Skip to content
Become a Patron Help Build a Better ComfyUI Knowledge Base
Tutorial SeriesComfyUI Advanced TutorialWan2.1 ComfyUI Workflow - Complete Guide

Wan2.1 ComfyUI Workflow

Wan2.1, open-sourced by Alibaba in February 2025, is a benchmark model in the video generation field. Released under the Apache 2.0 license, it comes in 14B (14 billion parameters) and 1.3B (1.3 billion parameters) versions, covering text-to-video (T2V), image-to-video (I2V), and other tasks. This model not only outperforms existing open-source models in performance but, more importantly, its lightweight version can run with just 8GB of VRAM, significantly lowering the barrier to entry.

Currently, all models are available for download on Hugging Face and ModelScope platforms:

This article will guide you through the following Wan2.1-related workflows:

  • Installing Wan2.1 related models
  • Wan2.1 image-to-video workflow
  • Wan2.1 text-to-video workflow setup

Additionally, community authors have created GGUF and quantized versions:

Tongyi Wan2.1 ComfyUI Official Workflow Examples

The following workflows come from the ComfyUI official blog, as ComfyUI now natively supports Wan2.1

1. Wan2.1 Related Installation

  1. Update ComfyUI to the latest version

Please make sure to upgrade your ComfyUI version to the latest. Refer to the How to Update ComfyUI guide to complete the upgrade.

  1. Diffusion models installation

Diffusion models

Please download a version of the model from Comfy’s repackaged collection based on your workflow needs:

  • i2v is for image to video models, t2v is for text to video models
  • 14B and 1.3B represent the parameter count; higher values require more hardware performance
  • bf16, fp16, fp8 represent different precision levels; higher precision requires better hardware performance, with fp8 being the lowest precision and least demanding, but with relatively poorer results
  • Generally, larger file sizes also indicate higher hardware requirements

Save the downloaded models to the ComfyUI/models/diffusion_models directory

  1. Text encoders installation

Click to download umt5_xxl_fp8_e4m3fn_scaled.safetensors

If the download doesn’t respond, visit this link to download

Save the downloaded model to the ComfyUI/models/text_encoders directory

  1. CLIP vision model installation

Click to download clip_vision_h.safetensors

If the download doesn’t respond, visit this link to download

Save the downloaded model to the ComfyUI/models/clip_vision directory

  1. Wan2.1 VAE model installation

Click to download wan_2.1_vae.safetensors

If the download doesn’t respond, visit this link to download

Save the downloaded model to the ComfyUI/models/vae directory

2. Wan2.1 Image-to-Video Workflow

The following two workflows have similar nodes, with differences in the Diffusion Model used and the corresponding video dimensions. You can use either of the workflow files provided below, or use one of them and modify the parameters according to my tutorial instructions.

2.1 Wan2.1 Image-to-Video 14B 720P Workflow

Workflow File Download Click the button below to download the corresponding workflow, then drag it into the ComfyUI interface or use the menu bar Workflows -> Open(Ctrl+O) to load it

Input Image Example Please download this image to use as input Wan2.1 Image-to-Video 14B 720P Workflow Input Image Example

Wan2.1 Image-to-Video 720P Workflow Explanation Wan2.1 Image-to-Video 14B 720P Workflow

Please refer to the numbered items in the image to ensure that the corresponding nodes and models are loaded correctly for the model to run properly:

  1. Please upload the input image example provided earlier
  2. Ensure that the [clip_vision_h.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/clip_vision/clip_vision_h.safetensors?download=true) model can be read and loaded normally
  3. Ensure that the [umt5_xxl_fp8_e4m3fn_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors?download=true) model can be read and loaded normally
  4. Ensure that you’re using the 720P model and it can be read normally. If not downloaded, click here to download one of the following, depending on your device performance:
    • wan2.1_i2v_720p_14B_bf16.safetensors
    • wan2.1_i2v_720p_14B_fp8_e4m3fn.safetensors
  5. Ensure that the [wan_2.1_vae.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors?download=true) model can be read and loaded normally
  6. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter to execute video generation
  7. Note that the dimensions in the wanImageToVideo node are 768x768

2.2 Wan2.1 Image-to-Video 14B 480P Workflow

The 480P workflow is similar to the 720P workflow, with the main difference being that the LoadDiffusion Model node loads a 480P model, and the wanImageToVideo node dimensions are 512x512

Workflow File Download Click the button below to download the corresponding workflow, then drag it into the ComfyUI interface or use the menu bar Workflows -> Open(Ctrl+O) to load it

Input Image Example Please download the image below to use as input flux_dev_example

Wan2.1 Image-to-Video 480P Workflow Explanation Wan2.1 Image-to-Video 14B 480P Workflow

Wan2.1 Image-to-Video 480P Workflow Explanation Wan2.1 Image-to-Video 14B 480P Workflow

Please refer to the numbered items in the image to ensure that the corresponding nodes and models are loaded correctly for the model to run properly:

  1. Please upload the input image example provided earlier
  2. Ensure that the [clip_vision_h.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/clip_vision/clip_vision_h.safetensors?download=true) model can be read and loaded normally
  3. Ensure that the [umt5_xxl_fp8_e4m3fn_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors?download=true) model can be read and loaded normally
  4. Ensure that you’re using the 480P model and it can be read normally. If not downloaded, click here to download one of the following, depending on your device performance:
    • wan2.1_i2v_480p_14B_bf16.safetensors
    • wan2.1_i2v_480p_14B_fp8_e4m3fn.safetensors
  5. Ensure that the [wan_2.1_vae.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors?download=true) model can be read and loaded normally
  6. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter to execute video generation
  7. Note that the dimensions in the wanImageToVideo node are 512x512

3. Wan2.1 Text-to-Video 1.3B 1080P Workflow

Click the button below to download the corresponding workflow, then drag it into the ComfyUI interface or use the menu bar Workflows -> Open(Ctrl+O) to load it

Wan2.1 Text-to-Video 1.3B 1080P Workflow

Please refer to the numbered items in the image to ensure that the corresponding nodes and models are loaded correctly for the model to run properly:

  1. Ensure that the [umt5_xxl_fp8_e4m3fn_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors?download=true) model can be read and loaded normally
  2. Ensure that you’re using the t2v model and it can be read normally. If not downloaded, click here to download one of the following, depending on your device performance:
    • wan2.1_t2v_1.3B_bf16.safetensors
    • wan2.1_t2v_1.3B_fp16.safetensors
    • wan2.1_t2v_14B_bf16.safetensors
    • wan2.1_t2v_14B_fp8_e4m3fn.safetensors
  3. Ensure that the [wan_2.1_vae.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors?download=true) model can be read and loaded normally
  4. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter to execute video generation

Wan2.1 Quantized Version Workflow

This version is provided by Kijai and requires plugins to use. You need to install the following two plugins:

Wan2.1 Quantized Version Model Downloads

All related models come from: Kijai/WanVideo_comfy

Text encoders models

Please choose either the fp8 or bf16 version based on your device, download the file below, and save it to the ComfyUI/models/text_encoders folder

The fp8 version has lower hardware requirements, while the bf16 version has higher hardware requirements

Diffusion models

Please download the corresponding Transformer model file based on your device and workflow needs, and save it to the ComfyUI/models/diffusion_models folder

Model file parameter description: 1. Model type

  • I2V: Image to Video, image-to-video model
  • T2V: Text to Video, text-to-video model

2. Model size

  • 14B: 140 billion parameter version, better results but higher hardware requirements
  • 1_3B: 13 billion parameter version, lower hardware requirements but relatively simpler results

3. Resolution

  • 480P: Generate 512x512 resolution videos
  • 720P: Generate 768x768 resolution videos

4. Precision format

  • bf16: Brain Floating Point 16, higher precision, requires newer GPUs
  • fp8: Floating Point 8, lower precision, lower hardware requirements
  • fp16: Floating Point 16, medium precision, suitable for most GPUs
  • e4m3fn: One of the 8-bit floating point formats, optimized for low precision calculations

VAE model

Please choose either the fp8 or bf16 version based on your device, download the file below, and save it to the ComfyUI/models/vae folder

Wan2.1 Quantized Version Workflow

Wan2.1 Quantized Version Text-to-Video Workflow

Please click the button below to download the corresponding workflow, then drag it into the ComfyUI interface or use the menu bar Workflows -> Open(Ctrl+O) to load it

Workflow Explanation Wan2.1 Quantized Version Text-to-Video Workflow

Please refer to the numbered items in the image to ensure that the corresponding nodes and models are loaded correctly for the model to run properly

  1. Ensure that the [umt5-xxl-enc-bf16.safetensors](https://huggingface.co/Kijai/WanVideo_comfy/blob/main/umt5-xxl-enc-bf16.safetensors) model can be read and loaded normally
  2. Ensure that the t2v model is used and can be read normally. If not downloaded, click here to download one of the following, depending on your device performance:
  3. Ensure that the Wan2_1_VAE_xxx.safetensors model can be read and loaded normally
  4. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter(回车) to execute video generation

Wan2.1 Quantized Version Image-to-Video 480P Workflow

CLIP Model Download

Workflow Download Please click the button below to download the corresponding workflow, then drag it into the ComfyUI interface or use the menu bar Workflows -> Open(Ctrl+O) to load it

Workflow Explanation Wan2.1 Quantized Version Image-to-Video 480P Workflow

Please refer to the numbered items in the image to ensure that the corresponding nodes and models are loaded correctly for the model to run properly

  1. Ensure that the umt5-xxl-enc-bf16.safetensors model can be read and loaded normally, download link: umt5-xxl-enc-bf16.safetensors
  2. Ensure that the open-clip-xlm-roberta-large-vit-huge-14_fp16.safetensors model can be read and loaded normally, download link: open-clip-xlm-roberta-large-vit-huge-14_fp16.safetensors
  3. Ensure that the Wan2_1_VAE_bf16.safetensors model can be read and loaded normally, download link: Wan2_1_VAE_bf16.safetensors
  4. Upload the image you want to generate a video from
  5. Ensure that the Wan2_1-I2V-14B-480P_fp8_e4m3fn.safetensors model can be read and loaded normally, download link: Wan2_1-I2V-14B-480P_fp8_e4m3fn.safetensors
  6. Input the description content you want to generate a video in the prompt input box
  7. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter(回车) to execute video generation

Wan2.1 Quantized Version Video-to-Video Workflow

Workflow Download Please click the button below to download the corresponding workflow, then drag it into the ComfyUI interface or use the menu bar Workflows -> Open(Ctrl+O) to load it

Workflow Explanation Wan2.1 Quantized Version Video-to-Video Workflow

Please refer to the numbered items in the image to ensure that the corresponding nodes and models are loaded correctly for the model to run properly

  1. Ensure that the umt5-xxl-enc-bf16.safetensors model can be read and loaded normally, download link: umt5-xxl-enc-bf16.safetensors
  2. Upload the original video you want to use for inpainting
  3. Ensure that the Wan2_1-T2V-1_3B_bf16.safetensors model can be read and loaded normally, download link: Wan2_1-T2V-1_3B_bf16.safetensors
  4. Input the description content you want to generate a video in the prompt input box
  5. Ensure that the Wan2_1_VAE_bf16.safetensors model can be read and loaded normally, download link: Wan2_1_VAE_bf16.safetensors
  6. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter to execute video generation

Related Resources