ByteDance has a cheaper, faster way to make AI video. Its Seedance 2.5 Flash model, announced by cloud arm BytePlus on November 12, trades top resolution for speed, and it is now live in Dreamina with developer access through BytePlus ModelArk.
What Flash changes
Flash is an accelerated cut of Seedance 2.5, ByteDance’s flagship video model. It generates clips of four to 30 seconds at 480p or 720p, with sound, from text, images or reference footage — but it stops there, with no 1080p or 4K option.
BytePlus says Flash runs at nearly twice the speed of the full model, with quality close to the flagship. The claim has no published benchmark behind it yet, so creators will have to judge the trade-off from their own renders.
Each generation can take up to 20 references: 10 images, five video clips and five audio clips, with video and audio totals capped at 30 seconds each. Clips can also be extended across rounds, and output comes in 16:9, 9:16 and five other aspect ratios.
Flash also drops the full model’s sample mode, which previews a shot before a full render. And its 8-bit colour at 480p and 720p sits a step below the flagship’s 10-bit 1080p, another reminder that this is the quick tier, not the polished one.
Price and access
The pitch is price. BytePlus’s ModelArk pricing puts Flash at about $0.15 per second of 720p video, roughly 34 percent below the flagship’s rate. At that price, an eight-second 720p clip costs about $1.20 for a single text-to-video generation today.
Flash is already available inside Dreamina, ByteDance’s CapCut-team studio, in its AI Video, AI Canvas and AI Agent workflows. It is not the default mode: creators pick it manually when speed matters more than final polish for a test render.
Developer access runs through BytePlus ModelArk, whose API listing opened on November 15, and through Volcengine Ark in China, where it is listed as doubao-seedance-2-5-flash-261115. The model is also rolling out to Jimeng, LibTV and other ByteDance apps this week.
Why it matters
Video generation is splitting into two tiers: flagship models for finished work, and cheaper fast models for the dozens of throwaway tests that come first. Flash is ByteDance’s answer to that split, aimed at creators who iterate constantly on tight budgets.
The pricing also pressures rivals. Kling, Kuaishou’s video model, has raised nearly $3 billion to scale, while Google’s Veo and OpenAI’s Sora set the quality bar — and all of them now face a flagship-grade model at a third less.
Our read
Flash makes sense as a preview tool, not a replacement. Dreamina says so itself: the model is for testing ideas fast, with final renders left to the full Seedance 2.5. That honesty is refreshing in a market full of models pitched as do-everything.
The real test is the unproven speed claim. ‘Nearly twice as fast’ with ‘quality close to the flagship’ is a vendor promise, not a measurement, and the 720p ceiling means Flash can never be the final cut. If the renders hold up, though, the price alone will pull users in.
Background
Seedance 2.5 arrived on July 31 as ByteDance Seed’s flagship video model, built for 30-second stories with sound and multimodal references. ByteDance has since pushed the family into Dreamina, CapCut and its developer platforms, while rivals from Kuaishou to Google race to cut their own generation costs.
Source: Dreamina’s Seedance 2.5 Flash page and Volcengine Ark documentation


0 Comments