ComfyUI Wan Video Guide: Text-to-Video & Image-to-Video Generation

What is Wan?

Wan is an open-source video generation model family from Alibaba, licensed under Apache 2.0 (commercial use allowed). It covers text-to-video (T2V) and image-to-video (I2V) generation with two main releases:

Hardware Requirements

| Wan 2.1/2.2 14B FP8 | 12–16 GB | Good balance | | Wan 2.1/2.2 14B FP16 | 16–24 GB | Best quality | | Wan 2.1 GGUF Q4 | 8 GB+ | Quantized for lower VRAM |

Wan 2.2: Recommended Starting Point

Wan 2.2 introduces a MoE (Mixture of Experts) architecture with separate high-noise and low-noise expert models for better quality. The 5B hybrid model is ideal for beginners — it handles both T2V and I2V in a single model.

Wan 2.2 5B Setup (Easiest)

Models (place in corresponding folders):

Workflow: Update ComfyUI to the latest version, then go to Workflows → Browse Templates → Video and select "Wan2.2 5B video generation".

Steps:

1. Load the diffusion model, text encoder, and VAE in the corresponding nodes
2. Write a video description in the CLIP Text Encoder node
3. (Optional) Load an image for I2V mode — enable the Load Image node with Ctrl+B
4. Adjust frame count via the length parameter
5. Click Run (Ctrl+Enter)

Wan 2.2 14B T2V Setup

For higher quality text-to-video, the 14B version uses two diffusion models (high-noise and low-noise experts):

Wan 2.2 14B I2V Setup

For image-to-video, download the I2V-specific diffusion models:

Wan 2.2 First-Last Frame Video

A unique mode that generates a video transitioning from a start frame to an end frame. Uses the same I2V models — load two images as first and last frames, and ComfyUI interpolates the motion between them.

Wan 2.1 Setup (Alternative)

Wan 2.1 remains a solid option with broader community tooling (Kijai wrapper, GGUF versions).

ComfyUI Native T2V

For I2V, also download:

• I2V diffusion model: 480p or 720p
• CLIP Vision: clip_vision_h.safetensors (place in models/clip_vision/)

GGUF Version (Low VRAM)

Requires the ComfyUI-GGUF plugin.

Saving Videos as MP4

ComfyUI's default output is .webp. To save as MP4, install the ComfyUI-VideoHelperSuite plugin and use the Video Combine node. All generated videos are saved to ComfyUI/output/.

Common Issues and Fixes

Out of memory during video generation

• Use the Wan 2.2 5B model (works on 8 GB VRAM)
• Use FP8 or GGUF quantized models
• Reduce resolution (480p instead of 720p)
• Reduce frame count

Video has visual drift or inconsistent motion

• Wan 2.2's MoE architecture significantly reduces drift compared to 2.1
• Write more specific motion descriptions in your prompt
• Try first-last frame mode for controlled transitions

T2V model specified but I2V model needed (or vice versa)

• T2V and I2V use separate diffusion models — make sure you download the correct one
• I2V workflows also require a CLIP Vision model that T2V does not

Models don't appear in node dropdown

• Verify files are in the correct folder (diffusion_models/, not checkpoints/)
• Restart ComfyUI after adding new model files

Related Guides

• HunyuanVideo Guide — Tencent's video generation model
• FramePack Guide — Low-VRAM video generation with FramePack
• Text to Image — Basic ComfyUI image generation