Wan-Move: Motion-Controllable Video Generation

Wan-Move is a motion-controllable video generation framework built on Wan2.1. Uses latent trajectory guidance for precise point-level motion control (NeurIPS 2025).

W

Wan-Move

Wan2.1Motion ControlI2VNeurIPS 2025

Simple and scalable motion-control framework for video generation built on Wan2.1. Uses latent trajectory guidance for precise point-level motion control. Generates 5-second 480p videos with controllability rivaling commercial solutions. Accepted at NeurIPS 2025.

DeveloperAli-Vilab (Alibaba)
Release Date2025-12
ArchitectureWan2.1-based with latent trajectory guidance
CapabilitiesPoint-level motion control, trajectory-guided I2V generation
Output Resolution832x480p at 5 seconds
VenueNeurIPS 2025

Overview

Wan-Move is a motion-controllable video generation framework built on Wan2.1 that brings precise point-level motion control to video generative models. Unlike existing methods that suffer from coarse control granularity and limited scalability, Wan-Move achieves fine-grained motion control through latent trajectory guidance.

The framework represents motion by taking features from the first frame and propagating them along user-defined trajectories. Users can mark points on objects in the first frame and specify where those points should appear in subsequent frames, enabling intuitive motion control similar to commercial tools like Kling 1.5 Pro's Motion Brush.

Wan-Move removes the need for auxiliary motion encoders, making fine-tuning of base models easily scalable. Through scaled training, it generates 5-second, 480p videos with motion controllability that rivals commercial solutions as validated by user studies.

ComfyUI Integration

Wan-Move is available in ComfyUI via the Kijai WanVideoWrapper custom nodes. Both fp16 and fp8 scaled model variants are hosted on Hugging Face under Kijai/WanVideo_comfy.

Resources

ResourceLink
PaperarXiv:2512.08765
GitHubgithub.com/ali-vilab/Wan-Move
HuggingFace (Kijai)Kijai/WanVideo_comfy
Project Pagewanmove.net
NeurIPSneurips.cc/virtual/2025/poster/116293

Guides and workflows related to this model series.

No articles found.