Seedance 2.5 vs Kling 3 on Sume: what each catalog row accepts

On Sume, Seedance 2.5 (seedance-2.5) runs 4 to 30 seconds at 480p, 720p and 1080p and accepts image, video and audio references. Kling 3 (kling-3) runs 4 to 15 seconds at 720p and 1080p and accepts no references.

Choose Seedance 2.5 for anything past 15 seconds or anything driven by reference assets. Choose Kling 3 for a short clip where you want a silent option and a flat per-second price.

Side by side

These values come from the capability records in the Sume Video Router catalog, which also feed GET /v1/videos/models.

What ByteDance says

ByteDance Seed's Seedance 2.5 page describes generation of up to 30 seconds in a single pass with the option to extend twice, and calls it an audio-video joint generation model (read 2026-10-07). The page does not state output resolution. Sume does not expose the extend step: each Sume job is one generation of 4 to 30 seconds.

Billing differs in kind

Kling 3 is simple: $0.14 or $0.21 per second after the 1.25 margin, so ten seconds is $1.40 or $2.10. Seedance is billed on video tokens, so cost scales with resolution and length in a way you read from pricing_skus, and reference videos add a priced allowance. For Seedance numbers, read pricing_skus from the models endpoint rather than guessing from a per-second rate.

The docs state that Sume reserves the billable amount on submit and captures on completion, so the amount you see at submit is the amount held.

Decision rule

Read the request you want to send and check three things: how long, whether you have reference files, and which resolution. Over 15 seconds or any reference file means Seedance 2.5. A short, silent-capable clip in 16:9, 9:16 or 1:1 means Kling 3.

For a draft pass, seedance-2-mini and seedance-2-fast are cheaper Seedance rows with a 4 to 15 second range.

• Over 15 seconds: Seedance 2.5 only (Wan 3.0 also goes to 30).
• Any reference file: Seedance 2.5, not Kling.
• Silent clip: Kling with generate_audio: false.