How to split a $100 AI video ad-test budget on Sume

A $100 test on Sume can pay for 20 eight-second Standard avatar hooks ($29.44), five Max re-renders of the winners ($22.00), ten product stills, the one-time avatar fee and trims, and still leave about $35 for retakes and a second round. The trick is to spend on volume with the cheap tier and put the expensive tier only on proven scripts.

This split is a worked example, not a recommendation of any creative approach. It uses the list rates on Sume's pricing page read on 2026-10-02, treats the image line as the catalog estimate, and assumes each avatar clip bills exactly 8 seconds.

The point of writing the split down is that every line has a unit price you can check. If a line looks wrong, change the quantity and re-run the arithmetic; nothing here depends on hidden fees.

What does round one cost?

Round one explores ideas: one avatar, twenty hooks on Standard, ten product stills and a trim for each hook you want to cut down.

Twenty hooks is a choice, not a rule. At $1.47 each on Standard, every extra hook is cheap, so the cost of testing one more idea is small compared with the cost of promoting a weak one later.

What do the winners cost?

Pick the five hooks that did best and re-render them on Max at 8 seconds each. Max lists at $0.55 per second, so each costs $4.40 and five cost $22.00. If those clips show a product image, the rate is $0.58 per second and the five cost $23.20.

Round two adds a timeline compose at $0.02 per job to assemble each winner into the final cut, so five more cost ten cents.

Max is the highest quality tier and slower. Sume does not publish a quality ranking for the tiers, so treat the Max pass as a finishing choice for clips you already like, not as a guarantee of better results.

How much buffer should I keep?

Rounds one and two total $55.19. A 20 percent buffer for retakes on the volume tier is about $6, but a failed or disappointing hook is a new job at the full rate, so a 30 percent buffer on round one is safer. That is about $10 of the $100.

The remaining roughly $35 is the real flexibility. It can fund a third round of ten more hooks on Standard ($14.72) and five more winners on Plus ($9.80), or sit unspent. Do not assume you must use it.

HOOK = 0.184 * 8          # Standard, 8 s
WINNER = 0.55 * 8         # Max, 8 s

round1 = 0.95 + 20 * HOOK + 10 * 0.23 + 20 * 0.02
round2 = 5 * WINNER + 5 * 0.02
buffer = round1 * 0.30
spare = 100 - round1 - round2 - buffer
print(f"r1 {round1:.2f} r2 {round2:.2f} buf {buffer:.2f} spare {spare:.2f}")

How do I stop the test from overspending?

Sume reserves the estimate at submit and captures the real amount afterwards, so the wallet balance is the hard boundary: a request that cannot be covered is refused with 402 insufficient_credits before any rendering. Fund the wallet with the test amount and nothing more.

For Formats, Actions and runs there are also spend caps, and a run's cap field in the usage summary shows what is left. Check summary.debited_usd after each round before you start the next one.

A practical safeguard is to make the budget a property of the wallet, not of a spreadsheet. A balance that cannot cover the next reserve stops spending on its own.

What does this split leave out?

• Media buying. The $100 covers creative only, not ad spend on any platform.
• Voiceover and music. Add them at the cheap-step rates if your hooks need a separate track.
• Rights and review. A clean render is not a cleared ad. Check platform labels and rules before you publish.