Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

ByteDance opens cheap Seedance 2.5 Flash API, with a 720p ceiling

ByteDance put Seedance 2.5 Flash on BytePlus ModelArk on November 15, opening API access to a video model priced at $7.0 per million tokens versus $10.70 for the full Seedance 2.5. It keeps 30-second clips with built-in audio but caps output at 720p and cuts reference files from 50 to 20.

Published

ByteDance has opened Seedance 2.5 Flash, the cheaper tier of its video generation model, to API users. According to Startup Fortune, the model was announced on November 12 and the API went live on BytePlus ModelArk on November 15, giving developers direct access to what amounts to the affordable workhorse of the Seedance line.

The pricing is the story. The report, citing CellCog's breakdown of the BytePlus announcement, puts Flash at $7.0 per million tokens for 480p and 720p output, against $10.70 for the full Seedance 2.5. A five-second 720p clip works out to roughly $0.76, or about $0.15 per second.

Chinese-language coverage describes the same arithmetic in yuan. PANews reported that Volcano Engine, ByteDance's cloud arm, prices 720p text-to-video and image-to-video at about 0.99 yuan per second, roughly 34% below Seedance 2.5, with generation speed about doubled. Consumer access opened earlier, on November 12, through Dreamina, LibTV and Updream.

The discount comes with hard limits. The full Seedance 2.5 launched on July 31 with native 1080p, clips of up to 30 seconds with built-in audio and room for up to 50 reference files in a single generation. Flash keeps the 30-second length and the audio, but drops the top of the resolution ladder.

Per CellCog's summary of the ModelArk documentation, 1080p, 4K and offline inference are unsupported, leaving 480p and 720p. Reference limits shrink as well: PANews says a Flash input can carry up to 10 images, 5 videos and 5 audio clips, 20 files in all, well under the flagship's 50.

ByteDance's own wording is careful. It says Flash can edit video and beats the previous flagship, Seedance 2.0, on instruction following, motion and audio, and that its output is the closest of any ByteDance model to Seedance 2.5. On speed, BytePlus claims nearly twice the generation speed of the flagship, though CellCog noted the documentation gave no timing measurements and that it had not tested the model itself when it published.

For teams building video products on someone else's model, the real change is not the discount but a published price sheet for the draft-then-finish workflow. A product team that wants six takes of a 15-second clip before picking one now spends least exactly where it burns the most, and 720p is fine for a preview. A broadcast cut or a paid placement that needs 1080p, or a brand that needs the same character or logo to hold steady across shots, still has to go back to the full model.

What to watch is the competitive response and the fine print. The report notes that OpenAI, Google and Kuaishou's Kling have not announced a like-for-like cheaper, faster 30-second model with native audio, leaving ByteDance to set the reference price for this class of clip. Volume discounts, geographic availability and benchmark results have not been disclosed, and the price figures come from vendor platforms and secondary write-ups rather than independent testing, so teams should run their own prompts through both models before moving a production workload.

Why it matters

Cheaper per-clip pricing shifts video generation toward metered, draft-first workflows, while final 1080p work stays with the flagship model.

ByteDanceAI Video
Back to realtime news

Nearby Updates

All