AI real estate walkthrough from photos: chain first/last-frame clips

For a walkthrough built from listing photos, Kling 4.0's Multiple Keyframes lets you set up to 10 images along one clip. Sume's video models take a first and a last frame per request, so the equivalent is one clip per pair of photos, then a join. Each clip ends on a photo the next one starts from.

Kling details are from its comparison post, read 2026-10-01; Sume's from the video generation and Timeline docs. Kling's post says 4.0 launches officially in October, with early access limited, so this is a plan, not a tested route.

What does Kling 4.0 add for this?

The post compares the two versions: Kling 3.0 supports start and end frames and no multiple keyframes; Kling 4.0 supports up to 10 keyframes and a 3 to 30 second duration, against 3 to 15 seconds for 3.0. Its example sets visual milestones at timestamps, such as a character entering at 0 seconds, which maps naturally to rooms in order.

What does Sume accept instead?

The docs say frame_images specifies first or last frame images for image-to-video, and catalog entries list supported_frame_images of first_frame and last_frame. Two anchors per clip, not ten.

How do I join the clips?

Import the finished clips as Sume-hosted media and place them in order on a Timeline 1.0 render. The Timeline docs offer transitions on slots after the first, such as dissolve, and note that stills are static holds, so a still can open or close the tour without a generation. The same idea at two frames is in Kling 4.0 multiple keyframes vs Sume first and last frame.

What should I check before publishing a listing video?

A generated move between two photos invents the space in between, so rooms can appear that do not exist. Review each clip against the property and disclose that the video is generated; the docs here make no claim about legal or listing-platform rules.