Seedance 2.5 Flash: multimodal video with native audio at 480p or 720p

Seedance 2.5 Flash is ByteDance Seed's fast, lower-cost Seedance 2.5 tier: multimodal video with synchronized native audio from text, image, video and audio references, at 480p or 720p.

Seedance 2.5 Flash is the fast, lower-cost tier of ByteDance Seed's Seedance 2.5 video model. It generates the picture and its synchronized native audio in one pass, so dialogue, effects and ambience land with the motion instead of being dubbed on later. It takes text, image, video and audio references, and edits or extends clips in place. Renders run 4 to 30 seconds at 480p or 720p — draft at 480p, finish at 720p for volume work.

ModelSeedance 2.5 FlashMultimodal AI video with native audio, editing, 480p/720p.

What it does well

Native audio generated with the motion

Seedance 2.5 Flash renders sound in the same pass as the picture. Keep generate_audio on and the diegetic effects, ambience and music track what happens on screen. In this table-tennis rally, each paddle strike cracks on the visible hit, the ball bounces on the table, and the shoes squeak on the varnished floor. Name the sounds you want in the prompt and the model places them in time.

Text-to-video with native soundfor Seedance 2.5 Flash
Two players rally at speed across a blue table-tennis table in a school gymnasium: the white ball snaps back and forth over the net, each paddle clips it with a sharp hollow pock, and rubber-soled shoes squeak on the varnished floor as the camera tracks the volley from the side. Natural diegetic sound - the crack of each hit, the ball bouncing on the table, squeaking shoes, and the hollow echo of the hall.
Try it

Keep a character consistent with a reference image

Give Seedance 2.5 Flash up to 10 reference images and cite them inline as @Image 1. The model holds the subject's identity through the shot without needing a first frame. Here one photo of a tin wind-up robot carries its exact design — riveted body, square eyes, yellow antenna knob and silver key — as it marches across the bench.

  • @Image 1
The reference image, cited in the prompt as @Image 1
Reference-image identity consistencyfor Seedance 2.5 Flash
Keep the blue-and-red tin wind-up toy robot from @Image 1 exactly as shown — its boxy riveted body, round yellow antenna knob, two square eyes and the large silver key in its back. The robot marches stiffly across a sunlit workshop bench past a row of colourful wooden building blocks, arms swinging and the wind-up key turning. Static camera. Clockwork ticking and the clack of little feet on wood.
Try it

Examples to try

One more worked example — load the prompt, keep the settings, and run it.

Native audio dialoguefor Seedance 2.5 Flash
Behind the counter of a small bakery, a baker in a flour-dusted blue apron holds up a crusty sourdough loaf, looks into the camera and says warmly, "Fresh from the oven — still warm." Wooden shelves of bread behind him. Medium close-up, static camera. Quiet bakery ambience with a coffee machine hissing softly. No music, no subtitles.
Try it

The baker's lips move in sync with the spoken line over quiet bakery ambience, so the speech is generated with the picture rather than dubbed on. Put the line in quotes and add 'no subtitles' to keep text off the frame.

Recommended settings

SettingValueWhen
duration5 (4–30, or -1)5s for quick shots; raise toward 30s for longer scenes; -1 (auto) is required for edits
resolution720p480p to iterate fast, 720p for finals; 1080p and 4K live on the sibling tiers
aspect_ratio16:9A fixed ratio for text-to-video; adaptive with a first frame, reference video, or edit
generate_audiotrueKeep on for any scene with speech, effects or music
seed42Fix a value to iterate the same scene; -1 for a random result
bitrate_modestandardSwitch to high for maximum fidelity and fewer compression artifacts
return_last_framefalseSet true to chain the next shot from the final frame

Good to know

  • The Flash tier renders at 480p and 720p only — use Seedance 2.5 for 1080p and Seedance 2.5 4K for 4K.
  • first_frame_url does not accept real human faces, per ByteDance content policy.
  • A reference video is billed for its full length at the video-input rate, on top of the output.
  • Reference-video and edit or extend runs need aspect_ratio set to adaptive and duration set to -1, so the output follows the input clip.

Use it from the API

Send the model slug and your inputs to the Segmind API. This is the default text-to-video example, with native audio on.

APIText-to-video with native audioAPI reference
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind

result = segmind.run(
    "seedance-2.5-flash",
    prompt="Two players rally at speed across a blue table-tennis table in a school gymnasium: the white ball snaps back and forth over the net, each paddle clips it with a sharp hollow pock, and rubber-soled shoes squeak on the varnished floor as the camera tracks the volley from the side. Natural diegetic sound - the crack of each hit, the ball bouncing on the table, squeaking shoes, and the hollow echo of the hall.",
    first_frame_url="",
    last_frame_url="",
    reference_images=[],
    reference_videos=[],
    reference_audios=[],
    duration=5,
    resolution="720p",
    aspect_ratio="16:9",
    generate_audio=True,
    seed=42,
    return_last_frame=False,
    skip_moderation=False,
    bitrate_mode="standard",
)
print(result.get("output"))
# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/seedance-2.5-flash" \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "prompt": "Two players rally at speed across a blue table-tennis table in a school gymnasium: the white ball snaps back and forth over the net, each paddle clips it with a sharp hollow pock, and rubber-soled shoes squeak on the varnished floor as the camera tracks the volley from the side. Natural diegetic sound - the crack of each hit, the ball bouncing on the table, squeaking shoes, and the hollow echo of the hall.",
  "first_frame_url": "",
  "last_frame_url": "",
  "reference_images": [],
  "reference_videos": [],
  "reference_audios": [],
  "duration": 5,
  "resolution": "720p",
  "aspect_ratio": "16:9",
  "generate_audio": true,
  "seed": 42,
  "return_last_frame": false,
  "skip_moderation": false,
  "bitrate_mode": "standard"
}' | jq -r .request_id)

# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
  -H "x-api-key: YOUR_API_KEY"

# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
  -H "x-api-key: YOUR_API_KEY"

Resources

  • #image-to-video
  • #bytedance-seed
  • #model-launch
  • #native-audio
  • #video-generation
AM

Abhilash Mirji

@abhilashmirji

Product engineer at Segmind, shipping new AI models faster than my cats can knock things off the desk. 11+ years turning ambiguous problems into dependable products. Off-call: drumming, scuba diving, and getting outvoted by four cats and two dogs.

View profile

Posts

More from Abhilash Mirji

Make your own with Segmind

Run every model and Pixelflow in this post, and build your own workflows.

Start creating