Skip to main content
Alibaba Cloud Model Studio home pageAlibaba Cloud Model Studio

Video generation

Alibaba Cloud Model Studio provides video generation models for general-purpose creation (text-to-video, image-to-video, reference-to-video, video editing) and vertical scenarios (digital human lip-syncing, image-to-action, video character swapping, emoji creation).

Videos generated on the Online Experience page of the Model Studio console always carry a watermark: the text "AI generated" in the lower-right corner of the video. Because of legal compliance requirements, this watermark cannot be turned off from the console UI. To obtain videos without a watermark, call the Wan video generation models through the DashScope API, where the watermark parameter defaults to false (no watermark added). For more information, see Wan 2.7 text-to-video API reference and Wan 2.7 image-to-video API reference.

Model overview

Service deployment scope
Compare scopes
Global
Compute resources for model inference are scheduled globally.
International
Compute resources for model inference are scheduled globally, excluding Chinese Mainland.
US
Compute resources for model inference are restricted to the US.
Chinese Mainland
Compute resources for model inference are restricted to Chinese Mainland.
RegionVirginiaSingaporeVirginiaBeijing
Supported modelsWan - text-to-videoWan - image-to-video - first frameWan - reference-to-videoWan - text-to-videoWan - image-to-videoWan - image-to-video - first frameWan - image-to-video - first and last framesWan - reference-to-videoWan - general video editingWan - image-to-actionWan - video character swapWan - text-to-videoWan - image-to-video - first frameWan - text-to-videoWan - image-to-videoWan - image-to-video - first frameWan - image-to-video - first and last framesWan - reference-to-videoWan - general video editingWan - digital humanWan - image-to-actionWan - video character swapAnimateAnyoneEMOLivePortraitEmojiVideoRetalkVideo style transform

Model selection

  • General video generation
  • Digital human lip-syncing: Animates static photos to speak, sing, or narrate --- background stays fixed while the face, head, and body move.
    • For the most natural results, including facial expressions, head, and body movements, use Wan - digital human. This model replaces EMO.
    • For videos longer than 20 seconds with simple head movements, such as news reports, use LivePortrait.
  • Video motion transfer : This feature keeps the background of the photo static and animates the person using motion from a reference video. Use Wan - image-to-action.
  • Video character swapping : This feature replaces the person in a video with a person from an image while keeping the original background. Use Wan - video character swap.
  • Dance replacement: Replaces the dancer in a video with a person from an image. For best quality, use Wan - image-to-action and Wan - video character swapping. If budget is limited, use AnimateAnyone.
  • Video lip replacement: This feature replaces the lip movements in an existing video to match new audio. Use VideoRetalk.
  • Emoji creation: This feature creates emojis using fixed-style templates. Use Emoji.
  • Video redrawing: To use fixed-style templates, use Video style transform. To describe styles freely using prompts, use Wan - video editing.
  • Video editing: For all the following tasks, use Wan - general video editing.
    • Local video editing: Replace elements such as subjects or clothing, or remove bystanders.
    • Video extension: Extend short videos, for example, from 1 second to 5 seconds.
    • Video frame expansion: Convert landscape videos to portrait mode or fill in missing borders.
    • Multi-image reference generation: Fuse background and subject images to create a video.

Supported models

Wan - text-to-video

Generates videos from text prompts. It supports text and audio input to create cinematic, multi-shot videos. API reference | Model pricing | Try online: Singapore, Virginia, Beijing
  • Global
  • International
  • US
  • Chinese mainland
If you select the Global deployment scope, model inference compute resources are dynamically scheduled worldwide. Static data is stored in your selected region. Supported regions: US (Virginia), Germany (Frankfurt),China (Hong Kong).

Model

Features

Input modality

Output video specifications

wan2.6-t2v Recommended

Video with audio

Multi-shot narrative, audio-video synchronization

Text, audio

Resolution options: 720P, 1080P

Video duration: 5s, 10s, 15s

Defined specifications: 30 fps, MP4 (H.264 encoding)

Input promptOutput video (wan2.6, multi-shot video)
Shot from a low angle, in a medium close-up, with warm tones, mixed lighting (the practical light from the desk lamp blends with the overcast light from the window), side lighting, and a central composition. In a classic detective office, wooden bookshelves are filled with old case files and ashtrays. A green desk lamp illuminates a case file spread out in the center of the desk. A fox, wearing a dark brown trench coat and a light gray fedora, sits in a leather chair, its fur crimson, its tail resting lightly on the edge, its fingers slowly turning yellowed pages. Outside, a steady drizzle falls beneath a blue sky, streaking the glass with meandering streaks. It slowly raises its head, its ears twitching slightly, its amber eyes gazing directly at the camera, its mouth clearly moving as it speaks in a smooth, cynical voice: 'The case was cold, colder than a fish in winter. But every chicken has its secrets, and I, for one, intended to find them '.

Wan - image-to-video

The Wan image-to-video model is upgraded with multimodal input (text/image/audio/video) and supports three tasks: first-frame-to-video, first-and-last-frame-to-video, and video continuation. API reference | Model pricing | Prompt guide
  • International
  • Chinese Mainland
If you select the International deployment scope, model inference compute resources are dynamically scheduled worldwide, excluding the Chinese mainland. Static data is stored in your selected region. Supported region: Singapore.

Model

Features

Input modality

Output video specifications

wan2.7-i2v-2026-04-25 Recommended

Video with audio

First-frame-to-video, first-and-last-frame-to-video, video continuation, video continuation with last frame control

Multi-shot narrative, audio-video synchronization

Text, image, audio, video

Resolution options: 720P, 1080P

Video duration: [2s, 15s] (integer)

Defined specifications: 30 fps, MP4 (H.264 encoding)

wan2.7-i2v

Video with audio

First-frame-to-video, first-and-last-frame-to-video, video continuation, video continuation with last frame control

Multi-shot narrative, audio-video synchronization

Text, image, audio, video

Resolution options: 720P, 1080P

Video duration: [2s, 15s] (integer)

Defined specifications: 30 fps, MP4 (H.264 encoding)

Wan - image-to-video - first frame

Generates a video from a specified first-frame image. This model accepts text, a first-frame image, and audio as input to generate cinematic, multi-shot videos. API reference | Model pricing | Try online: Singapore, Virginia, Beijing
  • Global
  • International
  • US
  • Chinese Mainland
If you select the Global deployment scope, model inference compute resources are dynamically scheduled worldwide. Static data is stored in your selected region. Supported regions: US (Virginia), Germany (Frankfurt),China (Hong Kong).

Model

Features

Input modality

Output video specifications

wan2.6-i2v Recommended

Video with audio

Multi-shot narrative, audio-video synchronization

Text, image, audio

Resolution options: 720P, 1080P

Video duration: 5s, 10s, 15s

Defined specifications: 30 fps, MP4 (H.264 encoding)

Input promptInput first frame image and audioOutput video (wan2.6, multi-shot video)
An urban fantasy art scene. A dynamic graffiti art character. A teenager made of spray paint comes to life from a concrete wall. He performs an English rap at high speed while striking a classic, energetic rapper pose. The scene is set under an urban railway bridge at night. The lighting comes from a single streetlight, creating a cinematic atmosphere with high energy and amazing detail. The audio of the video consists entirely of his rap, with no other dialogue or noise.rap-转换自-pngInput audio:

Wan - image-to-video - first and last frames

Generates a video that smoothly transitions between specified first and last frame images. This model accepts text, first and last frame images, and audio as input to generate cinematic, multi-shot videos. API reference | Model pricing | Try online
  • International
  • Chinese Mainland
If you select the International deployment scope, model inference compute resources are dynamically scheduled worldwide, excluding the Chinese mainland. Static data is stored in your selected region. Supported region: Singapore.
ModelFeaturesInput modalityOutput video specifications
wan2.2-kf2v-flash RecommendedSilent video
Improved stability and success rate compared to the 2.1 model.
Text, imageResolution options: 480P, 720P, 1080PVideo duration: 5sDefined specifications: 30 fps, MP4 (H.264 encoding)
wan2.1-kf2v-plusSilent videoText, imageResolution options: 720PVideo duration: 5sDefined specifications: 30 fps, MP4 (H.264 encoding)
Input first frame imageInput last frame imageInput promptOutput video
first_frame
last_frame
Realistic style. A small black cat looks up at the sky curiously. The camera starts at eye level, gradually rises, and ends with a top-down shot of the cat's curious gaze.

Wan - reference-to-video

Make characters from a specified video perform actions. Input a video and a text prompt to generate an output video that maintains character consistency. API reference | Model pricing
  • Global
  • International
  • Chinese mainland
If you select the Global deployment scope, model inference compute resources are dynamically scheduled worldwide. Static data is stored in your selected region. Supported regions: US (Virginia), Germany (Frankfurt),China (Hong Kong).

Model

Features

Input modality

Output video specifications

wan2.6-r2v Recommended

Video with audio

Single-role/multi-role video generation

Multi-shot narrative, audio-video synchronization

Text, video

Resolution options: 720P, 1080P

Video duration: 5s, 10s

Defined specifications: 30 fps, MP4 (H.264 encoding)

Input reference video 1 (role: little girl)Input reference video 2 (role: alarm clock)Input promptOutput video (multi-role dialogue)
character1 says to character2: “I’ll rely on you tomorrow morning!” character2 replies: “You can count on me!”

Wan - video editing

Video editing model. Accepts text, image, and video multimodal input to perform various video generation and editing tasks. Video editing 2.7 API reference | Video editing 2.1 API reference | Model pricing
  • International
  • Chinese Mainland
If you select the International deployment scope, model inference compute resources are dynamically scheduled worldwide, excluding the Chinese mainland. Static data is stored in your selected region. Supported region: Singapore.

Model

Features

Input modality

Output video specifications

wan2.7-videoedit Recommended

Video with audio, silent video (depends on the input video)

Instruction-based editing, video migration

Text, image, video

Resolution options: 720P, 1080P

Video duration: [2s, 10s] (integer)

Defined specifications: 30 fps, MP4 (H.264 encoding)

wan2.1-vace-plus

Silent video

Multi-image reference, video redrawing, local editing, video extension, video frame extension

Text, image, video

Resolution options: 720P

Video duration: Up to 5s

Defined specifications: 30 fps, MP4 (H.264 encoding)

  • Feature 1: Multi-image reference
    Input reference image 1 (reference entity)Input reference image 2 (reference background)Input promptOutput video
    image
    image
    		Video shows a girl gracefully walking out from the depths of an ancient, misty forest. Her steps are light, and the camera captures her every nimble moment. When she stops and looks around at the lush woods, a smile of surprise and joy blossoms on her face. This moment, frozen in the interplay of light and shadow, records her wonderful encounter with nature.
  • Feature 2: Video redrawing
    Input videoInput promptOutput video
    The video shows a black steampunk-style car, driven by a gentleman, adorned with gears and copper pipes. The background is a steam-powered candy factory with retro elements, creating a vintage and playful scene.
  • Feature 3: Local video editing
    Input videoInput mask image (the white area indicates the editing area)Input promptOutput video
    mask
    		The video shows a Parisian-style French cafe where a lion in a suit elegantly sips coffee. It holds a coffee cup in one hand, taking a gentle sip with a contented expression. The cafe is tastefully decorated, with soft hues and warm lighting illuminating the area where the lion is.
  • Feature 4: Video extension
    Input first video segment (1s)Input promptOutput video (extended video is 5s)
    A dog wearing sunglasses skateboards on a street, 3D cartoon.
  • Feature 5: Video frame extension
    Input videoInput promptOutput video
    An elegant woman passionately plays the violin, with a full symphony orchestra behind her.

Wan - digital human

Only the Chinese mainland service deployment scope is supported. Data storage is in the Beijing access region. Model inference compute resources are limited to the Chinese mainland.
Digital human lip-syncing animates a person or cartoon character in an image to speak, sing, narrate, or perform. You provide an image and an audio file, and the model automatically generates a video with synchronized lip movements, facial expressions, and head and body motions. Image detection API reference | Video generation API reference | Model pricing

Model

Features

Input modality

Output description

wan2.2-s2v-detect

Image detection

Image

Output detection status: Pass or Fail

wan2.2-s2v

Video generation

Video with audio

Image, audio

Resolution options: 480P, 720P

Video duration: Up to 20s (follows audio duration)

Defined specifications:

  • 480P: 16 fps, MP4 (H.264 encoding)

  • 720P: 30 fps, MP4 (H.264 encoding)

Input example (character image + audio)Output video (lip-sync)
mix_input_image
Input audio:

Wan - image to action

Animates a person from an image using motion from a reference video. You provide an image and a video, and the model generates a video that applies the motion from the reference video to the person, while keeping the background of the original image static. API reference | Model pricing
  • International
  • Chinese mainland
If you select the International deployment scope, model inference compute resources are dynamically scheduled worldwide, excluding the Chinese mainland. Static data is stored in your selected region. Supported region: Singapore.

Model

Features

Input modality

Output video specifications

wan2.2-animate-move

Video with audio, silent video (depends on the input video)

  • Standard mode wan-std: Fast generation, cost-effective.

  • Professional mode wan-pro: More realistic results.

Image, video

Resolution options: 720P

Video duration: 2s < duration < 30s

Defined specifications:

  • Standard mode wan-std: 15 fps, MP4 (H.264 encoding)

  • Professional mode wan-pro: 25 fps, MP4 (H.264 encoding)

Input character imageInput reference videoOutput video (standard modewan-std)Output video (professional modewan-pro)
move_input_image

Wan - video character swap

Replaces a character in a video with one from a reference image. You provide a source video and a reference image, and the model generates an output video that retains the original background. This feature is ideal for use cases like face swapping and full character replacement. API reference | Model pricing
  • International
  • Chinese mainland
If you select the International deployment scope, model inference compute resources are dynamically scheduled worldwide, excluding the Chinese mainland. Static data is stored in your selected region. Supported region: Singapore.

Model

Features

Input modality

Output video specifications

wan2.2-animate-mix

Video with audio, silent video (depends on the input video)

  • Standard mode wan-std: Fast generation, cost-effective.

  • Professional mode wan-pro: More realistic results.

Image, video

Resolution options: 720P

Video duration: 2s < duration < 30s

Defined specifications:

  • Standard mode wan-std: 15 fps, MP4 (H.264 encoding)

  • Professional mode wan-pro: 25 fps, MP4 (H.264 encoding)

Input videoInput character image for replacementOutput video (standard modewan-std)Output video (professional modewan-pro)
mix_input_image

AnimateAnyone

  • Only the Chinese mainland service deployment scope is supported. Data storage is in the Beijing access region. Model inference compute resources are limited to the Chinese mainland.
  • We recommend using Wan - image-to-action and Wan - video character swapping instead of AnimateAnyone. These models offer better quality, while AnimateAnyone is a more cost-effective option.
Designed specifically for dancing, this model replaces the dancer in a video with a person from an image. You provide an image and a video to generate an output video in one of two ways: 1. Retain the image background. 2. Retain the video background. Image detection API reference | Action template generation API reference | Video generation API reference | Model pricing
ModelFeaturesInput modalityOutput description
animate-anyone-detect-gen2Image detectionImageOutput detection status: Pass or Fail
animate-anyone-template-gen2Dance video template generation
Extracts an action template from a dance video.
VideoOutputs a dance action template ID.
animate-anyone-gen2Video generationSilent videoImage, video, dance action template IDVideo resolution options: 720PVideo duration: 2s ≤ duration ≤ 60sDefined specifications: 15 fps, MP4 (H.264 encoding)
Input character imageInput dance videoOutput video (generated with image background)Output video (generated with video background)
05-9_16

EMO

  • Only the Chinese mainland service deployment scope is supported. Data storage is in the Beijing access region. Model inference compute resources are limited to the Chinese mainland.
  • Consider using Wan - digital human as an alternative to EMO. Wan - digital human provides better results, while EMO is a more cost-effective option.
Generates singing and performance videos from an image. You provide an image and an audio file, and the model automatically generates a video with synchronized lip movements, facial expressions, and head motions. Image detection API reference | Video generation API reference | Model pricing

Model

Features

Input modality

Output description

emo-detect-v1

Image detection

Image

Output detection status: Pass or Fail

emo-v1

Video generation

Video with audio

Image, audio

Video resolution:

  • 1:1 aspect ratio: Fixed at 512 × 512

  • 3:4 aspect ratio: Fixed at 512 × 704

Video duration: Up to 60s

Defined specifications: 15 fps, MP4 (H.264 encoding)

Input example (portrait image + audio)Output video (lip-sync singing)
15_原图Input audio:

LivePortrait

  • Only the Chinese mainland service deployment scope is supported. Data storage is in the Beijing access region. Model inference compute resources are limited to the Chinese mainland.
  • Consider using Wan - digital human as an alternative to LivePortrait. Wan - digital human delivers higher quality results, while LivePortrait is a more cost-effective option. Note that LivePortrait is suitable for generating long videos (over 20 seconds).
Generates narration videos from an image by animating the person in the image to deliver news or tell stories. You provide an Image and an Audio file, and the model automatically generates a video with synchronized lip movements, facial expressions, and slight head motions. Image detection API reference | Video generation API reference | Model pricing

Model

Features

Input modality

Output description

liveportrait-detect

Image detection

Image

Output detection status: Pass or Fail

liveportrait

Video generation

Video with audio

Image, audio

Video resolution: Follows the input image, up to nearly 4K (4096 × 4096).

Video duration: 1s < duration < 180s

Video frame rate: 15 fps ≤ frame rate ≤ 30 fps

Video format: MP4 (H.264 encoding)

Input example (portrait image + audio)Output video (lip-sync voiceover)
Emoji男孩Input audio:

Emoji

Only the Chinese mainland service deployment scope is supported. Data storage is in the Beijing access region. Model inference compute resources are limited to the Chinese mainland.
Creates emojis using fixed emoji templates. You provide an image and an emoji template ID to generate an emoji video. Image detection API reference |  Video generation API reference |  Model pricing

Model

Features

Input modality

Output description

emoji-detect-v1

Image detection

Image

Output detection status: Pass or Fail

emoji-v1

Video generation

Silent video

Image, emoji template ID

Video resolution: Fixed at 512 × 512

Video duration: Up to 5s (follows template duration)

Defined specifications: 15 fps, MP4 (H.264 encoding)

Input portrait imageOutput video ("disgusted" emoji)
image.png

VideoRetalk

Only the Chinese mainland service deployment scope is supported. Data storage is in the Beijing access region. Model inference compute resources are limited to the Chinese mainland.
Lip sync: Replaces the lip movements in a video to match a new audio track. You provide a video and an audio file, and the model generates an output video with synchronized lip movements. API reference |  Model pricing

Model

Features

Input modality

Output video specifications

videoretalk

Video with audio

Video, audio

Video resolution: Follows the input video, up to nearly 2K (2048 × 2048).

Video duration: 2s < duration < 120s

Video frame rate: 15 fps ≤ frame rate ≤ 60 fps

Video format: MP4 (H.264 encoding)

Input example (character broadcast video + audio)Output video (lip-sync replacement)
Input audio:

Video style transform

Only the Chinese mainland service deployment scope is supported. Data storage is in the Beijing access region. Model inference compute resources are limited to the Chinese mainland.
Applies a new artistic style to a video based on a predefined style template. You provide a video and a style transfer ID to generate a restyled video. API reference |  Model pricing
ModelFeaturesInput modalityOutput video specifications
video-style-transformVideo with audio, silent video
Depends on the input video.
Video redraw style IDVideo resolution: Follows the input video, up to nearly 4K (4096 × 4096).Video duration: Up to 30sVideo frame rate: 15 fps ≤ frame rate ≤ 25 fpsVideo format: MP4 (H.264 encoding)
Input videoOutput video (style transfer: "Japanese manga")
Alibaba Cloud Model Studio home pageAlibaba Cloud Model Studio
Get Started
Billing
Token Plan
Model Playground
  • Music generation
Clients and Developer Tools
Model Inference
Fine-tuning
Deployment
Evaluation
Statistics and Monitoring
Security and Compliance
Best Practices
Support
Changelog