H3 Max can use a first image and an optional last image to guide where a generated shot begins and ends. The model creates the motion between them; your prompt should explain the path, timing, camera behavior, and sound.
The strongest pairs usually share:
Avoid changing the subject, location, camera angle, lighting, and art style all at once. If several things must change, use a longer duration and state the order clearly.
Begin exactly from the first image. [Describe the motion and camera path.] Preserve [identity and fixed details]. Transition continuously toward the composition of the final image. Reach the final image by [time] and hold it for the last half-second. Audio: [sounds tied to the visible events].Begin exactly from the first image. The camera moves slowly forward as afternoon light fades into blue hour. Leaves and hair respond to a gentle wind while practical lights turn on one by one in the background. Preserve the subject's identity, clothing, and proportions. Transition continuously toward the final composition, arrive by the last second, and hold steady for the final half-second. Audio: light wind, distant city ambience, and one soft tonal rise.Start at 480P to test whether the transition is structurally believable. Use 768P for the selected result when faces, products, or texture need more detail.
| Problem | Better approach |
|---|---|
| Subject changes identity | Use closer source frames and state which details must remain stable |
| Sudden jump near the end | Describe when the final pose should be reached and request a short hold |
| Warped product or logo | Reduce object motion and keep the camera move simple |
| Unnatural path | Choose endpoints with compatible perspective and geometry |