To generate an AI video, begin with one shot and one action. Choose whether the scene needs a visual reference, describe what moves and what stays still, then inspect the whole clip. Add complexity only after the basic shot works.
Afterform’s projects provide a place to organize sources and versions. Use a video workflow enabled in your account, and confirm its input types, duration, aspect ratio, sound options, and estimate before generating.
1. Write a shot card
A shot card turns a vague idea into something you can accept or reject. For a five-second product concept, your planning card might look like this; the actual generated duration depends on the available model.
| Field | Example |
|---|---|
| Purpose | Introduce a fragrance bottle |
| Starting image | Reviewed bottle on a stone shelf |
| Action | A small leaf shadow moves across the shelf |
| Camera | Slow, straight push toward the bottle |
| Must stay fixed | Bottle shape, cap, label and shelf position |
| End condition | Bottle remains fully visible with room for copy |
This is more useful than asking for a “cinematic viral product commercial” with no visual constraints.
2. Choose text-to-video or image-to-video
Text-to-video is useful for exploring scenes when a specific product identity is not the main constraint. Image-to-video starts from a visual anchor and is often the more controllable planning choice for a reviewed product scene. It still requires fidelity checks.
If your workflow supports a first frame, attach the image in that role. A general reference is not necessarily the first frame. Do not treat an end-frame control, a style reference, and an ordinary uploaded image as interchangeable.
3. Describe motion directly
With a supplied starting image, emphasize action and camera behavior rather than redescribing every visible object.
Begin from the supplied fragrance image. The bottle stays stationary. The camera makes one slow, straight push-in. A soft leaf shadow moves gently across the stone behind it. Keep the cap, bottle silhouette and printed label stable. Maintain the existing light direction. One continuous shot, no cuts, no new objects.
Avoid incompatible directions such as “locked camera” and “rapid orbit” in the same prompt. A request for several scene changes is also a sequence brief, not a simple single-shot prompt.
4. Keep the first attempt simple
Choose a supported duration and crop. Check whether audio is part of the selected option; do not assume every video model creates sound. Review the estimate and submit a small trial.
A first attempt should answer one question: can this workflow execute the movement while preserving the subject? A shorter, simpler shot can answer that more clearly than a long, elaborate sequence.
5. Inspect time, not only appearance
Watch the clip normally, then examine the opening, midpoint, and ending. Check for changing product geometry, flickering text, objects appearing or disappearing, and sudden shifts in light. Inspect interactions particularly closely: fingers touching a cap can expose errors that a static product shot avoids.
If audio is present, listen from start to finish. Reject stray speech, clipped words, or sounds that contradict the action. Preview images and silent playback cannot establish audio quality.
6. Fix the failing variable
If the product changes shape, reduce camera movement or use a stronger starting image. If the action is unclear, remove secondary motion. If the ending is unusable, shorten the usable portion in an appropriate editor or try another supported control.
Keep accepted takes separately from experiments. Record why a take failed, such as “label changes after the camera move,” so the next prompt responds to the problem rather than adding generic adjectives.
Prepare the final handoff
Export an accepted take using the supported format, then inspect the downloaded file. Check duration, dimensions, sound, and the intended crop. Add exact captions, branding, and a call to action in a finishing workflow that supports them; generation alone is not the complete ad.
Use the AI ad guide to plan a sequence, or the image-to-video guide to improve the starting frame before another attempt.