Try Seedance 2.5 for $1.08: a 4-second 480p first request on Sume

The cheapest way to try Seedance 2.5 on Sume is a 4-second 480p clip at $1.08. The request, the limits, and what 720p and 1080p would cost instead.

5 min readSume
All posts

The cheapest first request for Seedance 2.5 on Sume is a 4-second clip at 480p, which bills $1.08. Four seconds is the model's minimum duration on Sume, and 480p is its lowest resolution. Seedance 2.5 launched on 2026-07-31 and, per Magic Hour's tracker (read 2026-10-06), goes up to 30 seconds with native audio and video together.

On Sume the id is seedance-2.5, and its limits come from the video generation docs and the Video Router docs: 4 to 30 seconds, 480p, 720p and 1080p, first and last frames, and image, video and audio references. It does not take a seed or a bitrate_mode.

The first request

You need an API key and a funded workspace. The request below is valid as written; it returns 202 and a polling URL. Poll that URL until completed, then download unsigned_urls[0].

import os, requests

r = requests.post(
    "https://api.sume.com/v1/videos",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json={
        "model": "seedance-2.5",
        "prompt": "A paper boat drifts down a rain gutter, close-up, shallow focus",
        "duration": 4,
        "resolution": "480p",
        "aspect_ratio": "16:9",
    },
    timeout=60,
)
print(r.status_code, r.json())

What each step up costs

The price is per video token, so it grows with frame size and with seconds. These are billed totals at 16:9, list times 1.25, rounded up to the cent.

Billed Sume totals for Seedance 2.5 at 16:9, read 2026-10-06.
Resolution4 s8 s15 s30 s
480p$1.08$2.15$4.04$8.07
720p$2.32$4.63$8.67$17.34
1080p$5.69$11.38$21.33$42.65

What a first test should check

Use the 480p run to judge motion, composition and sound, not texture. Low resolution hides fine detail, so a clip that looks soft is not evidence about the model; a clip that moves wrongly is. If the motion is right, rerun at 720p, which costs about twice as much, and look at faces, hands and any on-screen text. Only then pay for 1080p.

Because there is no seed, the 720p rerun is a new take rather than a sharper copy of the draft. That is worth knowing before you fall in love with a 480p result: you may not get the same clip back. If a specific frame matters, generate the still first and pass it as a frame_images first frame, so the opening is fixed.

If the request fails

A 402 insufficient_credits means the workspace balance is below the reserve; top up and resubmit. A 400 unsupported_capability means a value is outside the model's list, for example a 3-second duration, which is below Seedance's 4-second minimum.

For the full set of aspect ratios, including auto and adaptive, see Seedance 2.5 aspect ratios. For the money side of the longest job, see a 30-second 1080p reserve.

How Seedance 2.5 compares with the cheaper ids

Seedance 2.5 is the model to pick for the longest clips, and it earns its rate with the widest input list: first and last frames, image, video and audio references, and eight aspect-ratio values. If you only need text-to-video and the lowest price, Wan 3.0 is $0.25 for the same 4 seconds at 480p, a quarter of the Seedance figure. If you need to lock the look with a reference clip or a voice track, Seedance is the one to test. Seedance 2.0 and its Fast and Mini variants stay in the catalog too, capped at 15 seconds, for cases where 30 seconds is not needed. A first test on 2.5 is a good way to decide whether the extra length and reference support are worth the premium for your work.

A second test worth running

After the plain text-to-video run, try one reference. Send a product photo in input_references as an image_url, and ask for a slow push-in. A single reference image with no frame is treated as reference-to-video and conditions the whole clip, while a frame_images entry with first_frame pins only the opening. Knowing which one you want avoids a surprising result.

Sources

Related posts

More in Models

All Models posts

Written by Sume