How to Use WAN Video in ComfyUI for AI Video Generation

ComfyUI now includes native support for WAN video models, offering both 1.3B models for low VRAM cards and 14B models for higher-end systems. These AI fundamentals make video generation accessible to creators with different hardware setups. GGUF formats are also available for additional flexibility.

Getting Started with WAN Video Templates

To access WAN video features, navigate to the workflow menu and click "browse templates." Select the video category to find five new WAN video templates: text to video, image to video, control video, outpainting, and first frame last frame.

If you don't see these templates, update your ComfyUI version using the ComfyUI Manager. Click the "update ComfyUI" button to access the latest features.

Setting up your Models

When opening workflows, you'll be prompted to download the appropriate models. For cards with limited VRAM, choose the 1.3B model and skip the larger 32GB download. If you have 24GB or more VRAM, the 32GB file runs smoothly despite its size.

GGUF files are available in various sizes but require a custom loader and manual download to your ComfyUI models directory.

Text to Video Generation

Load your chosen model using the "load models" option. The special LoRA available works like Hyper or Turbo LoRAs but for WAN 2.1, reducing steps from 20 to as few as 2 steps.

Enter your text prompt in the designated field. The WAN video node handles width, height, and frame count settings. The 14B model supports both 480p and 720p resolutions.

For optimal results without artifacts, use these settings:
- Sampler: DDIM
- Scheduler: DDIM uniform
- LoRA strength: 0.5
- Steps: 6

Image to Video Conversion

This workflow requires a reference image along with your text prompt. WAN video works best with images that have solid colored backgrounds or clean background images.

Use a background removal node to clean your reference image. Adjust the strength parameter in the WAN video node to control how closely the output matches your input image:
- Higher strength (1.0): Closer to original image, less motion
- Lower strength (0.67): More motion, potential first-frame issues

Control Video Features

This advanced workflow accepts both an input image and video. The video gets preprocessed using Canny edge detection or other preprocessors. The Canny output serves as the control video, allowing precise motion control while maintaining facial features and expressions.

Video Out-painting Capabilities

The outpainting template extends video borders by adding padding around the original footage. Use custom nodes for automatic width calculation rather than manual calculations. This feature can simultaneously out-paint and extend video duration.

First and Last Frame Generation

This template creates videos using only two images - one for the beginning and one for the end. Different aspect ratios work fine as the system resizes to match your specified video dimensions.

Using GGUF Models

GGUF models require a custom GGUF loader but maintain similar rendering speeds to standard models. Use the same settings as regular workflows, adjusting parameters to minimize artifacts.

Advanced Options

For additional features like block swap for lower VRAM cards, explore Kiji's WAN video wrapper, which offers extended functionality beyond the native ComfyUI implementation.

WAN video integration in ComfyUI provides powerful AI video generation capabilities across different hardware configurations. The five template workflows cover most video creation needs, from simple text-to-video to advanced control methods.

With proper settings optimization, you can createhigh-quality videos while minimizing common artifacts like first-frame blocking.