What 20 Free Sign-Up Credits Actually Buy on Seedance 2.5
20 credits cannot be converted into a number of videos, because no channel publishes what one credit buys. That is the honest answer, and it is the one the [attempts-needed calculator](/tools/free-tier-calculator) is built around. What the observation does establish is the *shape* of the grant: it is one-off, it is new-users-only, and it does not refill. Treated that way, 20 credits is a first-week buffer — not an income.
Here is the row, what it rules out, and the arithmetic you can run anyway if you are willing to label your assumption.
The observation
| Field | Value |
|---|---|
| Grant | 20 credits |
| Channel | Mina Labs |
| Metric | sign-up bonus |
| Nature | one-off, new users |
| Date observed | 11 Aug 2026 |
| Source type | report |
report, in this dataset, means a user reported it — not that the vendor published it. There is no vendor-side confirmation of the figure anywhere in the 22 observations. If you are going to quote 20 credits to someone, quote it with that qualifier attached.
One-off is the important word
The dataset records two categories of grant, and the difference between them is the whole point:
| Grant | Channel | Value | Refills? | Date observed |
|---|---|---|---|---|
| Daily free credits | Jimeng | 60–100 | yes — daily | 4 Jul 2026 |
| Daily free credits | Jimeng | 88 | yes — daily | 11 Aug 2026 |
| Daily generations | Doubao | 5 per account per day | yes — daily | 11 Aug 2026 |
| Sign-up bonus | Mina Labs | 20 credits | no — one-off | 11 Aug 2026 |
| Annual plan bonus | Lovart | up to 20 generations | no — tied to an annual subscription | 11 Aug 2026 |
A daily balance is a rate. A sign-up bonus is a stock. Confusing the two is the most common way to misjudge a free tier: the grant looks generous on day one and is irrelevant by day thirty. Our dataset's own note on the credit balance — *refill, not stockpile* — applies to the daily figure and deliberately does not apply here, because this one has no refill to speak of.
How big is 20, against something known?
The only way to size a stock is against a rate, and the dataset gives two rates to measure with — with a caveat each.
Against the daily credit balance. Jimeng's daily grant was recorded at 60–100 on 4 Jul 2026 and at 88 on 11 Aug 2026. Twenty credits is therefore somewhere around a fifth to a third of *one day's* balance on that channel. The two channels are different products with different credit systems, so this is orientation, not conversion — but it tells you the sign-up bonus is small relative to a single day of ordinary free use, let alone a month of it.
Against the daily generation cap. Doubao's free tier was recorded at 5 generations per account per day on 11 Aug 2026. If 20 credits bought 20 generations, that would be four days of Doubao's full allowance. It may not: nothing states that a credit and a generation are the same unit, and Mina Labs has no generation-cap observation in the dataset at all.
The arithmetic, with the assumption stated
If — and only if — one credit is one generation attempt, 20 credits is 20 attempts. Running that through the relationship used by the calculator, attempts = ceil(clips ÷ usable rate), against the one usable-rate figure in the dataset:
| First-attempt usable rate | Finished clips from 20 attempts | At 30 s per clip | Days of Doubao's 5/day |
|---|---|---|---|
| ~30% | 6 | ~3 minutes | 4 |
| >60% | 12 | ~6 minutes | 4 |
The usable rate — roughly 30% to over 60% — comes from a single host running five test groups on 12 Aug 2026, with structured prompts assigning one job per reference image, by one tester. It is not a benchmark. The 30-second single-pass figure is vendor-stated for ByteDance as of 31 Jul 2026.
Two things survive the assumption being wrong. The four days of allowance holds regardless of what a credit buys, because 20 attempts is 20 attempts against a 5-per-day ceiling. And the usable-clip count collapses by half between the two ends of the rate range — the same 20 credits, twice the output, determined by prompt structure rather than by the grant.
What the 20 is not
Three other one-off grants sit in the dataset, and none of them is this one:
- Lovart: up to 20 generations, observed 11 Aug 2026. The number matches, the economics do not — this one is tied to an annual subscription, so it is prepaid rather than free.
- Doubao Work: 30 days of Standard membership, list price CNY 68, observed 25 Aug 2026. Requires the desktop client and a phone-number login, and the inner quota was explicitly not verified — the membership is confirmed, the generation allowance inside it is not.
- Daily balances on Jimeng. Recurring, not one-off, and themselves unsettled: 60–100 against 88, five weeks apart, with no official statement reconciling them.
What to do with it
- Spend it on structure, not on volume. The gap between 6 and 12 finished clips is prompt structure and one job per reference, not a bigger grant.
- Do not budget it forward. It does not refill. Any plan that assumes 20 credits next month has no support in the data.
- Do not treat it as a sign-up decision. Compared against a daily balance of 88 credits or a daily 5-generation allowance, a one-off 20 is not what determines whether a free tier is usable.
Source and limits
Every figure above is a dated observation carrying the channel it came from. The critical gap — credits per generation — is not published by any channel in the dataset, which is why no conversion is offered here and why the calculator refuses to guess one. The usable rate is a single tester's five test groups and should be read as indicative. Free-tier terms move without announcement; check the date before quoting any number as current.
The complete set of 22 observations is published as CSV, JSON and Markdown at [the data page](/data/).