2026-09-21 · VideoFreeTier

Seedance 2.5's 5-Generation Daily Cap: What It Actually Buys You

Five free generations a day is not five videos. On the channel where the cap was observed, a generation is an attempt — and most attempts do not come back usable. At the first-attempt usable rate recorded in our dataset, a day's allowance of 5 generations turns into somewhere between 1 and 3 finished clips. Plan against that number, not against the 5.

Here is the arithmetic, and the observation it rests on.

The cap as it was observed

Two things worth separating before going further. The cap and the credit grant were recorded on different channels — the 5-generation ceiling on Doubao, the daily credits on Jimeng. Our dataset does not contain a generation-cap observation for Jimeng, so the 5 does not transfer to it by default.

And the cap is a count of attempts, not of results. That distinction is the whole story.

What one attempt is worth

The only usable-rate figure in the dataset comes from a single host running five test groups:

The range is wide because the two ends are different working conditions, not measurement noise. The upper end came with structured prompts that assigned one job per reference. The lower end is what happens without that structure. The same person, the same day, the same tool — a twofold difference in output.

The arithmetic

The relationship used by the attempts-needed calculator is:

attempts = ceil(clips ÷ usable rate)

Run it against a daily allowance of 5:

Read the middle column first. At the bottom of the observed range, one finished clip costs four of the five attempts. There is one attempt left over — not enough to reliably land a second clip, which needs four more. A bad day on the free tier is a one-clip day.

At the top of the range the picture changes completely: three finished clips, with the full allowance. The difference between those two days is not the tool and not the plan tier. It is prompt structure.

In seconds rather than attempts

Official material puts a single-pass clip at 30 seconds, up from 15 seconds in Seedance 2.0 (31 Jul 2026). Convert the rows above:

• At ~30% usable: about 1.5 usable clips × 30 s ≈ 45 seconds of usable footage per day.
• At >60% usable: 3 usable clips × 30 s ≈ 90 seconds.

So the honest ceiling on a free day is somewhere under two minutes of finished footage. The extension feature — officially "several minutes," and reported at 3 minutes on Jimeng alone — does not change that arithmetic, because our data does not say whether an extension consumes one generation or runs alongside the cap. It is recorded as unresolved.

What the cap does not tell you

These are the questions the observation cannot answer, listed rather than guessed at:

• What counts against it. Whether a rejected, failed or discarded attempt consumes one of the five was not observed. On the API side a related but different rule was recorded (4 Aug 2026): moderation and platform failures were not charged, while dissatisfaction with a successful output was. That is billing behaviour, not cap behaviour, and the two should not be conflated.
• Whether it is per account, per device or per network. The note says "per account per day." That is the extent of it.
• When it resets. No reset time or timezone was recorded.
• Whether the 429s seen on the API are the same ceiling. They are not. A 429 (4 Aug 2026) is a rate cap — too many requests too quickly. The 5-per-day figure is a daily allowance. Two different mechanisms.

What happens to an allowance you do not spend is a different question: whether Seedance 2.5 free credits expire covers the daily refill this cap sits on top of.

How to spend the five

Three decisions that change what five generations actually produce.

• Plan for 1–3 finished clips a day, not 5. Budgeting five assumes every attempt lands, which the observed usable rate contradicts.
• Put structure into the first attempt, not the third. The gap between ~30% and >60% came from prompt structure and one job per reference. Spending that effort up front is what moves a one-clip day to a three-clip day.
• Do not spend attempts on variants you already intend to discard. Each one costs a full generation.

Source and limits

Every figure here is a dated observation carrying the channel it came from. The usable rate is a single host's five test groups with one tester — it is the only such figure in the dataset and it should be read as indicative, not as a benchmark. Free-tier terms move without announcement; check the date before quoting any number as current.

The complete set is published as CSV, JSON and Markdown at the data page.

Not confirmed on this page: whether the 5-per-day cap applies beyond the channel it was observed on — no vendor statement covers it.