This is an easy example of wan2.2 comfyUI image to video that works with a 12GB VRAM card.
Wan 2.2 I2V (Image-to-Video) model, the high-noise model handles the initial, broader stages of video generation, focusing on foundational content and complex motion, while the low-noise model refines the details and sharpens the image in the later stages of the process. This "Mixture of Experts" (MoE) architecture divides the denoising steps into high-noise and low-noise processes, allowing for a larger model capacity to create higher-quality, more controlled, and more detailed videos.
High-Noise Model
Purpose: Used in the early steps of the video generation process.
Function: Handles the broader aspects of the video, such as creating complex motion and establishing the overall structure.
Key Benefit: Allows the model to handle complex scenes and generate various types of motion more effectively.
Low-Noise Model
Purpose: Used for the final steps of denoising, where details are refined.
Function: Focuses on improving image sharpness, clarity, and visual details.
Key Benefit: Enhances the final output by bringing greater detail to the generated video.
How They Work Together
1. Initial Generation:
The high-noise model creates the fundamental structure and motion of the video.
2. Refinement:
The low-noise model then takes over to add fine details, improving the overall quality and coherence of the final video.
This hybrid approach is essential for Wan 2.2's improved image sharpness, clarity, and visual details.