A video is not only a series of attractive images. It has timing, direction, and continuity. Start by deciding what should happen in one coherent shot. Then choose the input and workflow that communicate that intention most clearly.
01 / Video
Define the shot before the settings
Name the subject, the action, and the purpose of the shot. Decide whether the viewer should notice a performance, a product detail, a place, or a transition.
Keep the first scene simple enough to review. A short description of one clear action is more diagnostic than a multi-scene story packed into a single request.
02 / Video
Choose text or a visual starting point
Text-led generation gives you room to explore when the scene does not exist yet. An image-led mode can anchor the composition when the visual subject is already established.
The Reference to Video family adds other input roles, but supported combinations must come from the connected mode. Do not assume that every kind of reference is interchangeable.
03 / Video
Separate the sources of motion
Describe subject motion, camera behavior, and environmental motion independently. A person turning, a camera orbiting, and fabric moving in the wind are three different instructions.
Choose the important movement first. Add secondary motion only when it supports the shot and does not make the result harder to evaluate.
04 / Video
Give the shot an ending
An ending can be a held pose, a continued movement out of frame, or another intentional resolution. Saying how the moment ends can clarify the desired progression.
Do not rely on an arbitrary final frame as the only definition of completion. Review whether the pacing leading into that ending makes sense.
05 / Video
Review the entire result
Watch normally to judge timing, then pause around head turns, fast movement, occlusion, or geometry changes. Look for drift, unwanted cuts, inconsistent objects, and audio issues where audio is actually supported.
A poster or thumbnail does not establish motion quality. Keep the full output available when selecting or approving a result.
06 / Video
Choose a continuation deliberately
Use Video Extender when the shot works and needs a next beat. Use Video Edit when the clip needs a visual revision. Return to generation when the concept or action is fundamentally wrong.
Do not stack new processing steps simply because they exist. Decide what is worth keeping, what needs changing, and which workflow expresses that difference.
A direction you can adapt
integrated_multimodal_description: [Shot 1] A steady wide shot of a quiet green woodland at sunrise. Detailed ferns fill the foreground, with tall tree trunks behind them. Warm golden shafts of morning light fall diagonally between the trees, illuminating thin mist and leaves. Leaves and fern fronds sway gently and continuously in a light breeze throughout the five-second clip. The camera stays completely static with no cuts, pans, tilts or zooms. No people, speech, text or watermark. overall_soundscape: Soft natural birdsong with gentle rustling leaves and fern fronds in the breeze. No speech or artificial sounds. non_diegetic_music: N/A
Exact recorded prompt for the displayed five-second Studio woodland video. Open the recipe for its actual settings.
Keep this close
A brief worth keeping.
- Plan one coherent shot before adjusting parameters.
- Give subject, camera, and environment separate roles.
- Watch the whole result, not only the best frame.
- Use continuation and editing only when they solve a specific need.