MiniMax H3 first and last frame guide

MiniMax H3 first last frame control gives a video a visual start, a visual destination, or both. It accepts zero, one, or two images. Zero images activates text to video. One image anchors the shot. Two images define the transition between an opening composition and an ending composition.

This mode is useful when the final state matters as much as the motion. A character can begin outside a room and finish at a desk. A product can start in a closed package and end fully revealed. A visual-novel interface can begin with a choice screen and finish on a character reaction.

Supplied H3 demonstration: two interface states anchor a controlled transition.

Decide what each frame proves

The first frame should establish information the model must preserve: character identity, product geometry, camera height, scene layout, or a UI state. The last frame should show the required outcome, not merely another attractive image.

For a transformation, both frames need a believable relationship. If the camera angle, lighting, lens, and composition all change at once, the model must invent the path between two unrelated images. Keep at least some spatial anchors consistent.

The H3 manual states that first and last frame output follows the original aspect ratio of the input image. Prepare both images with matching dimensions and a ratio within the supported 5:2 to 2:5 range.

Write the motion between the frames

Do not ask the model to "transition smoothly" and stop there. Describe the mechanism. A camera may push into coffee foam until the texture fills the screen, then reveal a desert whose dunes match the foam's ridges. A UI panel may slide away while the camera moves toward the character, ending on the reaction frame.

Name the point where the transition happens. Texture, motion blur, an object crossing the lens, a lighting flash, or a matched shape can hide the change. Choose one method that fits the scene.

Keep the ending stable

Reserve enough time for the last frame to read. If a title, product, face, or interface state appears only in the final instant, the viewer cannot inspect it. Ask for a short hold after the action finishes.

Sound should support the same arc. A rising tone can lead into the transition, followed by one impact or a quieter ambience after the reveal. Native stereo audio is part of every H3 result described in the manual.

Review continuity, not just resemblance

Compare the first generated frame with the source opening, then compare the final generated frame with the source ending. Next, inspect the middle. The model may reach both endpoints while breaking object shape, body position, lighting direction, or camera logic along the way.