NEW: Wan 2.7-Image Released
QwenPowered by Alibaba Qwen

Create Stunning Videos & Images
with Wan AI

Wan 2.7 is here. The Image model generates lifelike faces, precise color palettes, and 3K-token text — #1 in blind tests. The Video model adds native audio, first/last frame control, 9-grid input, and instruction editing at 1080p. Wan 2.1 remains open-source with no content filters.

Start Creating
View Examples
15M+
Videos Generated
800K+
Active Users
4.9
User Rating
NEW RELEASE — April 2026

Wan 2.7-Image
Unified Image Generation & Editing

Wan 2.7-Image brings lifelike portrait generation, precise color control, and ultra-long text rendering. Human preference blind test #1 in China.

Thousand Unique Faces

Break free from AI 'standard faces'. Customize bone structure, eye shape (almond, deep-set, phoenix), face shape (oval, round, square), and fine facial details for truly unique, lifelike portraits.

Face ShapeEye FeaturesBone Structure

Color Palette Control

One-click extract color distributions from reference images. Reproduce master painting palettes or align with brand color guidelines while maintaining composition integrity.

Color ExtractionBrand AlignmentPrecise Migration

3K Token Text Rendering

Render up to 3,000 tokens of text across 12 languages with print-level clarity. Generate complex tables, mathematical formulas, and full A4 pages of academic content.

12 Languages3000 TokensPrint Quality

Interactive Click-to-Edit

Point and edit — simply select an area to add, align, or move elements with pixel-level precision. Native interactive editing module for intuitive image manipulation.

Click to EditPixel PrecisionAdd/Move/Align

Multi-Subject Consistency

Support up to 9 reference images for consistent character and style across storyboards, e-commerce photo sets, movie posters, and multi-angle architectural views.

9 Reference ImagesStyle UnityCharacter Consistency

Batch Generation (12 Images)

Generate up to 12 images at once for creating consistent series, PPT illustrations, storyboard sequences, e-commerce model photos, and multi-view renders.

Up to 12 ImagesSeries CreationMulti-View

Unified Generation + Understanding Architecture

Wan 2.7-Image features a leading unified generation-understanding architecture. Through shared latent space semantic mapping, it achieves a leap from pixel fitting to deep semantic cognition. Wan 2.7-Image-pro with larger-scale training is also available.

Try Wan 2.7-ImageDeveloper API

Powerful Features

Everything you need to create professional AI videos & images

Text to Video

Transform your text descriptions into stunning, high-quality videos with advanced AI understanding. Support for Chinese, English, Japanese, and more.

Image to Video

Bring your static images to life with natural motion and cinematic effects. Wan 2.7 supports 9-grid multi-image to video for richer scene composition.

Motion Control

Precise camera movement and object trajectory control for professional results. Pan, zoom, rotate with cinematic precision.

High Resolution

Generate videos up to 1080p/24fps and images with print-level clarity. Wan 2.7 VBench score 90%+, ranking #1 in human preference blind tests.

Multi-Language

Native support for prompts in English, Chinese, Japanese, Korean, and German with excellent semantic understanding.

Open Source

Wan 2.1 is fully open-source under Apache 2.0. Run locally on consumer GPUs (8GB+ VRAM) or use cloud API for newer versions.

AI Image Generation

Wan 2.7-Image generates and edits images with a unified model. Supports face customization, color palette control, 3K token text rendering, and batch generation of up to 12 images.

Uncensored / NSFW

Wan 2.1 runs locally on your GPU with zero content filters. Full control over generation — no restrictions, no censorship. Community fine-tuned versions available on GitHub and Hugging Face.

Advanced Technology

State-of-the-Art Architecture

Built on cutting-edge Diffusion Transformer technology with MoE (Mixture of Experts) and native multimodal capabilities, achieving top-tier performance in VBench benchmarks.

Diffusion Transformer (DiT)

Advanced transformer-based diffusion model enabling superior temporal coherence and complex motion understanding for realistic video generation.

TransformerDiffusionSOTA

Causal 3D VAE (Wan-VAE)

Efficient spatiotemporal compression with 4×8×8 ratio, supporting arbitrary length 1080p video encoding while preserving precise temporal information.

4×8×8 Compression1080p SupportTemporal Coherence

Mixture of Experts (MoE)

27B total parameters with 14B activation, reducing computation by ~50% while improving complex scene generation and multi-character interactions.

27B Parameters50% EfficiencyMulti-Expert

Native Multimodal

Unified architecture for text, image, video, and audio processing. Native lip-sync support with precise mouth movement matching to speech.

Lip-SyncAudio-VisualUnified Model

Model Specifications Comparison

MetricWan2.1Wan2.2Wan2.6Wan2.7
Max Resolution720P720P1080P1080P+
Max Duration5s5s15s2-15s
Frame Rate24fps24fps24fps24fps
Parameters14B27B27B+27B+
VBench Score86.22%87.5%89%+90%+
Image Generation--BasicUnified Gen+Edit

Evolution of Wan Series

Continuous innovation from Wan 2.1 to Wan 2.7. Both the Image model (unified generation & editing) and Video model (1080p, native audio, 15s) are now available.

2025.02

Wan 2.1

  • 14B parameter flagship model
  • VBench score 86.22% (Global #1)
  • Chinese/English text effects
  • Consumer GPU support (6GB+)
2025.07

Wan 2.2

  • MoE architecture (27B total)
  • 60+ cinematic parameters
  • Character replacement tech
  • 50% computation savings
2025.10

Wan 2.5

  • Native multimodal architecture
  • Audio-visual synchronization
  • 10-second generation
  • Photo singing & dancing
2025.12

Wan 2.6

  • 15-second video (China's longest)
  • Multi-shot narrative system
  • Role-playing & voice cloning
  • Full lip-sync support
Latest
2026.04

Wan 2.7

  • Wan 2.7-Image: unified gen+edit model
  • Video: 1080p 15s with native audio
  • First/last frame, 9-grid, instruction edit
  • Open-source expected mid-to-late Q2 2026
Online Ready

AI Video Playground

Start creating your AI video in seconds

Model:
5s
Wan 2.65sCinematic
1080p•16:9•24fps
Preview
NSFW Mode(18+ content)
Estimated Time~15s
Credits Cost
10 creditsLimited FREE

Quick prompts:

Unlimited Creative Possibilities

From personal creativity to professional production, Wan AI empowers creators across all industries.

Short Video Creation

Create engaging short-form content for TikTok, YouTube Shorts, Instagram Reels. Generate creative videos from simple text prompts.

Lifestyle vlogs
Food recipes
Travel highlights
Comedy sketches

Advertising & Marketing

Produce professional product demos, brand commercials, and marketing materials at a fraction of traditional costs.

Product showcases
Brand stories
Social media ads
E-commerce videos

Film & Animation

Generate concept videos, storyboard previews, and animated sequences for film pre-production and indie projects.

Concept visualization
Storyboard animation
VFX previews
Indie films

Education & Training

Create educational content with physics simulations, process demonstrations, and interactive learning materials.

Science simulations
Historical recreations
Language learning
Tutorial videos

Digital Human & Avatar

Generate realistic digital humans for news broadcasting, virtual assistants, and interactive entertainment.

Virtual anchors
AI assistants
Virtual influencers
Customer service bots

Gaming & Entertainment

Create game trailers, cutscenes, character animations, and promotional content for gaming industry.

Game trailers
Character reveals
Cutscene previews
Esports highlights

Wan vs Competition

Comprehensive comparison: Wan 2.7 vs SeedDance 2.0, Sora 2, Kling 3.0, Veo 3.1, and Runway Gen-4.5. Based on April 2026 data. Wan 2.7-Image ranks #1 in human preference.

MetricsWan 2.7
Recommended
SeedDance 2.0Sora 2Kling 3.0Veo 3.1Gen-4.5
Max Duration2-15s15s25s10s10s10s
Resolution1080p1080p1080p4K/60fps1080p1080p
Open Source
(wan 2.1 open source)
Real Person Input
Video Reference Clips51—1—1
Free to Use
Instruction Editing
Lip Sync★★★★★★★★★☆★★★★☆★★★★☆★★★★☆★★★☆☆
Style Consistency★★★★★★★★★☆★★★★☆★★★★☆★★★★☆★★★☆☆
CostFree$$$$$$$$$$$
Wan AI vs Seedance →Wan AI vs Sora →Wan AI vs Kling →Wan AI vs Runway →Wan AI vs Pika →Wan AI vs Veo →

Frequently Asked Questions

Everything you need to know about Wan AI video generator.

Wan AI is the most advanced AI video generation model series developed by Alibaba. Wan 2.1 is fully open-source (Apache 2.0) and can run locally on consumer GPUs. The series offers unique features like 15-second generation, multi-shot narrative, and native lip-sync support.
Wan 2.1 is completely free and open-source. You can download the model weights from GitHub or Hugging Face and run it locally. For newer versions like Wan 2.6, we offer a cloud API service.
For the lightweight 1.3B model, you need only 6-8GB VRAM (RTX 3060 or better). For the full 14B model, 24GB+ VRAM is recommended (RTX 4090, A100). The model supports INT8 quantization to reduce memory requirements.
Wan 2.7 includes two models — Image and Video — both available now. Wan 2.7-Image (April 1, 2026) is a unified image generation & editing model with: ① Lifelike face customization ('thousand faces'); ② Color palette control; ③ 3K token text rendering in 12 languages; ④ Interactive click-to-edit; ⑤ Multi-subject consistency (up to 9 refs); ⑥ Batch generation up to 12 images. Ranked #1 in human preference blind tests. The Wan 2.7 Video model (late March 2026) generates 1080p, 15-second videos with native audio built in. Five improvements: commercial-grade visual quality, motion coherence, native audio integration, style consistency, and temporal stability. New capabilities: first/last frame control, 9-grid image-to-video, subject+voice reference cloning, instruction-based editing, and video recreation. Open-source video model release expected mid-to-late Q2 2026. Credits on our platform never expire.
Wan 2.7 has several advantages over SeedDance 2.0: ① Real person image input support (SeedDance doesn't support); ② Up to 5 video reference clips (SeedDance only 1); ③ Flexible 2-15s dynamic duration; ④ 1080P video generation; ⑤ Wan 2.1 is open-source and can run locally without restrictions.
Wan 2.6 supports up to 15 seconds of video in a single generation. Wan 2.7 offers flexible 2-15 second dynamic duration. For longer videos, use the multi-shot narrative feature to create coherent sequences.
Yes! Wan 2.5 and later versions feature native multimodal architecture with full audio-visual synchronization. Characters can sing, speak, and their lip movements perfectly match the audio input.
While Sora excels at physics simulation and longer sequences (25s), Wan offers advantages in multi-shot narrative, lip-sync, Chinese language support. Wan 2.1 is free and open-source and runs on consumer GPUs, while Sora requires expensive cloud infrastructure.
Absolutely! Wan is released under the Apache 2.0 license, which allows commercial use without restrictions. You can use it for advertising, film production, content creation, and any other commercial purpose.
Wan has excellent support for English, Chinese (native-level), Japanese, Korean, and German. The model can understand complex prompts in these languages and generate accurate videos accordingly.
This is Wan's biggest advantage. Wan 2.1 is fully open-source and runs on your own hardware with zero content filters — no restrictions whatsoever. You have complete control over the model. While Sora, Kling, Runway, and Pika all enforce strict content policies, Wan 2.1 lets you generate anything locally. Community-tuned NSFW versions are widely available on GitHub and Hugging Face. This alone makes Wan the top choice for creators who need unrestricted generation.
Limited Time Offer

Ready to Create Amazing Videos?

Join thousands of creators using Wan AI to bring their ideas to life. Free to use, Wan 2.1 is open-source.

$1 FREE Credit

25% Cashback

50 Free Generations

Claim Your Bonus Now

No credit card required

10M+

Videos

500K+

Users

99.9%

Uptime

24/7

Support