Seedance 2.5 Is Here: 30-Second Video With Audio and 50 References

The timeline

• June 23, 2026: announced at Volcano Engine FORCE in Beijing.
• Mid-July: broader access for BytePlus API customers.
• July 31: release in ByteDance's Jimeng AI and Doubao apps.
• August 7: public developer API.

What Seedance 2.5 does differently

30 seconds in one pass

Most premium video models stop at 8 to 10 seconds. Seedance 2.5 produces up to 30 seconds, and ByteDance says clips can be extended over several rounds into videos several minutes long. For anything with a beginning, middle and end, that removes the need to stitch short clips and fight inconsistencies between them.

Audio generated with the picture

Sound is produced in the same pass as the video, so impacts, footsteps and ambience line up with what is on screen instead of being added afterwards.

Up to 50 references

A single request can carry up to 30 images, 10 video clips and 10 audio clips. In practice that means a character sheet, a product from several angles, a motion reference and a music cue can all inform the same generation.

Editing, not just generation

Timestamp-level control lets you change a specific moment of a clip's audio or video. ByteDance also highlights green screen, camera perspective changes and reference-based editing.

Seedance 2.5 vs Seedance 2.0 and 2.0 Mini

The trade-off is time: a 30-second clip takes several minutes to render, so Seedance 2.5 belongs in background jobs rather than live previews. Keep Seedance 2.0 for iteration and switch the model key for the final render.

Pricing

BytePlus bills Seedance by tokens rather than seconds. Published rates for Seedance 2.5 are around $10.70 per million tokens for requests without video input and $6.40 with video input; the cost of a clip therefore depends on its length and resolution. On VideoGenAPI, Seedance models are billed per video within your plan instead, so a long clip still counts as one video. See the AI video API pricing comparison for the full picture.

Using Seedance 2.5 on VideoGenAPI

Seedance 2.5 is rolling out in the VideoGenAPI catalog; the Seedance 2.5 API page shows its current status and parameters. Calls use the same endpoint as every other model:

{
  "model": "seedance-25",
  "prompt": "A street food vendor cooks noodles from first flame to finished bowl, wok fire, camera drifts around the stall",
  "duration": 30
}

Frequently asked questions

Sources