ByteDance Seedance 2.5

One-take video storytelling with native audio and timestamp-level editing

About model

Seedance 2.5 is ByteDance's new-generation video creation model, built on the unified multimodal audio-video joint-generation architecture of Seedance 2.0 and centered on turning clips into complete creative works. It generates 30-second audio-video clips in a single pass, doubled from 15, organizing multiple logically connected shots into a story with setup, development, and resolution, and supports multi-round extension that keeps characters, environments, and pacing consistent across multi-minute videos. Reference capacity grows to 30 images, 10 video clips, and 10 audio clips per pass, joined by clay render, motion, and creative referencing that let textureless 3D models drive composition, blocking, camera moves, and physically consistent lighting. Timestamp-level control directs narrative, camera, and rhythm for specific time ranges and supports targeted edits after generation, alongside green screen, camera perspective, and reference-based editing. Available on Together AI.

‍

Single-Pass Generation

30s

Multi-round extensions build multi-minute stories with consistent characters and pacing

Max Reference Inputs

Up to 30 images, 10 video clips, and 10 audio clips in one pass

Editing Control

Timestamp

Direct narrative, camera, and pacing per time range, and edit specific moments after generation

Model key capabilities

One-Take Storytelling: 30-second single-pass clips structured as complete stories, extendable round by round into multi-minute videos with consistent subjects
Multimodal Referencing: Up to 30 images, 10 videos, and 10 audio clips per pass, plus clay render, motion, and creative references that control composition, blocking, and lighting
Timestamp-Level Editing: Per-time-range control during generation and targeted edits after, with green screen, camera perspective, and reference-based editing
Production-Ready Infrastructure: 99.9% SLA, available on Together AI serverless infrastructure

API usage

cURL
Python
Typescript

Endpoint:

ByteDance/Seedance-2.5

curl --request POST \
  --url https://api.together.xyz/v2/videos \
  --header "Authorization: Bearer $TOGETHER_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "model": "ByteDance/Seedance-2.5",
    "prompt": "some penguins building a snowman"
  }'

from together import Together

client = Together()

# Create a video generation job
job = client.videos.create(
    prompt="A serene sunset over the ocean with gentle waves",
    model="ByteDance/Seedance-2.5"
)

import Together from "together-ai";

const together = new Together();

async function main() {
  // Create a video generation job
  const job = await together.videos.create({
    prompt: "A serene sunset over the ocean with gentle waves",
    model: "ByteDance/Seedance-2.5"
  });

Model card
Architecture Overview:
• Unified multimodal audio-video joint-generation architecture, carried forward from Seedance 2.0
• Single-pass generation of up to 30 seconds with synchronized audio, extendable across multiple rounds
• Accepts text, image, video, and audio inputs, including textureless clay renders that define spatial structure, motion paths, and camera angles

Training Methodology:
• Built by the ByteDance Seed team as the successor to Seedance 2.0, centered on foundational generation and reference-based generation
• ByteDance reports systematic optimization of object textures, skin and eye detail, lighting, and color, along with fewer uncontrolled subtitles and background music in outputs

Performance Characteristics:
• ByteDance reports smoother shot transitions and scene changes, with subjects stable across cuts and audio staying in sync through long-form videos
• Clay render referencing produces lighting that follows physical laws, including source direction, color temperature, intensity, and shadow projection
• Green screen editing preserves the subject while adapting clothing movement, hair, gait, and lighting interaction to the new environment
‍
Prompting
Together AI API Access:
• Access Seedance 2.5 via Together AI APIs using the endpoint ByteDance/Seedance-2.5
• Authenticate using your Together AI API key in request headers
• Create a video request with a text prompt, optionally combining image, video, audio, and clay render references, then poll until the video completes
• Reference materials support up to 30 images, 10 video clips, and 10 audio clips per generation, addressable in prompts
• Direct specific time ranges in the prompt to control narrative, camera, and pacing, or extend an existing output with a follow-up request
• Available on Together AI serverless infrastructure
‍
Applications & use cases
Film & Advertising Production:
• Produce one-take sequences with multiple shots, then extend them into multi-minute pieces
• Block complex scenes with clay render references so composition and camera match the storyboard
• Replace backgrounds with green screen editing while keeping subjects physically consistent

Education & Training Content:
• Turn lessons, historical events, and experimental procedures into vivid instructional video
• Keep recurring characters and settings consistent across a series of clips
• Localize and customize teaching materials by swapping references rather than reshooting

Simulation & Synthetic Data:
• Generate synthetic video for training perception and manipulation systems
• Simulate long-tail driving scenarios such as extreme weather and complex traffic
• Produce industrial process demonstrations and equipment walkthroughs from reference material
‍

Model specifications

Model data

Model provider
ByteDance
Type
Video
Main use cases
Text-to-Video
Image-to-Video
Resolution/Duration
Up to 30s per generation
Deployment
Serverless
Endpoint
ByteDance/Seedance-2.5
Price
$0.115 / video
Input modalities
Text
Image
Video
Audio
Output modalities
Video

Released
July 30, 2026
Category
Video

Run in Playground

Quickstart docs

Deploy model

ByteDance Seedance 2.5

About model

API usage

Model card

Prompting

Applications & use cases