Kling 4.0 Creative Recreation vs Sume reference ingest manifest

Kling describes Creative Recreation as borrowing a reference video's pacing, shot structure and presentation style, then replacing the characters, products or other assets. Sume's docs do not list a generation mode like that; what they do list is the first step, reference ingest, which reads one clip you own into a manifest of shots[], text_tracks[] and audio facts. It is dest first and opt-in on production.

Vendor text from Kling's comparison post; Sume facts from the reference ingest docs, read 2026-10-01.

What does Creative Recreation need from a reference?

Per Kling, creators can borrow pacing, shot structure and presentation style, for example rebuilding a trending product video around their own product. Before any model call, a plan needs those structural facts written down.

What does Sume's reference ingest return?

One call on a media.sume.com clip (up to 300 s) answers a manifest with shots[], text_tracks[], audio, an overview strip, and coverage and uncertain[] fields that say what ran and what to look at again. It is read-only: the source is never re-encoded and no MP4 is produced.

Does it interpret the video for me?

No. purpose accepts reference_remix (the default), brief_format, face_swap or qa, and the docs say it is stored, not interpreted. semantic: true is refused with reference_ingest_semantic_unavailable until the enrichment pass ships.

What do I do with the manifest?

Use the cut list as the beat sheet for your own shots, keep or rewrite the text lines, and decide per line whether the source audio is worth reusing; audio.silent: true means plan new music. The full flow is in UGC remix from a TikTok reference video. Remember the surface is dest first, so confirm it is listed for your environment.