How to plan a multi-shot AI video
A multi-shot AI video is a sequence of short generated shots edited together, and it only works if it is planned before anything is generated. The planning steps are: write a beat sheet, turn it into a numbered shot list with shot size, action and length, set continuity rules for light, direction and wardrobe, decide what to generate per request, and generate in an order that lets you catch problems early. This guide uses a 45-second example to show each step.
Step 1: write the beats
1. Write what happens in five to eight short lines, one per story beat, before thinking about shots. For a 45-second coffee brand piece: she wakes before sunrise; she grinds beans in a dark kitchen; she pours water, steam rises; she carries the cup to the balcony; the sun comes up as she drinks; the bag of coffee on the railing, logo to camera.
2. Give each beat a rough duration. Six beats in 45 seconds is roughly 7 seconds each, with the final product beat shorter.
Step 2: turn beats into a shot list
3. Break each beat into one to three shots. For every shot write a number, shot size, subject, action, camera move, and length. "4A. Medium, her, pours water from a kettle, static, 5 s. 4B. Close-up, the cup, steam rises, slow push-in, 3 s."
4. Plan coverage. Editors cut between shot sizes, so alternate: a wide to establish, a medium for the action, a close-up for detail. Two shots of the same size and angle side by side look like a jump cut.
5. Keep individual shots short. Three to eight seconds is the comfortable range for most shots. Faces and products tend to hold up better in short clips, and short shots give you more choices in the edit. A 45-second piece of this kind usually lands at 8 to 12 shots.
Step 3: set continuity rules
6. Light. Decide the light for each location and write it the same way in every prompt for that location: "blue pre-dawn light from the window behind her, one warm lamp on the counter". Light that jumps between shots is one of the most visible continuity errors in AI video.
7. Screen direction. If she walks from left to right in one shot, she should still be moving left to right in the next. Write it into the prompt. Breaking direction confuses viewers about where things are.
8. Wardrobe and props. Fix the outfit and props through the character and product references, not the prompt. If the cup is part of the story, give it its own reference images or keep it the same simple object in every prompt.
9. Time of day. Tie it to the beats: the sky should be darker in beat 2 than in beat 5. Note it per shot.
Step 4: decide what goes in each request
10. Choose between one shot per request and several shots per request. Kling 3.0 Omni generates up to 15 seconds per request and supports multi-shot generation, according to Kling's user guide as of October 2026.
Grouping two or three shots of the same location in one request keeps light and setting consistent across them. One shot per request gives you more control and makes it cheaper to redo a single shot.
In RefVid, one clip holds up to 3 shots, and the clip's length is split evenly across them: a 10-second clip with three shots gets 4, 3 and 3 seconds. Next shot starts a new clip from the last frame of an earlier one, which suits continuous action across a cut.
A practical rule: group shots that share a location and continuous action; generate separately anything that needs a specific end frame, a product close-up, or that you expect to retry several times.
Choose the aspect ratio once for the whole video, 16:9, 9:16 or 1:1. Mixing 16:9 and 9:16 shots means cropping later.
Step 5: generate in a smart order
11. Generate the hardest shot first. Usually that is the product close-up or the shot with the most movement. If it does not work, you may need to change the plan, and it is better to know before spending credits on the easy shots. In RefVid, a free still of the scene shows how a shot will be framed before you pay for the clip.
12. Then generate one location at a time, all shots for the kitchen, then all shots for the balcony, so you compare light and setting side by side.
13. Rough-cut as you go. Drop each accepted shot into the editor timeline in order. You will see quickly if a shot is too long, if two shots do not cut together, or if a beat needs one more angle. Generate the fix while you are still in that location.
14. Finish in the editor with music, sound, color correction and a real packshot for any logo moment.
Common mistakes and checklist
Mistakes: generating before writing the shot list; shots of 12 seconds or more where 5 would do; the same shot size twice in a row; light described differently for one location; the easy shots generated first and the hard shot last.
Checklist: beats written with durations; numbered shot list with size, action, camera and length; coverage alternates sizes; continuity rules for light, direction, wardrobe and time of day; grouping decided per shot; one aspect ratio; hardest shot generated first; rough cut built as you go.
RefVid keeps the characters and products for a project as saved subjects, so every shot in the list starts from the same photos; up to 3 subjects can share a shot, each named in the prompt, and each subject's page lists its clips. Editing happens in your own tool.
Frequently asked questions
- How many shots does a 30-second AI video need?
- A 30-second AI video typically needs six to ten shots of three to five seconds each. Short shots hold faces and products better and give you more options in the edit.
- Should I generate several AI video shots in one request?
- Group shots that share a location and continuous action to keep light consistent; generate separately the shots that need a precise end frame or several retries. RefVid fits up to 3 shots in one clip.
- How do I keep the lighting consistent between AI video shots?
- Write the light for each location once and paste the same words into every prompt for that location, and generate all shots for one location together so you can compare them.