Seedance Draft Mode: When a 480p Preview Before 1080p Actually Saves Money
Credits and specs follow TWno1's live billing settings. Last checked: 2026-10-05
In October 2026, a 5-second Seedance 2.5 draft at 480p costs US$0.514, about 18% of the 1080p final.
Cheap is not the same as thrifty. Draft mode adds a step you pay for every time. It only wins when it stops you from paying full price for takes you would have thrown away. Below is the arithmetic, using BytePlus list prices checked on 2026-10-05, plus a break-even keep rate you can apply to your own projects.
What draft mode does, and what it does not
BytePlus ModelArk describes draft mode as two separately billed tasks:
- Draft. You generate a 480p preview. It is billed exactly like a normal 480p video. No other resolution is accepted.
- Final. If you like the draft, you send its task ID back and get the final. On Seedance 2.5 the final is 1080p only.
The docs list draft mode for Seedance 2.5 and 1.5 pro. The 2.0 series is not on that list.
Here is the part people miss. Step 2 is not an upscale of your 480p file. It is billed at the full 1080p rate, the same as rendering 1080p from scratch, and nothing we read in the BytePlus docs promises the final will match the draft frame for frame. Treat it as a fresh generation that inherits the draft's settings. So every cent of savings comes from drafts you rejected. The final itself costs exactly what it always did.
The prices
Seedance 2.x bills by video tokens. BytePlus's estimate: tokens ≈ seconds × width × height × fps ÷ 1024, times the per-million rate. Failed tasks and tasks blocked by moderation are free.
| Seedance 2.5, no video input (2026-10-05) | Per million tokens | BytePlus 5 s 16:9 example | Per second |
|---|---|---|---|
| 480p | US$10.70 | US$0.514 | about US$0.103 |
| 720p | US$10.70 | US$1.156 | about US$0.231 |
| 1080p | US$11.7 | US$2.843 | about US$0.569 |
Check the 1080p row: 1920 × 1080 × 5 × 24 ÷ 1024 = 243,000 tokens. At US$11.7 per million that is US$2.843. Vertical 1080×1920 has the same pixel count, so a 9:16 shot costs the same.
The ratio that drives everything: 0.514 ÷ 2.843 ≈ 0.18. Price scales linearly with duration, so the ratio holds for a 5-second shot and a 15-second shot alike.
The break-even: an 82% keep rate
Two numbers decide this.
- Keep rate (p): out of every 100 direct 1080p renders, how many you would actually use.
- Match rate (r): of the drafts you approve, how many produce a usable final.
Going direct, each usable shot costs 2.843 ÷ p on average. With drafts, it costs 0.514 ÷ p to find a good draft, plus 2.843 ÷ r for the final.
If every final matched its draft (r = 1), drafts win whenever p < 1 − 0.18. That is a keep rate below roughly 82%. If you already keep eight or more out of ten direct renders, skip draft mode.
Finals will not always match. With r in the formula, drafts win when r > p × 2.843 ÷ (2.843 − 0.514). Call it r > 1.22 × p. Here is what that looks like for a 5-second shot:
| Direct keep rate | Direct 1080p, per usable shot | Drafts, every final matches | Drafts, 70% of finals match |
|---|---|---|---|
| 80% | US$3.55 | US$3.49 | US$4.70 |
| 50% | US$5.69 | US$3.87 | US$5.09 |
| 25% | US$11.37 | US$4.90 | US$6.12 |
At an 80% keep rate the best case saves six cents. If three in ten finals need a redo, you lose money. At 25%, drafts still cut the cost by almost half even with a 70% match rate.
What is your keep rate? Only your own logs can say. Our one data point is an internal review: we went shot by shot through a 15-shot episode and found zero clean shots, ten usable with flaws and five unusable. That was not Seedance and not this workflow. It does suggest that keep rates for story shots run lower than most people assume.
What you can judge at 480p
Your match rate depends on whether a 480p preview shows the problems that matter. Our read:
| On a 480p draft | Reliable? | Why |
|---|---|---|
| Camera drifting or pushing in too far | Yes | Whole-frame motion survives low resolution |
| Action order, head count, blocking | Yes | Composition-level errors |
| Stray subtitles or text appearing | Mostly | You can see that text exists, not whether it is correct |
| Whether a small, distant face looks like the character | No | The face may be a few dozen pixels wide |
| Fingers, jewelry, glasses | No | Detail needs 1080p |
The top rows are where most failures live. When we audited 244 storyboard frames from a 16-episode demo drama, the most common error was action or pose not matching the script: 220 frames, or 90.2%. In a separate 15-shot video review, the camera moved on its own in at least 12 shots, and 11 of those prompts explicitly asked for a static camera. Both kinds fall in the rows of the table above that we judge readable at 480p. Note, though, that those two reviews looked at full-size storyboard images and final video; we did not test whether a 480p draft shows the same errors.
If your shots live or die on a close-up expression or small product text, a 480p draft will not tell you much. Your match rate drops, and you are in the right-hand column of the table above.
Where the usual advice goes wrong
"Test at low res, then render high res" only works with a linked draft. If you send a 480p request and then a separate 1080p request with the same prompt, you get two independent generations. BytePlus documents the seed parameter only for 1.5 pro and the 1.0 series, not for 2.0 or 2.5. The same prompt can produce very different takes. You tested the prompt, not the shot.
Draft task IDs expire. BytePlus keeps task results for 7 days. A Monday draft you want to finalize the following Tuesday is gone.
Video inputs break the 18% rule. When a request includes reference video, BytePlus applies minimum token floors. Its own example: a 5-second 1080p output with 2 to 30 seconds of input video costs US$3.062 to US$11.907. Rerun the numbers in BytePlus's calculator.
Other vendors have drafts too. Luma's pricing page on 2026-10-05 lists Ray3.14 at 4 credits per second for Draft, 10 for 540p, 20 for 720p and 80 for 1080p. That draft is 5% of the 1080p price, which widens the break-even zone a lot. How a Luma draft becomes a final is not covered in the material we read.
Limits: when draft mode is not a good fit
- Your delivery spec is 720p. Seedance 2.5 drafts only finalize to 1080p. Compare "draft + 1080p" with "direct 720p": drafts win only if 2.843 < (1.156 − 0.514) ÷ p, which means a 720p keep rate under about 23%. Usually direct 720p is cheaper.
- You are on Seedance 2.0. No draft mode is listed. If you need 4K, 2.0 is the only Seedance option anyway.
- Your keep rate is already high. Product turntables and static establishing shots often keep eight or nine in ten. Render direct.
- The shot depends on fine detail. Expressions, small text and hands cannot be judged at 480p.
- File format of the final. Seedance 2.5's 1080p output is 10-bit H.265. BytePlus warns some players cannot open it. Before posting to social platforms, we would re-encode to 8-bit H.264 mp4. That is our inference from the vendor warning.
Strip draft mode down and it is a way to learn whether a shot deserves the 1080p money before you spend it. TWno1 puts a similar check in front of the submit button. The screen shows the estimated points for a generation before you send it, and if that generation fails, the points reserved at submit come back in full. A take that renders fine but that you dislike is another matter. Like a rejected draft in the math above, it is charged, so the keep-rate logic still applies. Packages and rules are on the pricing page. To log a keep rate on your own shots, open a project in Studio.
Related reading: Seedance 2.5 vs 2.0: Clip Length, Reference Limits, 4K and Price Compared.
FAQ
How much does a Seedance 2.5 draft cost?
A Seedance 2.5 draft is billed as a normal 480p video. BytePlus ModelArk lists US$10.70 per million tokens (checked 2026-10-05), and its 5-second 16:9 example is US$0.514, about US$0.103 per second. That is roughly 18% of the US$2.843 for the same shot at 1080p.
Is the Seedance draft-mode final just an upscaled draft?
No. The second step of Seedance 2.5 draft mode is billed at the full 1080p rate (US$2.843 for 5 seconds in BytePlus's example), the same as a direct 1080p render, and the BytePlus docs do not promise a frame-for-frame match. Treat it as a new generation that reuses the draft's settings.
When is Seedance draft mode cheaper than going straight to 1080p?
Using BytePlus prices from 2026-10-05, Seedance draft mode is cheaper when you would keep fewer than about 82% of direct 1080p renders, assuming finals always match their drafts. If some finals need a redo, the final-match rate has to exceed about 1.22 times your direct keep rate.
Should I use Seedance draft mode if I only need 720p?
Usually not. Seedance 2.5 drafts can only finalize to 1080p. With BytePlus's 5-second prices (480p US$0.514, 720p US$1.156, 1080p US$2.843), a draft plus 1080p final beats direct 720p only when your 720p keep rate is below about 23%.
Does Seedance 2.0 have a draft mode?
Not according to BytePlus ModelArk documentation checked on 2026-10-05. Draft mode is listed only for Seedance 2.5 and Seedance 1.5 pro. Task results are kept for 7 days, so a draft must be finalized within that window.
References
- BytePlus ModelArk: model pricing (checked 2026-10-05)
- BytePlus ModelArk: Seedance 2.5 (checked 2026-10-05)
- BytePlus ModelArk: create a video generation task (API) (checked 2026-10-05)
- Luma pricing (checked 2026-10-05)