Seedance 2.5 Native Audio: Generate Video with Sound via API (Tutorial)

Published August 25, 2026 · 6 min read

Seedance 2.5's headline feature is native audio: the soundtrack is generated together with the video in a single pass — no separate TTS step, no lip-sync post-processing. This tutorial shows exactly how to enable it and what the metering looks like.

The full flow in Python

import requests, time

BASE = "https://aiapi-pro.com/v1"
H = {"Authorization": "Bearer sk-novai-...", "Content-Type": "application/json"}

# 1) submit the generation task
task = requests.post(BASE + "/video/generations", headers=H, json={
    "model": "doubao-seedance-2.5",
    "prompt": "A street-food market at dusk, steam rising, vendor calling out, "
              "sizzle of the wok, ambient crowd murmur",
    "duration": 10,
    "resolution": "720p",
    "generate_audio": True,
}).json()

# 2) poll until finished (status: running -> succeeded)
while True:
    r = requests.get(BASE + f"/video/generations/{task['task_id']}", headers=H).json()
    if r["status"] in ("succeeded", "failed"):
        break
    time.sleep(10)

print(r["video_url"])   # MP4 with synchronized soundtrack
print(r["duration"], r["resolution"])   # echo of what was delivered

The same flow works as two curl calls — submit, then poll — which is how our playground drives it under the hood.

What the metered response tells you

Upstream bills in tokens, not seconds. A real 10-second, 720p, audio-on generation returned:

{
  "status": "succeeded",
  "resolution": "720p",
  "duration": 10,
  "generate_audio": true,
  "usage": { "completion_tokens": 216900, "total_tokens": 216900 }
}

On NovAI you never do that math: billing is a flat $0.32/second at 720p and $0.14/second at 480p, audio included, with automatic refunds on any failed generation.

Prompting for sound

The audio model listens to your prompt. Describe the soundscape explicitly — "rain on a tin roof, distant thunder, a dog barking once" — and it will stage those sounds against the visuals. Prompts with no audio description still get a plausible ambient track, but explicit cues produce markedly better sync.

Cost tip: iterate on framing and motion at 480p ($0.14/s), then render the final cut at 720p ($0.32/s). Two 480p drafts cost less than one 720p take.

Trial accounts can render one 5-second 480p clip per day with the $2 welcome credit; your first top-up of $1+ unlocks 720p, 15-second clips and unlimited daily renders.

Try it free on NovAI

$2 free credit · no card required · OpenAI-compatible API

Start Free →

Or test it in the playground →

Related: Seedance 2.5 per-second pricing · Seedance 2.5 API basics