Wan 2.7

Video generation and editing with multi-reference control and temporal transfer

About model

Wan 2.7 is Wan AI's video generation and editing model, introducing instruction-based and reference-based video editing alongside temporal feature transfer. It supports text-to-video, image-to-video, and reference-to-video workflows with up to 5 simultaneous reference inputs for multi-subject compositions. The model accepts joint image, video, and audio references for synchronized subject+voice control, supports real human inputs as references or first frames, and generates native 1080p video from 2 to 15 seconds.

Max References

Multi-subject with mixed image/video/audio inputs

Native Resolution

1080p

Across all generation and editing modes

Flexible Duration

2-15s

T2V and I2V with 2-10s for R2V

Model key capabilities

Video Editing: Instruction-based and reference-based editing to modify subjects or scenes globally via text prompts or reference media
Temporal Feature Transfer: Clone motion, camera moves, effects, and style directly from a reference video into new generations
Multi-Reference Control: Up to 5 simultaneous references with joint subject+voice referencing via combined image, video, and audio inputs
Flexible Generation: T2V, I2V, and R2V modes with first/last frame control, 3x3 grid-to-video, real human inputs, and native 1080p up to 15s

Quickstart guides

Video

Videos Overview

API usage

cURL
Python
Typescript

Endpoint:

Wan-AI/Wan2.7

curl --request POST \
  --url https://api.together.xyz/v2/videos \
  --header "Authorization: Bearer $TOGETHER_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "model": "Wan-AI/Wan2.7",
    "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="Wan-AI/Wan2.7"
)

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: "Wan-AI/Wan2.7"
  });

Model card
Architecture Overview:
• Unified video generation and editing model supporting text-to-video (T2V), image-to-video (I2V), and reference-to-video (R2V) workflows
• Instruction-based and reference-based video editing: modify subjects or scenes globally via text prompts or reference media
• Temporal feature transfer: clone motion, camera moves, effects, and style directly from a reference video
• I2V supports first/last frame control and 3x3 grid-to-video generation
• Multi-reference support with up to 5 references for multi-subject, image+voice, and mixed image/video modes
• Joint subject+voice referencing via combined image, video, and audio inputs
• Real human image and video inputs supported as references or first frames
• Native 1080p output across all generation modes

Training Methodology:
• Built on the Wan model family with expanded capabilities for video editing and temporal transfer
• Trained for consistent subject identity preservation across multi-reference and multi-subject scenes
• Optimized for real human inputs maintaining natural appearance and motion

Performance Characteristics:
• T2V and I2V support 2-15 second generation with flexible duration control
• R2V supports 2-10 second generation
• Up to 5 simultaneous reference inputs for complex multi-subject compositions
• Temporal feature transfer preserves motion dynamics, camera work, and visual effects from source video
Prompting
Together AI API Access:
• Access Wan 2.7 via Together AI APIs using the endpoint Wan-AI/Wan2.7
• Authenticate using your Together AI API key in request headers
• Supports text-to-video, image-to-video, and reference-to-video generation modes
• Reference inputs accept image, video, and audio for joint subject+voice control
• Available on Together AI serverless infrastructure
Applications & use cases
Video Editing & Post-Production:
• Instruction-based editing: modify subjects, scenes, and visual elements via text prompts
• Reference-based editing: transfer style, motion, and camera work from source videos
• Temporal feature transfer for replicating specific motion dynamics and effects

Marketing & Brand Content:
• Campaign video production with consistent character identity across multiple assets
• Product videos and social media content at native 1080p
• Brand mascot and spokesperson videos using real human reference inputs

Creative Production:
• Multi-subject scene composition with up to 5 reference inputs
• First/last frame control for precise narrative sequencing
• 3x3 grid-to-video generation for storyboard-to-video workflows
• Subject+voice referencing for synchronized character performances

Related models

Model specifications

Model data

Model provider
Alibaba
Type
Video
Resolution/Duration
1080p / 2-15s
Deployment
Serverless
Endpoint
Wan-AI/Wan2.7
Input modalities
Text
Image
Video
Audio
Output modalities
Video

Category
Video

Run in Playground

Quickstart docs

Deploy model

Wan 2.7

About model

API usage

Model card

Prompting

Applications & use cases