Test six ad hooks with 4-second avatar clips on standard: cost

Six hook tests at 4 seconds each cost $4.416 on the standard Avatar 1.0 tier without a product image, and $4.656 with one. That is 24 seconds of avatar video at $0.184 or $0.194 per second. Sume avatars are async jobs: you submit a request, get a job id back, and read the finished video later. There is no live video session.

Four seconds is also the shortest video Sume accepts. A planned duration under 4 seconds is rejected, so a 3-second hook has to be padded, either with a longer line or with a silence beat.

The cost grid

If you want to compare more than one tier, here is the same six-hook test at each tier. Use standard for exploring and keep plus and max for finalists.

How to run the test

Write six first lines that differ in one thing each, such as a question, a claim, a number, a complaint, a demo cue, and a direct address. Keep the avatar, scene, aspect ratio and tier the same for all six, so the hook is the only variable. Submit six jobs with six idempotency keys, poll them, and collect the six media URLs. The jobs are independent, so they can run side by side within your concurrency limit.

Then put each clip into your ad platform as a separate creative and let the numbers decide. When you have a winner, extend it: render the winning hook plus the rest of the script as a full 15 to 30 second video at plus or max. The hook clip tells you which opening works. The final render is where you spend for quality.

Using a silence beat for a short hook

If your hook is a two-second line, use multi-scene video_inputs: one spoken scene of 2 seconds and one silence scene of 2 seconds gives a planned duration of 4 seconds. Silence scenes take a duration and no text. The extra two seconds still bill at the same per-second rate, so a short hook with a pad costs the same as a 4-second line.

curl -X POST https://api.sume.com/v1/avatar-1.0/talking-video \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: hook-test-03" \
  -d '{
    "avatar_handle": "product_host",
    "quality": "standard",
    "video_inputs": [
      {"id": "hook", "voice": {"type": "text", "script": "Stop scrolling.", "duration": 2},
       "background": {"type": "prompt", "prompt": "Casual bedroom framing, native UGC lighting"}},
      {"id": "beat", "voice": {"type": "silence", "duration": 2},
       "background": {"type": "prompt", "prompt": "Casual bedroom framing, native UGC lighting"}}
    ]
  }'

Reading the result

Do not decide on a few hundred views. Run the six hooks long enough that the differences are larger than the noise, and compare the same metric for each, such as the share of viewers who stay past the first three seconds. A hook test only measures the hook, so keep the body of the ad identical. If two hooks tie, pick the cheaper one to produce, which usually means the shorter script.

The test is cheap on purpose. Six clips at standard cost $4.416, less than a single 30-second clip at plus, which is $7.35. That ratio is the argument for testing small and finishing big.

What a 4-second clip cannot do

It cannot carry a message. A hook is a tease, so do not judge it on whether it explains the product. Judge it on whether the viewer keeps watching. Once a hook wins, write the next 10 to 25 seconds to pay it off, and render the full clip as one job, up to the 60-second limit.

Scaling the test

If six hooks go well, twelve is cheap too: $8.832 at standard without a product image. Cost grows with the number of hooks, and so does the time to read the results, so cap the test at the number you can judge. Two rounds of six, where the second round varies the winner, usually teaches more than one round of twelve.