Video models

Seedance 2.5 Flash: 30-second AI video, twice as fast

BytePlus has announced Seedance 2.5 Flash, a cheaper tier of its Seedance 2.5 video model that it says runs about twice as fast and costs about 34% less. It is live in Dreamina now, with the ModelArk API opening on 15 November.

· Announced 12 November 2026 · By

Scroll
A long-haired figure in dark robes stands in a moonlit forest, with a bolt of lightning in one hand and a glowing sigil in the other, an image from Dreamina's Seedance 2.5 Flash page
Image: Dreamina (Seedance 2.5 Flash page)

BytePlus, ByteDance’s cloud arm, announced Dreamina Seedance 2.5 Flash on 12 November 2026. It is a second, cheaper tier of the Seedance 2.5 video model, aimed at people who need many drafts rather than one polished render.

What Flash offers

BytePlus describes Flash as “results close to flagship Seedance 2.5, with nearly twice the generation speed”. According to its announcement thread and ModelArk price page:

  • Length: one clip of up to 30 seconds per request.
  • Resolution: 480p or 720p only. 1080p and 4K stay with the flagship model.
  • Inputs: up to 10 images, 5 video clips and 5 audio clips in a single request, plus video editing and precise instruction following.
  • Speed and cost: about twice as fast as the flagship, and about 34% cheaper.

Price and access

For 720p text-to-video and image-to-video, BytePlus gives a price of USD 0.152 per second, which it says is roughly 34% below Seedance 2.5.

Flash is already selectable in Dreamina, ByteDance’s creator app, in its video tool, its canvas and its agent. The API on BytePlus ModelArk opens at 14:00 (UTC+8) on 15 November. BytePlus’s documentation says an account needs a balance above USD 30, a savings plan or a resource pack before enabling the Seedance 2.5 series.

Limits

The speed and quality claims come from BytePlus. We have not seen independent tests, and BytePlus has not published a speed benchmark. Flash gives up the top resolutions, so it is a drafting tier, not a final-render tier. Dreamina’s own page frames it that way: test motion, framing and camera moves first, then commit to a full render.

Our take

Cheap, fast video matters for interactive worlds because a world needs many clips, not one. Cut-scenes, character reactions and place establishing shots can all be tried at low cost, and the full model kept for what ships. A 30-second limit at 720p is enough to test a scene before committing to it.

It also fits a pattern we see across generative interactive film: the creative work moves from making a single perfect shot to choosing among many. That makes the authored story, the state that decides which clip is needed next, the part worth protecting. Video models are the renderer, not the author.


Sources

← All news