Home Blog

How to Use the Seedance 2.5 Flash API

Seedance 2.5 Flash API blog cover

Nov 15, 2026

EmpirioLabs AI

Disclosure: This article was written with AI assistance and reviewed by EmpirioLabs AI.

Seedance 2.5 Flash is available on EmpirioLabs. It is ByteDance's faster Seedance 2.5, released on November 12, 2026: the same text to video, image to video, multimodal references, editing and extension as Seedance 2.5, at 480p and 720p, with clips of up to 30 seconds and native synchronized audio. It works in the API and in the Playground.

Watch the launch film

Every clip in the film was made by Seedance 2.5 Flash through the EmpirioLabs API at 720p, including a whale breaching beside a boat, a steam train bursting out of a tunnel, a figure skater's jump and a storm wave over a lighthouse; the surfer shot plays its own generated sound. The score was generated by ACE-Step on EmpirioLabs. Watch it on YouTube.

How it compares

Seedance 2.5 FlashSeedance 2.5Seedance 2.0 Fast
Longest clip30 seconds30 seconds15 seconds
Resolutions480p, 720p480p, 720p, 1080p480p, 720p
Reference imagesUp to 10Up to 30Up to 9
Reference videos and audio clipsUp to 5 of eachUp to 10 of eachUp to 3 of each
Audio as the only referenceYesYesNo, it needs an image or a video
Editing and extending a clipYesYesYes
Draft modeNoYesNo
Output formatsMP4, MOVMP4, MOVMP4

In our tests, a 5 second 720p clip with audio finished in about 90 seconds from submit to the finished file.

What it supports

  • Text to video. A prompt on its own.
  • First frame, or first and last frame. Send image, and optionally image_end, to animate a still or to travel between two.
  • Multimodal references. Send up to 10 reference_images, 5 videos and 5 audios in one request. Audio works with no image or video at all.
  • Editing. Attach a clip and say what to change: add, remove or replace something in the footage.
  • Extension. Attach a clip and ask the model to continue it.
  • Native audio. Dialogue, effects and ambience are generated with the picture. Set generate_audio to false for a silent clip.

Clips run from 4 to 30 seconds at 24 frames per second, in 16:9, 9:16, 1:1, 4:3, 3:4 or 21:9, or adaptive to let the model choose. Output is MP4 by default; MOV carries higher color precision for grading and compositing.

Generate a video

Video generation is asynchronous. Submit the job, then poll it.

curl -X POST https://api.empiriolabs.ai/v1/videos/generations \
  -H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2-5-flash",
    "prompt": "A black steam locomotive bursts out of a mountain tunnel into a snowstorm, white steam billowing over the carriages. Sound: a piercing whistle, heavy chugging, wheels clattering on the rails.",
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "duration": 5
  }'

# -> {"job_id": "...", "status": "processing", "poll_url": "/v1/jobs/..."}

curl https://api.empiriolabs.ai/v1/jobs/JOB_ID \
  -H "Authorization: Bearer $EMPIRIOLABS_API_KEY"

When status is completed, the clip is in result.data[0].url, and result.usage shows what was billed:

"usage": {
  "mode": "reference",
  "resolution": "480p",
  "aspect_ratio": "16:9",
  "duration_seconds": 5.0,
  "billable_units": 5.0416,
  "billing_unit": "second",
  "pricing_tier_label": "T2V/I2V 480P"
}

The same flow in Python:

import time
import requests

API = "https://api.empiriolabs.ai/v1"
HEADERS = {"Authorization": "Bearer YOUR_EMPIRIOLABS_API_KEY"}

job = requests.post(f"{API}/videos/generations", headers=HEADERS, json={
    "model": "seedance-2-5-flash",
    "prompt": "A humpback whale breaches beside a small boat, crashing back in a towering splash. Sound: the splash, a deep whale call.",
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "duration": 5,
}).json()

while True:
    status = requests.get(f"{API}/jobs/{job['job_id']}", headers=HEADERS).json()
    if status["status"] in ("completed", "failed", "cancelled"):
        break
    time.sleep(5)

print(status["result"]["data"][0]["url"])

Build from references

Attach images, videos and audio clips and refer to each one in the prompt by its position, per type: @Image1, @Image2, @Video1, @Audio1.

{
  "model": "seedance-2-5-flash",
  "prompt": "The brass compass from @Image1 rests on the picnic blanket from @Image2 while a breeze ripples the cloth. Slow push in. Sound: wind in the grass, birdsong.",
  "reference_images": ["https://example.com/compass.jpg", "https://example.com/blanket.jpg"],
  "resolution": "720p",
  "duration": 4
}

Audio can lead on its own:

{
  "model": "seedance-2-5-flash",
  "prompt": "A cozy animated kitchen: a kettle on the stove whistles and steams in time with @Audio1.",
  "audios": ["https://example.com/sound.mp3"],
  "resolution": "480p",
  "duration": 5
}

Edit or extend a clip

Attach the clip and describe the change or the continuation in the prompt. The model reads the intent from your wording.

# Edit: follows the source clip's length and aspect ratio
{"model": "seedance-2-5-flash",
 "prompt": "Edit @Video1: make it snow heavily, keep everything else the same.",
 "videos": ["https://example.com/street.mp4"]}

# Extend: continue the shot
{"model": "seedance-2-5-flash", "duration": 8,
 "prompt": "Continue @Video1 as the camera keeps drifting forward.",
 "videos": ["https://example.com/street.mp4"]}

The model sets an edit's length from its source clip, so duration does not apply to it. Edits, extensions and first-frame jobs take their aspect ratio from the input.

Good to know

  • 480p and 720p. For 1080p, use Seedance 2.5.
  • No draft mode. A request with draft or draft_task_id returns an error. Seedance 2.5 offers drafts.
  • Reference limits. Reference videos and audio clips run 2 to 30 seconds each and 30 seconds in total.
  • Real faces. Reference uploads that contain real human faces are rejected by the model.
  • Generated audio is checked. A clip whose generated audio is too close to existing music is not returned and is not billed; rewrite the sound part of the prompt and try again.

Pricing

Per second. Seedance 2.5 Flash is billed per second of generated video at the rate for its resolution. When you attach a reference video, the reference clip's seconds are billed along with the generated seconds at the lower video-input rate. You can pay as you go, or choose a plan with weekly video allowances. Current rates are on the model page and the pricing page.

Start building

Try Seedance 2.5 Flash in the Playground, read the full parameter reference in the API docs, or compare it with the other video models in the model catalog. For the full-resolution model of the family, see How to Use the Seedance 2.5 API.

Ready to use better endpoints?

Explore our models, or contact us about business inquiries, custom deployments, or anything else.