10 hooks times 3 endings: 30 video ads from 13 clips, cost on Sume

Ten 5-second hooks and three 5-second endings make 30 different 10-second ads, and on Sume they cost $11.19 with Wan 3.0 at 720p: 13 generated clips at $0.63 plus 30 timeline renders at $0.10. Generating 30 full 10-second clips would cost $37.50. The saving comes from paying for 13 clips and combining them, instead of paying for 30.

This is the cheapest way to test hooks, which is where an ad's performance usually differs. Keep the ending fixed, vary the first five seconds, and each variant is a $0.10 render of two clips you already own. The prices are Wan 3.0's list rate times 1.25, rounded up per job (read 2026-10-06), and the render is $0.10 per started output minute.

What stitching gives up

Two clips joined with a cut or a short transition are not one continuous shot. The hook and the ending are generated separately, so characters, lighting and the product can differ between them, and a viewer will notice a mismatch if the brief does not hold them constant. Stitching works best for ads built from separate beats: a hook, a product shot, a call to action.

Use a still as the bridge when the shots do not match. A Timeline slot can hold a still, a clip or a transition of up to one second, so the join can be a fade rather than a hard cut.

• Write the ending once and reuse it for every hook, so the offer line is identical across variants.
• Generate the hooks at 480p first. A hook that does not work at $0.32 does not deserve a $0.63 re-render.
• Use the unbilled plan call to check each render's minute count before you submit the batch.
• Read the whole matrix into one spreadsheet, with one row per ad, so you can tie results back to the pair.

Cost a matrix before you generate it

The function below prices any hooks-by-endings matrix and prints the saving against generating every ad as one full clip.

HOOK_CLIP = 0.63  # Wan 3.0, 720p, 5 seconds
FULL_CLIP = 1.25  # Wan 3.0, 720p, 10 seconds
RENDER = 0.10     # per started output minute


def matrix(hooks: int, endings: int) -> dict:
    ads = hooks * endings
    stitched = (hooks + endings) * HOOK_CLIP + ads * RENDER
    full = ads * FULL_CLIP
    return {"ads": ads, "stitched": round(stitched, 2),
            "full": round(full, 2), "saved": round(full - stitched, 2)}


for h, e in ((3, 2), (10, 3), (20, 5)):
    print(h, e, matrix(h, e))

Render only the pairs you will test

You do not have to render the whole matrix. A render is $0.10, so rendering all 30 pairs costs $3.00, but a test usually needs a handful. Pick the three best hooks by eye, render the 9 pairs they make with the 3 endings for $0.90, and add more pairs only when a hook earns its place.

The generated clips are the larger cost, so spend the review time there. A hook that cannot hold a viewer for five seconds should be dropped before you pay for the other half of its pair. Ten hooks at $0.63 is $6.30. If four of them are weak after one look, you have spent $2.52 on them and avoided $1.20 of renders and the time to check them.

When full clips are still the right call

Even a 3 x 2 matrix pays: six ads for $3.75, against $7.50 for six full clips. The saving grows as the matrix grows, because the generated clips grow with the sum of hooks and endings while the ads grow with the product. At 20 x 5 you pay for 25 clips and get 100 ads, $25.75 against $125.00.

If the story needs one continuous shot, generate the full clip. The model prices and limits are in the Video Router catalog, and the slot and transition rules are in the Timeline docs.