15 cuts in 30 seconds: Wan 3.0 2 s shots vs Seedance 2.5 4 s shots

For a 30 second ad with the most cuts, use Wan 3.0 on Sume: its shortest clip is 2 seconds, so 15 shots fit, and 15 shots at 720p cost $3.75. Seedance 2.5 and Kling 3 both start at 4 seconds, so the same 30 seconds holds at most 7 shots, at $16.17 (Seedance 2.5, 720p) or $3.92 (Kling 3, audio off).

The question comes up because the new Kling 4.0 is being sold on long single takes and many keyframes. The fal explainer (read 2026-10-05) gives 3 to 30 seconds for the full model and 3 to 20 seconds for Flash. Those are not on Sume yet. What you can run is below, and the floor of each model decides how fast you can cut.

Floors and rates

A short shot is not the same as a short clip. Each generation job has a minimum length per model, and you pay for that minimum even if you only use part of it. The Sume catalog reports supported_durations, and the video docs state the ranges: wan-3.0 accepts 2 to 30 seconds, seedance-2.5 accepts 4 to 30, and kling-3 (Kling Video v3 Pro) accepts 4 to 15.

Wan 3.0 is also billed per second in the Sume pricing tables: $0.0625 at 480p, $0.125 at 720p and $0.25 at 1080p, after the 1.25 margin. A 2 second shot is therefore $0.125, $0.25 or $0.50. Seedance 2.5 uses a video-token formula, so its 4 second floor costs $1.07, $2.31 or $5.69 at the same three resolutions.

What 30 seconds of cuts costs

What the table leaves out

Two caveats keep the comparison honest. Seven 4 second shots reach only 28 seconds, so you need either a 6 second tail or a trim to reach 30. And kling-3 does not list 480p, and the estimator prices it per second, not by resolution, which is why its figures repeat across the 720p and 1080p columns.

Wan also supports a wider spread of references: up to 10 images, 5 videos and 5 audio clips per the model notes, but no 21:9. If the cut needs a wide frame, Seedance 2.5 is the only one of the three that lists 21:9.

Joining fifteen shots

Fifteen shots is more than the 12 slots Timeline renders in one pass, but that is not a problem: the default auto strategy chunks past 12 segments, and only the explicit single strategy is refused. For hard cuts there is no transition limit to hit, because a hard cut is not a chained fade. The render is $0.10 for the minute.

This script prints the table above, so you can change the shot length or the count.

WAN_PER_SECOND = {"480p": 0.0625, "720p": 0.125, "1080p": 0.25}
KLING_PER_SECOND = {"off": 0.14, "on": 0.21}

def shots(total, floor):
    return total // floor

n = shots(30, 2)
for res, rate in WAN_PER_SECOND.items():
    print("wan-3.0", res, n, "shots", round(n * 2 * rate, 2))

k = shots(30, 4)
for audio, rate in KLING_PER_SECOND.items():
    print("kling-3 audio", audio, k, "shots", round(k * 4 * rate, 2))

Which to pick

Pick the model by what the cut needs. Fast montage with plain motion: Wan 3.0 at 480p for a draft, then 720p for the final. Wide frame or a long take: Seedance 2.5. Dialogue or sound on short shots: Kling 3 with audio on is the cheapest of the audio-capable models in this table. The video generation docs list the ranges, and the Timeline docs cover the join.

A practical plan for a 30 second montage is to write the shot list first, with a duration for each shot, and check every duration against supported_durations before you submit. With Wan 3.0 you can mix 2, 3 and 5 second shots, as long as the total reaches the spine length, and a 30 second Wan job is also available if you would rather have one continuous take at $3.75 at 720p.

Submit each shot as its own job with an Idempotency-Key, since a failed shot should be retried alone, and poll the jobs as in the jobs docs.