Kling 4.0 five reference videos vs Sume's 1 to 3 video URLs

Kling 4.0 allows up to 5 reference videos with 30 seconds combined, according to Kling's comparison table. On Sume, the legacy Video 1.0 shape takes 1 to 3 reference_video_urls, and each catalog model sets its own limit, so read capabilities before you send.

This post covers only the reference-video count and total length. For the wider 15-asset limit see Kling 4.0 Omni reference: 15 inputs.

What does Kling say about reference videos?

Kling's table compares Kling 3.0 and 4.0. For reference videos, 3.0 lists up to 1 video, max. 15 sec. Kling 4.0 lists up to 5 videos, 30 sec combined. The page does not say how the 30 seconds are split between clips.

What does Sume accept for reference videos?

The legacy Video 1.0 docs list reference_video_urls as 1 to 3 video URLs. On the Video Router, Gemini Omni Flash 1.1 takes up to 3 reference videos, each up to 3 seconds. The kling-3 row takes no reference_*_urls at all.

The limit is therefore a count that depends on the model, not one Sume-wide number. A Kling 4.0 five-clip workflow does not map onto any of these shapes as written.

Where do I check the limit for a model?

The Video Router docs say to read capabilities from GET /v1/video-router/models rather than assuming one envelope. On POST /v1/videos, the model listing reports supported_input_references, which says which reference types a model accepts.

What should I do with five reference clips?

Count the clips against the model you plan to call. If you have more than the model accepts, pick the most useful clips, or split the work into several requests and join the results. Keep each request within the limits the catalog reports for that model.