Kandinsky 6.0 stops at 5 s: a 30 s ad as one job or six clips

A 30-second ad from Kandinsky 6.0 Video means six separate 5-second generations that you join yourself, because the repository lists 5 seconds as the clip length. On Sume the same 30 seconds can be one request: Wan 3.0 accepts 2 to 30 seconds and Seedance 2.5 accepts 4 to 30 seconds, per the docs. At the 720p Wan rate in Sume's pricing tests, one 30-second job is $3.75.

Kandinsky facts are from its GitHub repository, read 2026-10-11. Sume facts come from Video Generation, Timeline 1.0, Audio detach and Timeline audio. Sume does not list Kandinsky, so the stitch route below applies to clips you made elsewhere and import.

Route A: one long job

One request keeps a single continuous shot and one prompt. The catch is the limit: the docs say each other catalog model has a maximum of 15 seconds, and Gemini Omni Flash 1.1 tops out at 10. Only wan-3.0 and seedance-2.5 reach 30. Wan 3.0 at 720p bills $0.625 for 5 seconds in Sume's golden-amount test, which is $0.125 per second, so 30 seconds is 30 x $0.125 = $3.75 on a straight-line assumption.

Route B: six 5-second clips and Timeline

Stitching gives you six cuts, so plan for them as an edit rather than one shot. Timeline 1.0 takes one audio spine plus up to 200 ordered video slots and bills $0.10 per output minute, rounded up. A 30-second render is one billable minute. Because the render uses a single spine, native clip audio does not ride along by itself. The documented way to keep it is to pull each clip's audio with Audio detach ($0.01 per job), join the six files with Timeline audio concat ($0.01 flat, up to 20 parts), and pass that file as the spine.

Which route when

Both routes are valid; they answer different needs.

• One continuous move or a single subject across 30 seconds: Route A, because a stitch will show seams.
• Six distinct beats with different prompts: Route B fits, and the cuts are the design.
• Open weights and zero per-clip fee: Route B with clips from your own GPU, plus about $0.17 in Sume assembly fees. Local clips must be importable by URL first, since Sume's media tools refuse off-host files.
• Most predictable price: Route A. Six takes can also mean six retries, and each retry is a clip you re-render.

Caveats

The $3.75 figure assumes the rate scales linearly with seconds, which matches how the pricing tests describe a per-second SKU but is arithmetic here, not a quote. Confirm the live price with GET /v1/catalog, or run a Timeline plan call, which is unbilled and returns an estimate. Sound at the seams also needs checking by ear, since the concat is sample-exact but the music and ambience inside each clip were generated separately.