Revolutionary AI Video Engine: One model for video generation, editing, and re-creation.
Wan2.1-VACE is more than just video generation; it's an all-in-one video creation partner. Its single model architecture gives you unprecedented control over video.
Create brand new video content from text descriptions or single images, transforming your imagination into dynamic visuals.
Perform in-depth editing on existing videos, including style transfer, object replacement, background extension, etc., giving old footage new life.
No need to switch between different tools. Wan2.1-VACE efficiently completes all video processing tasks from generation to editing with its unified architecture.
Wan2.1-VACE gives you fine-grained control over every frame of the video, freeing your creativity.
Action, posture, direction, all under your control.
Layout, motion trajectory, freely set.
Video style, overall look and feel, customize as you wish.
Supports multiple input methods, flexibly combined to meet your diverse creation needs.
The power of Wan2.1-VACE lies in the flexible combination of its functions, easily handling complex creation demands.
Combine "Image Reference + Background Extension + Duration Extension" to easily convert a vertical image into a horizontal long video with intelligently filled harmonious background.
Combine "Reference Image + Local Inpainting" to replace only specific objects in the video while perfectly preserving other elements, achieving seamless editing.
Find answers to common questions about the Wan2.1-VACE model here.
Wan2.1-VACE is an open-source multimodal video generation and editing foundational model developed by Alibaba Wan-AI Lab. It employs a unified architecture supporting various complex tasks like Text-to-Video (T2V), Image-to-Video (I2V), Video-to-Video (V2V) editing, Reference-guided generation (R2V), and Masked Video Editing (MV2V).
"All in One, Wan for All" is the core design philosophy of Wan2.1-VACE. "All in One" refers to its single model architecture capable of handling multiple video creation and editing tasks without needing to switch tools. "Wan for All" emphasizes its inclusivity, enabling a broader range of users to access and use advanced AI video technology through open source and support for consumer-grade hardware.
Main features include:
There are two main versions: Wan2.1-VACE-1.3B and Wan2.1-VACE-14B.
Wan2.1-VACE-1.3B: A lightweight version with about 1.3 billion parameters. Primarily supports 480p resolution video and is friendly to consumer-grade GPUs (e.g., T2V inference requires about 8.19GB VRAM). Suitable for individual creators and rapid prototyping.
Wan2.1-VACE-14B: A larger parameter scale version with about 14 billion parameters. Supports 480p and higher quality 720p resolution video. Offers stronger performance but has higher hardware requirements (e.g., I2V inference requires about 35GB VRAM). Suitable for professional video production and high-quality content generation.
Yes, Wan2.1-VACE is licensed under the Apache 2.0 open source license.
You can obtain the model and code from the following main channels:
Basic requirements include:
Detailed setup steps typically involve cloning the repository, installing dependencies, and downloading model weights.
Application prospects are broad, including: