Seedance 2.5: write the sound into the prompt (4s, 5s, 6s, 720p, 47cr)

Seedance 2.5 has native audio. A silent prompt still gets a soundtrack you did not choose.

Seedance 2.5 is ByteDance's Dreamina text-to-video row: native 30-second single-shot video at up to 720p from one text prompt, reasoning about the whole shot at once so motion, lighting and subject identity stay coherent from first frame to last. Audio is on. Credits: 47. Durations: every second from 4s through 30s. If you type a picture and omit the room, you still get a score. You just did not pick it.

Native audio is not a toggle you forgot

audio is true. The feature list includes audio_sync. This is not a silent B-roll model with a "maybe sound" footnote. Write the sound into the prompt: who speaks, in what language, how loud the street is, whether the hit is a glass or a door, whether the last second is breath or music.

A silent prompt on a native-audio row is how you spend 47 credits on a take you will mute in the editor. Mute is not a strategy. If you needed mute, pick a row whose audio field is false or empty.

The glossary term is native audio. The generator door is AI video generator. Models with audio is the ranked list if you are still shopping families. This page is only Seedance 2.5.

One shot, up to thirty seconds

The description is the constraint: a native 30-second single-shot. The duration picker is not a timeline. It is the length of one reasoned take — 4s, 5s, 6s, all the way to 30s. There is no cut, no second angle, no "and then the product spins" as a new clip inside the same generate.

If the brief is a sequence, you are writing a sequence into one prompt and hoping the single shot covers it. That is usually a worse idea than two shorter Seedance takes. If the brief is one continuous move — a walk, a pour, a hold — this is the row that was built for the hold.

Resolution tops out at 720p. Qualities listed: 480p and 720p. Do not prompt 4K. You will not get it here. Aspect ratios include 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, adaptive, auto. Pick the one the platform needs before you spend 47 credits.

720p and 47 credits are the row

Quote it and stop decorating:

• text-to-video
• native audio
• 4s–30s
• 720p
• 47 credits
• requires_image: false (prompt-only)

ByteDance stills are Seedream. ByteDance long takes with sound are this. Mixing them in your head is how a 4K poster prompt lands on a 720p talking clip. The roster is ByteDance.

Because the model reasons about the whole shot at once, the prompt has to describe the whole shot: first frame, last frame, what the subject does in between, and the soundtrack that covers it. A prompt that only names the subject is how identity holds and the audio invents a radio.

Write the room before you generate

Room tone is a line. Dialogue is a line. Foley is a line. If there is no speech, say so. If there is speech, write the line — do not write "someone talks." Thirty seconds of unnamed talking is a tax.

Start short when you are learning the row: 4s, 5s, 6s. The title of this post names those because they are the cheap way to hear whether the soundtrack matches. Then extend toward 30s only when the sound and the identity already work. The catalog will let you jump to 30s on the first try. That is not the same as a good idea.

Seedance 2.5 is a long single shot with a score. Prompt both, or do not press generate.

FAQ

Can I make Seedance 2.5 silent?

Not as a catalog fact. Audio is true. A prompt that never mentions sound still returns a soundtrack. If silence is the brief, this is the wrong model.

Why 720p and not 4K?

Because the spec's max output is 720p. Qualities are 480p and 720p. If the plate needs 4K, generate a still elsewhere and pick a 4K video row — or accept 720p as the Seedance 2.5 job.

Is 47 credits per second or per clip?

The catalog lists 47 credits on the model. Do not convert that into a USD price. Check the live quote in the app before a 30-second run if you need the exact job total.

Does it take a start image?

requires_image is false and the category is text-to-video. This is a prompt-only single shot. If the still is already locked, use an image-to-video row instead of asking Seedance to re-imagine the product for 30 seconds.