Planning B-Roll for AI Product Videos: A Shot List That Actually Works
Most AI product videos fail in the same place: the script makes five claims, and the footage supports maybe two. The gap is not a generation problem — modern video tools can render almost anything you describe — but a planning problem. A b-roll shot list built from the script closes that gap before the first clip is generated, and it turns a day of prompt-and-pray into an hour of systematic production.
**Start From Claims, Not From Ideas**
Read your script and underline every concrete claim: "waterproof to 30 meters," "sets up in under two minutes," "fits in a jacket pocket." Each claim becomes at least one b-roll shot that demonstrates it visually. A claim without a shot is a claim the viewer must take on faith; a shot without a claim is filler that dilutes the video. Write the list as three columns — claim, shot description, duration — and the structure of your video appears by itself.
**Describe Shots the Way a Camera Would Take Them**
AI video generators respond to camera language. For each shot, write the subject, the action, the framing, and the camera movement: "close-up of hands snapping the case shut, slight push-in, soft window light." Three details separate usable generations from rejects: the light source ("soft window light," not "good lighting"), the framing word (close-up, medium, wide, overhead), and one movement at most. Shots described with two camera movements or two subjects nearly always come out wrong, and fixing them costs more generations than writing them precisely the first time.
**Lock Consistency Before You Batch**
B-roll feels amateur when the product changes shape, color, or finish between shots. Before generating a batch, create one reference image of the product in your hero angle and reuse it as the reference for every shot, and write the product description once as a reusable phrase — same words, same order, every prompt. Keep backgrounds to two or three named environments ("white studio," "oak desk," "outdoor trail") so the video reads as one shoot, not ten. GetVideoStudio's reference-image workflow keeps the product recognizable across an entire batch when you attach the same reference to every generation.
**Shoot for the Edit, With Handles**
Ask for slightly more than you need: a four-second shot used for three seconds gives the editor room to trim around a shaky first frame. Plan transitions on paper — which shot follows which — because motion continuity (a left-to-right pan followed by another left-to-right pan) makes cuts invisible. And keep every shot's prompt saved next to its clip; when one shot needs a regeneration, you re-run the exact prompt instead of rebuilding it from memory.
**The Checklist Before You Generate**
Every claim has a shot. Every shot has a framing, a light source, and at most one movement. One reference image governs the product's look. Backgrounds are limited to a short named list. Durations include editing handles. Pass those five checks and the batch you generate is the batch you publish — with most of the editing already decided on paper. GetVideoStudio handles the generation half: batch clips from one reference, batch audio extraction for narration reuse, and thumbnail extraction for the cover. Questions or feedback: sapsap@qq.com.