24 Wan 3.0 boards at 480p, then 6 finals on Seedance 2 at 1080p
Twenty-four 6-second Wan 3.0 boards at 480p cost $9.00; finishing six shots on Seedance 2 at 1080p adds $30.618. Mixed-model cost versus Seedance-only drafts.

A cheap model can storyboard and a better one can finish. Twenty-four 6-second storyboard shots on Wan 3.0 at 480p cost $9.00 ($0.375 each), and finishing six chosen shots on Seedance 2 at 1080p, 9:16, adds $30.618 ($5.103 each): $39.618 in total.
Drafting the same 24 shots on Seedance 2 at 480p instead would cost $25.3109 for the board alone, so the Wan board saves $16.3109 before the finals.
The two plans
Plan A uses Wan 3.0 for the board. Plan B uses Seedance 2 for both stages. Seedance rows are 9:16.
| Stage | Plan A: Wan board | Plan B: Seedance 2 only |
|---|---|---|
| 24 boards at 480p, 6 s | $9.00 | $25.3109 |
| 6 finals at 1080p, 6 s | $30.618 | $30.618 |
| Total | $39.618 | $55.9289 |
| Saving of Plan A | $16.3109 |
What the Wan board does and does not prove
It proves staging, framing, the subject and the pacing of the cut. It does not prove how Seedance 2 will render that prompt: the two models read prompts differently and you cannot carry a seed across them (no v1 model takes a seed). Treat the board as a shot list with moving pictures.
A practical rule is to keep prompts structural on the board (who, where, camera) and add the model-specific style language only in the final prompt. Expect to re-roll some finals, and budget one retry per three finals: that is $10.206 for two extra.
Run the board
Use one request template and an idempotency key per shot, so a rerun of the loop never double-submits.
import os, requests
H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
shots = ["Wide: empty cafe at dawn", "Close: espresso pour", "Pan: barista smiles"]
for i, s in enumerate(shots, 1):
body = {"model": "wan-3.0", "prompt": s, "resolution": "480p", "duration": 6}
r = requests.post("https://api.sume.com/v1/videos", json=body, timeout=60,
headers={**H, "Idempotency-Key": f"board-shot-{i:02d}"})
print(i, r.status_code)Gotchas
- A shot that works on Wan may need a different prompt on Seedance 2, so keep the final's prompt separate.
- Seedance 2 accepts up to 15 seconds; Wan 3.0 accepts 2 to 30. Keep board and final lengths equal.
- A job that fails before capture is released back to your balance; a rejected board shot is spent.
Sources
Related posts
More in Use cases
- 24x36 poster from Nano Banana 2.1 at 4K: 3392x5056, about 140 dpi
Google's size table gives Nano Banana 2.1 at 4K and 2:3 as 3392x5056. On a 24x36 inch print that is about 140 dpi. Sume bills the 4K image at $0.20.
- 28-second vertical ad from four 7 s Omni clips: $5.35 with the render
Four 7-second Gemini Omni Flash 1.1 clips at 1080p cost $5.25 on Sume; a Timeline render of the 28-second cut adds $0.10, so the ad is $5.35 in total.
- 3-second product hook: Seedance 2.5 bills 4 s, Wan 3.0 bills 3 s
Seedance 2.5 on Sume starts at 4 seconds, so a 3-second hook at 720p bills 4 x $0.5778 = $2.31. Wan 3.0 accepts 2 seconds: 3 x $0.125 = $0.375.
- A 30-minute lecture video to text: three detach ranges, 33 cents
A 1,800-second video is the audio detach source ceiling. Three 600-second ranges at $0.01 plus 30 minutes of Sume STT at $0.01 a minute come to $0.33.
Written by Sume