Guide motion with first and last frames

Prepare two compatible images

Use a consistent character, camera angle, framing, and background treatment. Leave room for the whole motion, including hands, weapons, hair, and shadows. A sharply different body scale or viewpoint asks the model to solve appearance changes as well as motion.

An end frame is a constraint for video generation, not a guarantee that every intervening frame will be usable or that the last output frame will be an exact pixel copy. It is not a rig, pose interpolation editor, or frame-by-frame storyboard.

Create the transition in Studio

1. Select first + last frame mode and a compatible model.
2. Add the starting image and ending image. Reuse owned image resources when appropriate.
3. Describe one clear transition, such as “raise the shield, hold the final defensive pose, fixed camera.”
4. Review supported duration, ratio, resolution, and the current quote.
5. Approve generation, then inspect the entire preview, including both endpoints.

If you only want the first image to define the start, use first-frame mode. Adding references belongs to a different workflow.

Make space before generating

Use input canvas when either pose is tight against the image border. The same canvas configuration is applied to both endpoints so they remain aligned. Start with centered placement, then check whether the end pose needs more space.

Canvas expansion scales and pads the existing images. It cannot recover a hand already cut out of the source image, and it does not guarantee the model will keep every future pose inside the frame.

For Alpha Key, both input images need meaningful Alpha. A transparent start plus an opaque end is not a valid transparent-image generation workflow. See transparent animation.

Check the seam instead of assuming a loop

Two similar endpoints may help a repeating action, but they do not guarantee a seamless loop. Watch the last-to-first transition, silhouette, foot position, and acceleration. Repeating a duplicate boundary frame can create a visible pause.

Use trim and frame selection for a continuous range, or Advanced Editor to remove, repeat, or reorder source frames. The editor changes sequence timing; it does not synthesize missing poses.

Public request shape

Supply two independently valid image inputs under input.first_frame and input.last_frame. A file ID must come from a completed upload owned by the current account. Select a model that supports first_last_frame and its actual settings before quoting.

If the selected model does not support the end frame, choose a compatible model or remove that input intentionally. Do not silently downgrade a user-approved two-image request to a one-image generation.