duration, aspect ratio and other options actually offered by the selected model. The timing in your storyboard is an editing target and may require trimming generated clips.

Build a three-shot product sequence

The following example uses a fictional reusable bottle and makes no product-performance claim.

Shot	Purpose	Source and motion	End and acceptance check
01: establish	Show the complete product	Approved full-bottle image; slow camera approach	Full silhouette remains visible; cap shape is stable
02: detail	Show the cap design	Approved close-up image; restrained lateral camera movement	Cap and neck stay recognizable; no invented parts
03: close	Create space for the final message	Approved wider composition; gentle camera pullback	Clear area for a headline added in the editor

Choose shot lengths after checking available generation options and the final edit. If a detail shot needs a different angle, prepare a suitable source image; do not assume a camera instruction can reveal an unseen side accurately.

Keep an asset record beside the storyboard

Record the source image ID, permission status, fixed subject attributes, intended crop and approved version for every shot. Use meaningful names such as bottle-front-A and bottle-cap-A. Keep a shared description of the subject across shots, and record deliberate changes in angle or lighting.

Prepare stills with the image prompt and composition guide, then inspect them with the reference-image preparation checklist. For a person or recurring character, use the character consistency guide to define identity checks. A shared description helps review continuity; it does not guarantee matching identities across separate generations.

Write a motion brief for each shot

Use this structure as a short planning brief:

Shot 01. Approved full-bottle reference. Slow camera approach toward the stationary bottle. Soft background movement only. End with the whole bottle still visible and space above the cap. Review the cap, silhouette and background edges throughout the clip.

Move acceptance checks into your review notes if the generation prompt becomes crowded. The model prompt should emphasize a coherent action. Avoid combining a rapid orbit, a zoom, object rotation and a scene change before you have a usable baseline.

Runway’s image-to-video prompting guide recommends starting simply and distinguishing subject, environment and camera movement. This informs the brief structure; its model-specific controls are not a promise about every VidVid model. See the video prompt guide for individual motion prompts.

Decide what the cut must communicate

Approve adjacent shots as a pair. Two clips can pass their individual checks while their join suggests an unintended jump in position, time or product design. Write the relationship before generating; a deliberate change of view does not need to look like one uninterrupted take.

Intended relationship	Plan at the boundary	First response when the join fails
Change of view: whole bottle to cap detail	Make the new framing clearly purposeful; keep the same approved cap design	Check whether the detail image depicts the correct product. A framing change is acceptable; a changed cap is not
Match cut: connect similar shapes or positions	Mark subject position, visible size and movement direction in both boundary images	Try a different clean trim point or a crop preview. If alignment removes essential detail, revise the shot plan
Continuous-looking action	Record the outgoing pose, direction and stage of action that the next shot must continue	Review nearby frames and playback. Matching still images alone cannot establish matching movement
Deliberate time or location change	Identify the story cue that explains the change, such as a new setting or an editorial title	Strengthen that cue in the edit; do not spend retries trying to hide a change the story needs

For the fictional bottle sequence, a wide view followed by a cap close-up can be an intentional detail cut. A cap that becomes a different shape must be rejected even if the cut is smooth. Conversely, a size jump alone is not a defect when the purpose is to reveal a detail. These are editorial decisions, not measured model-performance results.

Review the outgoing and incoming frames together

1. Place the last retained frame of shot A beside the first retained frame of shot B. Use the intended delivery crop for both, rather than comparing one full-width image with one cropped image.
2. Record what must persist: product parts or character identity, prop state and relevant lighting. Separately mark deliberate changes in viewpoint, scale or location.
3. Play the end of A and the start of B at normal speed. Check direction, speed changes and whether an action is repeated or skipped. Scrubbing is useful for locating a defect; normal playback checks whether the join communicates the intended beat.
4. Try an edit-only correction when both clips contain usable material: change the retained range or preview an appropriate crop. A trim that removes the required action, or a crop that hides the product, is not an acceptable fix.
5. If the join still fails, identify the responsible shot and change one requirement or source image. Keep the accepted neighbor as the comparison reference. If the plan demands continuity the available clips cannot support, choose an explicit cut or redesign the sequence before buying more attempts.

When an exported frame helps—and when to reject it

Runway’s image-to-video guide describes extracting a final frame as the next clip’s image input and assembling clips in an editor, with attention to the shared boundary frame. That describes its workflow; VidVid users must extract frames externally and confirm the next selected model accepts image input. It does not establish continuous motion or automatic assembly in VidVid.

Use the last retained clean frame, not automatically the final frame in the downloaded file. Reject a distorted cap, face or hand before passing it forward. If you choose an earlier frame, update the retained range in the edit so that the next clip does not resume from a state the audience has already passed. Inspect for an unintended repeated-frame pause after assembly; retain a pause only if the timing calls for it. A clean approved source is preferable when no usable boundary frame remains.

Hand the editor a decision record

Copy this untested editorial template alongside the approved clip files:

Field	What to record
Shot pair and file versions	A → B, exact approved filenames; do not leave a rejected revision as the current file
Retained ranges	In/out timecodes for each source file, with the frame-rate convention if available
Intended join	Detail cut, match cut, continuous action, or explicit story change
Required continuity	Specific cap/prop/identity features and action state; separate planned changes
Boundary evidence	Exported comparison images and a note describing the issue or approval
Delivery and audio	Crop/aspect target; whether audio should continue, change or be added externally
Decision and next action	Accept, edit only, revise A, revise B, or redesign; owner and remaining revision budget

Do not ask the editor to infer which version passed review. If the visual join works but the soundtrack jumps, record an audio edit separately; another visual generation is not automatically the right remedy.

Review clips individually and together

1. Inspect the beginning, middle and end, then play the whole clip for flicker and shape changes.
2. Check fixed details against the approved source image. Reject misleading product changes or identity drift.
3. Place clips in sequence and check direction, subject size, lighting and the transition boundary.
4. Add verified captions, music and branding in your editor; confirm rights for those assets.
5. Review the final crop, caption readability and exported file at the intended viewing size.

When a clip fails, classify the problem as source image, motion, continuity or final layout. Change the relevant brief field before trying again. Consult the artifact troubleshooting guide for visible defects.

Budget revisions before expanding the sequence

Record the selected model, request settings, displayed cost, result and reason for each revision. Start by validating the most difficult shot, then expand the sequence once its acceptance criteria are achievable. Read the pricing guide and check current cost in the interface before submitting. This workflow does not imply free retries or a fixed cost per finished video.

Editorial review: October 11, 2026. Added shot-pair decisions, boundary review and an editor handoff record; no new paid model tests.