How to use first and last frames in AI video
Learn when to give an AI video model both start and end images, how to plan the motion between them, and when to join shots with a cut.
Choose a destination you can show in a still
First-and-last-frame video generation gives a supported model two images for a single clip: where the shot begins and where it should land. Use it when a precise end pose matters. For example, a tabletop shot starts with a box closed and ends with the lid lifted. The final image can establish how far the lid opens and what the viewer sees inside.
Align both stills before sending them. Keep the same crop, lens perspective and main lighting direction unless the change is the point of the shot. Use a starting image alone if the final pose is flexible. Google Cloud's first-and-last-frame guide shows how to attach these inputs to supported Veo endpoints. Check the exact model and mode you plan to use.
Describe the movement between the stills
For the box reveal, try: “Locked tabletop camera. One hand lifts the lid steadily from the near edge, then holds it open for the final beat. The box stays in the same place.” Use the source photos to establish the look of the product. If the contents must be exact, compare the result with a real photo before using the shot.
Scrub through the beginning, middle and end. A good final pose cannot rescue a middle frame where the hand changes shape or the box slides. When the motion is unstable, first bring the start and end poses closer together or simplify the movement. Retest one change at a time so you can identify what helped.
Connect clips with a shared frame only when the action continues
End-frame conditioning happens inside one clip. A last-frame bridge uses the last clean frame of an approved clip as the start image for the next clip. Match camera direction and motion speed, trim a repeated seam frame if needed, then watch the join at normal speed.
When moving from a wide tabletop view to a label close-up, cut on the lid movement. A cut gives the two framings a clear relationship and avoids forcing the model to invent the whole camera move. In Scena Studio's video workflow, choose the input mode per shot, then queue and review the shots that belong together.