Bytedance Seedance 2.5 Text To Video API Documentation

Bytedance Seedance 2.5 Text To Video API Documentation

Playground

Seedance 2.5 (Text-to-Video) generates Hollywood-grade cinematic videos from text prompts with native audio-visual synchronization, director-level camera and lighting control, and exceptional motion stability. Built on Seed’s unified multimodal architecture, it leads on instruction adherence, motion quality, and visual aesthetics.

Features

Seedance 2.5 Text-to-Video generates cinematic videos from text prompts with native audio, strong motion stability, and director-style control over camera movement, lighting, mood, subject action, and scene progression. It also supports optional reference images, videos, and audio for stronger visual, motion, or soundtrack guidance.

Why Choose This?

• Text-to-video generation
  Generate cinematic videos directly from natural-language prompts.

• Native audio generation
  Generate synchronized audio together with the video in a single workflow.

• Director-style prompt control
  Control camera movement, lighting, shadows, character performance, atmosphere, and visual style through prompts.

• Reference-guided generation
  Use reference images, videos, or audio to guide style, characters, composition, motion, or soundtrack direction.

• Flexible output settings
  Choose aspect ratio, duration, resolution, and audio generation settings for different creative workflows.

• High-resolution output
  Generate videos in 480p, 720p, 1080p, or 4k.

Parameters

How to Use

1. Write your prompt — Describe the scene with cinematic detail, including subject, action, lighting, mood, camera movement, and style.
2. Choose aspect ratio — Use 16:9 for widescreen, 9:16 for vertical content, or another supported layout.
3. Set duration — Choose a video length from 4 to 30 seconds.
4. Choose resolution — Select 480p, 720p, 1080p, or 4k.
5. Add references optional — Use reference images for style or character guidance, reference videos for motion guidance, and reference audio for soundtrack direction.
6. Configure audio optional — Keep generate_audio enabled when synchronized audio is needed.
7. Submit — Generate the final video and retrieve the output URL.

Writing Effective Prompts

Most weak results come from thin prompts. Seedance reads a prompt like a shot list — the more you direct it, the more it delivers. Structure your prompt in three layers.

1. One-sentence summary

Open with a single line that fixes the essentials:

> subject + setting + action + style + camera

A lone astronaut walks across a red desert at dusk, cinematic sci-fi, slow low-angle tracking shot.

2. Shot-by-shot timeline

For anything with a beginning, middle, and end, break the duration into timed segments (1-second granularity). Keep the timeline continuous — no gaps.

> 0-3s: The astronaut crests a dune, silhouetted against the setting sun. Slow push-in. > 3-7s: They stop, turn to face the camera, and raise a hand to shield their eyes. Wind lifts red dust across the frame. > 7-10s: Wide pull-back reveals a half-buried spaceship behind them. Warm rim light, long shadows.

Give each range enough to do but not too much — cramming a dozen actions into three seconds causes cuts or dropped beats.

3. Consistency notes

End with anything that should hold across the whole clip: camera style, lighting, color palette, atmosphere, audio.

Handheld feel throughout, shallow depth of field, warm golden-hour palette. Ambient wind and footsteps only — no music.

Camera language

Write these directly; the model understands them:

• Shot size — extreme wide, wide, medium, medium close-up, close-up
• Movement — push in, pull out, pan, track, follow, orbit, tilt up, handheld shake
• Angle — low angle, overhead, eye-level, first-person
• Techniques — one-shot / long take, dolly zoom, bullet time, speed ramp

For a niche term, add a plain-language gloss: “rack focus: the sharp foreground softens as the background comes into focus.”

Referencing uploaded assets

When you attach images, videos, or audio, bind each one explicitly in the prompt by its upload order — @image1, @video1, @audio1 — and say what it’s for. Don’t rely on labels drawn inside the image itself.

• “The knight in @image1 walks through the castle in @image2.”
• “Refer to @video1 for the camera movement only; keep its shot order.”
• “@image1 and @image2 are Character 1, voiced by @audio1.”

When a reference is already accurate, just point to it — no need to re-describe it in detail.

Negative control

Positive descriptions work best, but you can suppress subtitles and audio:

• “No subtitles.”
• “No BGM — ambient and action sounds only.”
• “No audio.”

Weak vs. strong

Pricing

Without reference videos

Output-only pricing:

Prices scale linearly with output duration from 4 to 30 seconds.

With reference videos

Normalized reference-video duration and output duration use the same per-second rate:

Each reference video is normalized to at least 2 seconds, each clip and the combined input are capped at 30 seconds, and the normalized combined input duration is rounded up to a whole second for billing.

Example: 5 seconds of normalized reference input plus 5 seconds of output costs $1.10 at 480p, $2.20 at 720p, $5.50 at 1080p, or $11.00 at 4k.

Best Use Cases

• Cinematic video generation — Create polished video scenes from detailed text prompts.
• Commercials and ads — Generate high-quality promotional clips, product scenes, and campaign visuals.
• Social media content — Produce vertical, square, landscape, or ultra-wide short-form videos.
• Concept visualization — Turn creative ideas into motion previews for pitching, planning, and storyboarding.
• Reference-guided generation — Use images, videos, or audio to guide style, motion, characters, or soundtrack direction.
• Music and audio-driven scenes — Generate clips with synchronized sound for music, ambience, or narrative moments.

Pro Tips

• Write prompts like a film director, including subject, action, camera angle, lighting, atmosphere, and mood.
• Use 16:9 for cinematic widescreen and 9:16 for vertical social content.
• Include specific visual details such as lighting direction, color palette, environment, and camera movement.
• Use reference images when character, product, or style consistency matters.
• Use reference videos when motion, pacing, camera movement, or scene structure matters.
• Use reference audio when soundtrack, ambience, or voice style matters.
• Start with shorter durations for prompt testing, then increase duration once the direction is confirmed.
• Use asset tags such as @image1, @image2, and @audio1 to reference uploaded assets in the prompt.
• Reference video duration is rounded up, with a minimum of 2 seconds and a maximum of 30 seconds.

Related Models

• Seedance 2.5 Image-to-Video — Generate video from a start image and prompt.
• Seedance 2.5 Video Edit — Edit an existing video with a prompt.
• Seedance 2.5 Video Extend — Extend an existing video with a new generated segment.
• Seedance 2.0 Text-to-Video — Previous-generation Seedance text-to-video model.

Authentication

For authentication details, please refer to the Authentication Guide.

API Endpoints

Submit Task & Query Result

set -euo pipefail

export WAVESPEED_API_KEY="your-api-key"

REQUEST_BODY=$(cat <<'JSON'
{
  "prompt": "A cinematic ocean wave at sunrise, highly detailed",
  "aspect_ratio": "16:9",
  "resolution": "720p",
  "duration": 5,
  "generate_audio": true
}
JSON
)

# 1. Submit the prediction.
SUBMIT_RESPONSE=$(curl --silent --show-error --fail-with-body \
  -X POST "https://api.wavespeed.ai/api/v3/bytedance/seedance-2.5/text-to-video" \
  -H "Authorization: Bearer ${WAVESPEED_API_KEY}" \
  -H "Content-Type: application/json" \
  -d "${REQUEST_BODY}")

TASK=$(printf '%s' "${SUBMIT_RESPONSE}" | jq 'if type == "object" and has("data") then .data else . end')
PREDICTION_ID=$(printf '%s' "${TASK}" | jq -r '.id // empty')
if [ -z "${PREDICTION_ID}" ]; then
  printf 'Submission response did not contain a prediction id
' >&2
  exit 1
fi
RESULT_URL="https://api.wavespeed.ai/api/v3/predictions/${PREDICTION_ID}/result"

# 2. Poll until the prediction finishes.
while true; do
  RESPONSE=$(curl --silent --show-error --fail-with-body \
    "${RESULT_URL}" \
    -H "Authorization: Bearer ${WAVESPEED_API_KEY}")
  RESULT=$(printf '%s' "${RESPONSE}" | jq 'if type == "object" and has("data") then .data else . end')
  STATUS=$(printf '%s' "${RESULT}" | jq -r '.status // empty')

  case "${STATUS}" in
    completed) printf '%s\n' "${RESULT}" | jq '.outputs'; break ;;
    failed|cancelled|timeout|deleted) printf '%s\n' "${RESULT}" | jq . >&2; exit 1 ;;
    *) sleep 2 ;;
  esac
done

Parameters

Task Submission Parameters

Request Parameters

Response Parameters

Result Request Parameters

Result Response Parameters