How many seconds of AI video does $100 buy? Every Sume model

Price per second is the number every video model shares, even when the bill is computed differently. Dividing $100 by it gives the number that makes a budget concrete: how many seconds of footage you can generate. This post does that for each Sume video model and tier, from rates read on 2026-10-03.

Seedance 2.5 is billed by tokens, so its per-second figure depends on frame size; the table uses 16:9 and a 10-second clip.

What does $100 buy?

The figures are exact divisions before rounding and assume every job succeeds and is kept. Read them as an upper bound.

What do the numbers hide?

A second of footage is not a second of finished ad. Hooks need alternatives, finals need retakes, and unusable clips still cost. A common planning rule is to divide by your keeper rate, which you can read off your own usage rather than guess. If one clip in four is kept, $100 of Wan 3.0 at 720p buys 200 seconds of footage you would use, not 800.

The table also mixes products. Avatar Video prices a talking presenter and Genjutsu prices a source clip you already have, so they are not substitutes for text-to-video. Compare within a purpose.

What is the next step?

Pick the tier by what the clip must do, then read the price of one clip, not the average. The wallet reserve on each job is the table rate times the clip length, and a balance too small for it returns 402 insufficient_credits. For funding amounts see how much to top up.