About Wan 2.7
Wan 2.7 is the latest evolution of Alibaba's open-source AI video generation model, designed to make cinematic video creation accessible to everyone.
What Is Wan 2.7?
Wan 2.7 is a state-of-the-art AI video generation model developed within Alibaba's Qwen ecosystem. Built on a 27-billion-parameter Mixture-of-Experts (MoE) architecture, Wan 2.7 generates cinematic 1080P HD videos from text descriptions and images. It represents a major leap forward in open-source AI video generation, combining visual fidelity, audio synchronization, and motion consistency in a single unified model.
Unlike closed-source alternatives, Wan 2.7 is fully open source, giving developers, researchers, and creators complete access to the model weights, architecture, and training methodology. This transparency has made it one of the most popular AI video models on GitHub, with over 15,000 stars and an active community of contributors.
The Evolution of Wan Video Models
The Wan model family has rapidly evolved through multiple generations, each bringing significant improvements in video quality, capabilities, and efficiency:
Wan 2.1 — The Foundation
The original open-source release established the core architecture with text-to-video and image-to-video capabilities at 480P and 720P resolutions using a 14B parameter model.
Wan 2.2 — Mixture-of-Experts
Introduced the MoE architecture (27B total, 14B active), trained on 65% more images and 83% more videos. Added LoRA support, VACE features, and a new VAE with 16x16x4 compression ratio.
Wan 2.5 — Speed & Scale
Optimized for fast content creation with improved generation speed and quality refinements, becoming a practical tool for daily content workflows.
Wan 2.6 — Multi-Shot Storytelling
Introduced connected multi-shot video generation, real-person image inputs, up to five video references, and 1080P output. Released December 2025.
Wan 2.7 — The Current Generation
Major upgrades across visual quality, audio, motion dynamics, stylization, and consistency. New features include 9-grid image-to-video, subject + voice reference, first-and-last-frame control, and instruction-based video editing. Launching March 2026.
Key Capabilities
Wan 2.7 delivers the most comprehensive feature set of any open-source AI video generator:
- Text-to-Video: Generate cinematic videos from natural language descriptions with control over camera, lighting, and mood
- Image-to-Video: Transform still images into dynamic videos with realistic motion and physics
- Multi-Shot Storytelling: Create connected scenes with consistent characters and environments
- Audio Synchronization: Native audio synthesis with automatic lip sync, foley effects, and music matching
- First & Last Frame Control: Specify start and end frames for precise creative direction
- 9-Grid Synthesis: Structured multi-angle input for enhanced composition control
- Instruction-Based Editing: Modify existing videos using natural language commands
- Subject & Voice Reference: Maintain character visual and vocal consistency across scenes
Open Source & Community
Wan 2.7 is built on the principle that powerful AI video generation should be accessible to all. The model weights are available on Hugging Face, the code is on GitHub under the Wan-Video organization, and an active community contributes extensions, fine-tuning workflows, and integration tools for platforms like ComfyUI.