First, last and keyframes

First and last frames control a segment's endpoints. Multiple keyframes plan several distinct states. Both need instructions for what happens between images; pictures alone cannot reliably explain complex cause and effect.

Use first and last frames

1. Switch the shot-generation area to First/Last Frame mode.
2. Select or upload the opening frame. Add a final frame when the ending needs a fixed visual target.
3. Check that identities, environment, ratio and props can remain continuous between them.
4. Describe action direction, camera movement and the final pose. Allow enough time.
5. Review endpoint matching and the transition, particularly deformation, prop changes and unnatural speed.

Good uses include entrances and exits, slow pushes, orbits, blooming flowers, outfit changes, transformations and spatial transitions. Split stories with rapid reversals or frequent cuts into simpler segments.

Instructional example: plan continuous camera movement through opening and ending compositions.

Capture a frame from existing footage

Find the desired moment in the video preview, capture it and save it to the frame asset area. Reference it from the shot editor. For a continuous sequence, use the previous segment's final frame as the next opening frame and describe how the movement continues.

Using the same image at both ends can help attempt a loop. Check motion, lighting and sound at the join; identical endpoints do not guarantee a seamless loop.

Describe multiple keyframes

Upload images in order and describe the state defined by each. Ordered keyframes are not a guarantee of exact frame-level or millisecond timing.

Use @Image1 through @Image4 as keyframes in that order.
@Image1: the boat is centered, the ferryman stands at the stern,
and the traveler sits at the bow.
@Image2: the traveler turns toward the water on screen left;
the boat's position remains continuous.
@Image3: a vague figure appears in the mist as the camera moves closer.
@Image4: the traveler stands; the ferryman stops his pole;
both look ahead.
Keep identities, clothing, the boat and morning mist consistent.
Use natural continuous transitions. Add no characters or subtitles.

Multiple-keyframe example: ordered visual states guide a complete segment.

Use a storyboard grid

For shot order, prefer a grid of 15 panels or fewer. This is quality guidance, not a confirmed hard upload limit. State its reading order, such as left to right and top to bottom. Keep it clean; excessive labels and unclear placeholder figures can interfere.

Use the grid for sequence and composition, separate character images for identity, and a location image for the environment. Explicitly exclude the grid's line-art style and annotations from the final render. Ensure the total duration can accommodate all actions and dialogue.

See Prompt templates and examples for more patterns.

Six reusable patterns

Replace every @Image and @Video with an actual mention. Choose a duration supported by the model. Time ranges guide generation; they do not provide frame-exact timeline control.

Task	Prompt starting point
Product orbit	Start from @Image1 and slowly orbit the watch. Preserve dial structure, reflections and background.
Transformation	Start from @Image1. An original armored shell forms from the feet upward; preserve position and describe the final state.
Flower opening	@Image1 is the closed-bud start, @Image2 the open-flower end. Petals unfold as sunlight brightens; plan 5 seconds.
Outfit transition	@Image1 defines casual clothes and @Image2 the formal ending. Preserve identity and blocking with a continuous fabric transition; plan 10 seconds.
Three states	Follow @Image1, @Image2, @Image3: start moving, raise a hand, then settle. Describe timing and cause.
Multi-part passage	@Video1, @Video2, @Video3 define successive spaces. Specify order, entry/exit points, camera direction and transitions.

For a loop using the same first and last image, the action should also return to its starting pose. An action prompt can follow “subject + start-to-end action + camera,” such as a skater jumping onto a rail while the camera follows behind. Endpoints do not replace instructions for the motion between them.