Seedance 2.5 on Flixly: 30-Second Shots with Synchronized Audio

Seedance 2.5 is ByteDance's newest video model, and it is live on Flixly in all three modes: text-to-video, image-to-video, and reference-to-video.

What changed since Seedance 2.0

• Up to 30 seconds in one generation. Seedance 2.0 topped out at 15. Thirty seconds is long enough for a complete beat — a setup, a turn, and a landing — without stitching separate clips and fighting continuity between them.
• Audio in the same pass. Dialogue, music, ambience, and effects are generated with the picture rather than dubbed over it afterward, so timing and lip movement line up. Put spoken lines in double quotes in your prompt.
• Up to 50 references. Reference-to-video takes images, video clips, and audio together — up to fifty items — and you address them directly in the prompt as [Image1], [Video1], [Audio1].

Specs

How to use it

1. Open the Video Generator in the dashboard and pick the tab you need — Text, Image, or Reference.
2. Pick Seedance 2.5 from the model selector.
3. Set resolution and duration. Draft at 480p and short lengths — it costs less than half of 720p — then re-run the prompt you like at 720p and full length.
4. Write the prompt as direction, not description: the subject, the action, the camera move, and any dialogue in double quotes.
5. For image-to-video, upload a start frame; add an end frame if you want the shot to land on a specific composition.

Two things worth knowing about cost

Seedance 2.5 is billed by output length, so a 30-second clip costs roughly six times a 5-second one. The credit estimate on the page always reflects the exact settings you've chosen before you generate.

The second is less obvious: in reference-to-video, a video reference adds its own length to what you're billed for. A 20-second reference driving a 6-second output is billed on 26 seconds, not 6. Image and audio references don't do this — only video.

Writing a prompt for it

Seedance 2.5 responds to direction, not description. A prompt that reads like a shot list beats one that reads like a caption.

• Say what moves. "A chef plates a dish" gives the model nothing to animate. "A chef sets the last herb on the plate, then slides it across the pass as the camera pushes in" gives it a beginning, an action and an end.
• Put dialogue in double quotes. Anything quoted is generated as spoken audio, lip-synced in the same pass. She turns to the window and says "we should have left an hour ago" gets you the line and the mouth movement together.
• Direct the sound. Ambience, score and effects are all generated. Naming them — rain on glass, a low sustained string, the click of a latch — is the difference between a scene with sound and a scene with room tone.
• Address references by index. In reference-to-video the prompt refers to your uploads as [Image1], [Video1], [Audio1]. Say what each one is for: "the woman from [Image1], moving with the camera rhythm of [Video1]."

Which mode to pick

If you're deciding between Seedance 2.5 and Seedance 2.0, the trade is length and references against resolution: 2.5 goes to 30 seconds and 50 references but stops at 720p, while 2.0 reaches 1080p and 4K in a 15-second window.

What it costs, and the one surprise

Seedance 2.5 is billed by output length, so a 30-second clip costs roughly six times a 5-second one, and 480p runs a little under half of 720p. The cheapest way to work is to draft short at 480p, then re-run only the prompt you like at full length and 720p.

The surprise is in reference-to-video: a video reference adds its own length to the bill. A 20-second clip driving a 6-second output is billed on 26 seconds, not 6. Image and audio references don't do this. Trim reference clips to the part that actually carries the motion you want.

Going deeper

For the full spec sheet — what ByteDance confirmed, what the API settled, and which claims circulating about this model are still unsupported — read Seedance 2.5: What's Real, What Isn't, and How to Run It.