BytePlus has opened API access to Seedance 2.5 Flash, a lower-cost version of ByteDance’s video generation model that tops out at 720p. The November 15 rollout adds another option for developers who want to generate short clips programmatically and do not need the higher resolution available on some standard Seedance 2.5 routes.
BytePlus ModelArk now lists the standard and Flash models as available in its video generation task documentation. BeLive AI Hub lists its own Flash route as available from 06:00 UTC on November 15, and fal offers separate Flash endpoints for text, images and mixed reference material. These are different services, so an integration written for one should not assume the same model name or billing rules on another.
Short clips with optional audio
Fal documents clips from four to 30 seconds at 480p or 720p on its text-to-video endpoint. It also lists native audio, meaning the model can generate a soundtrack along with the image rather than leaving sound entirely to an editing step. Teams can still choose to add sound separately if they need more control over dialogue or a final mix.
BeLive AI Hub documents the same duration range and resolution ceiling on its Flash route. Its standard route allows 1080p and more input assets. For reference-based generation, BeLive lists up to 10 images, five videos and five audio clips for Flash, versus 30 images, 10 videos and 10 audio clips on its standard route. Those limits are specific to BeLive; they are not a guarantee for a different provider.
Why the bill depends on the input
Fal puts the cost of its text-to-video route at roughly 14 cents per second for 480p or 31 cents for 720p. A five-second 720p clip on that basis is about $1.55. That is an estimate for a text request on fal, not a universal price for the model or a price for an edited, approved clip.
For jobs with reference video, fal’s pricing notes describe a lower multiplier on the token rate, but the calculation counts the reference video’s duration as well as the generated output. A request with a long reference can therefore cost more than a text-only request for the same output length. BeLive also lists different rate rows depending on whether a video reference is supplied.
Where the standard version still fits
Flash gives developers a lower-resolution option for short social clips, rough cuts or shots that will be viewed mainly on small screens. The standard model remains relevant when a workflow requires native 1080p on a route that supports it or needs more reference assets than the Flash integration allows. An upscaled Flash clip is a separate processing step and should not be counted as the same deliverable as a native higher-resolution render.
Provider specifications establish where the model is listed and what requests it accepts. They do not establish how many attempts a particular shot will take. That distinction matters most for a production budget: a cheap render can still become an expensive approved clip if the team needs several takes to get the motion, composition or sound right.



