AI 影片生成前的逐鏡頭預檢清單

Most AI video rework does not start in the generator. It starts earlier, in a prompt that looked complete but left one small decision unmade. A shot can have a beautiful setting and still fail if no one is clearly doing anything, if the camera has no direction, or if the style changes halfway through. Checking each shot before generation is cheaper than regenerating it three times.

Use this checklist shot by shot, not once for the whole video. A long video is a chain of short clips, and a gap in shot 4 will not be fixed by a strong shot 1.

1. Scene: where are we, exactly?

"Outside" is not a scene. "A rainy old street at night, neon signs reflected in wet pavement" is. Ask: could a stranger draw this place from your words alone? If not, add one concrete place, one time of day, and one visual detail.

2. Subject: who or what holds the frame?

Name the subject and one stable trait: a young traveler in a yellow raincoat, not just "a person." If the same character appears again, reuse the same description word for word. That repetition is not boring; it is what keeps the character recognizable across shots.

3. Action: what is happening right now?

Every shot needs one clear action, in the present tense: she looks back, he opens the door, the lantern swings. Avoid two competing actions in one short clip. If a shot tries to do everything, it usually shows nothing clearly.

4. Camera: how does the shot move and frame it?

Choose both framing and movement: a close-up that slowly pushes in, a wide shot that follows from behind, an aerial shot that drifts forward. A missing camera note leaves the most visible choice to chance. If two shots cut together, check that their camera language still feels like one film.

5. Fixed style words: does this shot still belong to the same film?

End with a short, repeated style anchor: the same palette, light, and mood words in every prompt. Check them last because they are easy to drop when you are busy fixing the story. One changed style word can be enough to make a shot feel imported from another video.

A two-minute pass

Read the storyboard top to bottom and mark each shot: scene, subject, action, camera, style. If any box is empty, add a word before you generate, not after you see the result. Fix the missing piece at the prompt stage, then generate the shots in order and join them.

A free, no-sign-up tool can make this pass easier. Storyboard Workshop turns a one-line theme into a storyboard table: it splits a long video into shots, lets you tap words from a swipeable word tree, search across categories, and use a quick-pick area or a one-click lazy mode. Selected words stay only in your browser's localStorage. It outputs only the storyboard table and prompts; it does not generate video directly, and it is an independent tool, not an official tool of any AI video product.

That is the point of a pre-flight check: not more words, but no missing words. When every shot passes the same five questions, generation becomes execution, and rework becomes the exception.