Last updated Nov 20, 2026

Seedance 2.5 Flash Specs: Flash vs Flagship 2.5 Compared

ModelBrief · Published 2026-11-12 · Fact-check 2026-11-16 · Reference: Volcano Engine, BytePlus, IT之家, Pandaily, WaveSpeedAI

Seedance 2.5 Flash's spec sheet is short, but the trade is clean: the price is cut ~34%, length, audio and editing are all kept, and the cuts are the resolution ceiling, Draft mode and the reference cap. This article compares Flash with the flagship Seedance 2.5 item by item — what is kept, what is cut, and what is vendor-claimed.

The comparison: Flash vs Seedance 2.5 (flagship)

CapabilitySeedance 2.5 FlashSeedance 2.5 (flagship)
Clip lengthUp to 30 sUp to 30 s (extensions available)
Resolution480p / 720p480p / 720p / 1080p, native 4K (enterprise / Runway)
Generation speedVendor claims near-2× the flagshipBaseline
Reference cap per request20 (≤10 images + ≤5 videos + ≤5 audio)50 (images + video + audio)
Built-in audioYes (synchronised audio kept)Yes
Draft modeNoneYes (480p draft → approve → final)
Video editingYesYes
Offline inferenceNot supportedFlagship 4K includes enterprise offline deployment
720p price per second¥0.99¥1.51
480p price per second¥0.44¥0.67
ReleasedNov 12, 2026 (API Nov 15)Jul 31, 2026

Three things that survived

First, 30 seconds: most "fast" tiers cut length, but Flash kept the same 30s maximum as the flagship — the most notable line on the spec sheet. Second, built-in synchronised audio: per Startup Fortune and Atlas Cloud, Flash keeps built-in audio, so finished clips need no separate soundtrack. Third, video editing: per Atlas Cloud and BytePlus, Flash supports video editing for quick variants without full regeneration.

Three things that were cut

First, resolution: Flash supports 480p/720p only — no 1080p, no 4K, no offline inference (BytePlus documentation). 4K belongs to the flagship: opened to selected enterprise customers on Oct 23 and launched natively via Runway on Nov 11 (10-bit colour, outside the US for now). Second, Draft mode: the flagship's Draft workflow is a cheap 480p draft, approve it, then render the final — Flash does not support it, an explicit trade-off of the cheap tier. Third, the reference cap: Flash allows 20 per request (≤10 images, ≤5 videos, ≤5 audio clips) against the flagship's 50; complex multi-subject, multi-asset prompts will hit the ceiling first on Flash.

The vendor's speed and quality claims

Volcano Engine makes three claims, all vendor statements with no independent verification: generation speed "nearly twice" the flagship; quality "closest to Seedance 2.5" of any tier; and "overall better than the previous flagship Seedance 2.0." WaveSpeedAI's "announced vs still open" table marks speed explicitly as "no independent measurements yet." Volcano Engine's same-prompt telescope-ad comparison clip is vendor demo material, not an independent test. The flagship's AA board scores (No. 1 image-to-video, No. 5 text-to-video) belong to the flagship, not Flash — Flash is not listed.

What doesn't exist: parameters, architecture, model card

ByteDance publishes no parameter count, architecture detail, model card or technical report for Flash; no weights, no open-source license — API and app access only. That is normal for video-model vendors, but it also means the spec sheet ends here — ask "how big" or "what architecture" and the answer is that there is none. What to watch: an AA video-board listing for Flash, the first independent same-prompt test, and whether a 1080p tier appears.

FAQ

Can Flash output 1080p?

No. The ceiling is 720p; 1080p, 4K and offline inference are all unsupported. For 1080p or 4K, use the flagship Seedance 2.5.

Does Flash have a Draft mode?

No. Draft is a flagship feature (cheap 480p draft → approve → final render); Flash does not support it.

How many reference images can I use at once?

Up to 20 references per request: ≤10 images, ≤5 videos, ≤5 audio clips. The flagship allows 50.

Is the 30-second length real?

Yes. Flash generates up to 30s per request, the same as the flagship — what sets it apart from most "fast" tiers.

ModelBrief's take: Flash's spec trade is honest: ~34% cheaper, with length, audio and editing kept, and the resolution ceiling, Draft mode and reference cap cut. The only dishonesty risk sits in the two vendor claims about speed and quality — with no independent measurement, they are advertising copy, not specs.

Reference: Volcano Engine official WeChat account (launch announcement, Nov 12, 2026) · BytePlus ModelArk docs (as cited by Startup Fortune, CellCog and Atlas Cloud, Nov 13–15, 2026) · IT之家 (Nov 12, 2026) · Pandaily (Nov 12, 2026) · WaveSpeedAI (Nov 13, 2026) (rewritten and organized by ModelBrief; vendor claims are labelled as such)