Black Friday hook tests: 20 ten-second variants on seven Sume rates

Twenty ten-second Black Friday ad variants cost between $12.50 and $50.00 on the Sume catalog models, depending on the model and resolution. Wan 3.0 at 480p is the cheapest row at $0.0625 per second, and Wan 3.0 at 1080p is the dearest at $0.25 per second.

A hook test is the cheapest honest way to find a winner before the sale starts. You render many short variants of the same offer, run them as paid ads, and spend the real budget on the survivor. This post prices the render side only, so you can decide how many variants a budget buys.

How the prices are built

Sume bills each catalog video model at the provider list price times 1.25, per output second. The Video Router docs state the house margin, and the Video generation docs list the duration and resolution envelope of each model. The provider list rates below come from the Sume pricing tables in the repo. They are not a quote, so read your own GET /v1/videos/models response before you commit a budget.

A ten-second clip fits inside every envelope in the table. wan-3.0 accepts 2 to 30 seconds, gemini-omni-flash-1.1 accepts 3 to 10 seconds, and minimax-h3-max accepts 5 to 15 seconds. Omni Flash 1.1 tops out at ten seconds, which is why ten seconds is the common length here.

Where to spend and where to save

Do not render all twenty at 1080p. A hook test judges the first two seconds: does the product land, does the motion read. That is visible at 480p or 720p. Render the winner again at the resolution your ad platform wants.

Wan 3.0 and Omni Flash 1.1 cost the same at 720p, so pick by what the clip needs. Per Google's launch post (read 2026-10-05), Omni 1.1 Flash has a 360p draft mode that is up to 60 percent faster and a third of the cost of standard 720p. Sume lists 360p at $0.0375 per second, so twenty ten-second drafts cost $7.50. That is a third of the 720p row, which matches Google's claim.

Wan 3.0 is the longer one. Alibaba's Wan 3.0 repository (read 2026-10-05) describes native 30-second generation, and Sume accepts up to 30 seconds. If your hook needs a setup and a payoff, ten seconds of Wan at 720p is $1.25.

• Draft at 360p on Omni Flash 1.1 ($7.50 for twenty), or 480p on Wan 3.0 ($12.50).
• Promote the top three variants and re-render only those at 1080p: $2.50 each on Wan, or $1.875 each on Omni.
• Keep the prompt identical across variants except for the hook line, so the test measures the hook.

One variant, one request

Send one job per variant. The request is the same shape for every row; only model, resolution and prompt change.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: bf-hook-07" \
  -d '{
    "model": "wan-3.0",
    "prompt": "Vertical product hook: gift box opens, text-free, warm light",
    "resolution": "720p",
    "duration": 10,
    "aspect_ratio": "9:16",
    "mode": "async"
  }'

Guardrails

Give each variant its own Idempotency-Key. A retry with the same key and payload returns the same job instead of a second charge. The mismatch most teams hit is a new key on every retry, which creates a second paid job.

Price the whole test before you start: variants times seconds times the rate. Twenty variants is a $12.50 to $50.00 decision, not a budget line that needs approval. The larger cost is the finals you render after the test, so keep the draft tier cheap.

Reading the test

Read the results by hook, not by model. If the same hook wins on two models, the idea is strong, and you can pick the cheaper model for the final. If a hook wins on only one model, check whether the model changed the motion or the framing, since that is a render effect and not a message effect.

Log each variant with its job id, model, resolution and hook line in a sheet. When the sale ends you will want to know which rate bought the winning ad, and the sheet is the cheapest way to keep that record. The job result also shows the billed amount, so you can reconcile the sheet against what Sume charged.

If you want the same test as a repeatable run rather than twenty hand-built requests, the Format API takes a queue of up to 100 runs, covered in the posts linked below.