Wan 2.2 Animate 14B ComfyUI on Kaggle

A full solution that installs and runs Kijai's workflows for Wan 2.2 animate 14B GGUF model in a constrained Kaggle environment. The script also links or copies pre-downloaded model files (GGUF, text encoders, VAEs, LoRAs, and vision modules) directly from Kaggle Datasets while maintaining a clean directory structure compatible with ComfyUI Manager.

There are 2 different types of notebooks and 3 different variants of Kijai's modified workflows here.

• wan-2-2-animate-comfyui.ipynb: Uses Local Tunnel to expose ComfyUI for accessibility
• wan-2-2-animate-comfyui-gradio.ipynb: Uses Gradio and ComfyUI API for direct use of the workflow (More stable than Localtunnel, but less flexible in configuration because no direct access to ComfyUI interface)

ComfyUI Kaggle Environment (CUDA 12.8 / PyTorch 2.9 / Multi-GPU)

This is a fully-automated setup script for running ComfyUI on Kaggle Notebooks with:

• CUDA 12.8
• PyTorch 2.9.0 + cu128
• Multi-GPU support
• ONNXRuntime GPU
• All required ComfyUI plugins
• Automatic symbolic linking of models
• Full dependency management + numpy constraints

System Requirements

• Kaggle Nvidia 2 x T4 GPU (16 GB VRAM)
• CPU with 32 GB RAM

Python Environment

Numpy Constraint

• numpy<2.0

Constraint file: /kaggle/working/constraints.txt

Environment variable: PIP_CONSTRAINT=/kaggle/working/constraints.txt

Kaggle Runtime

• GPU accelerator enabled (T4 or better)
• Ubuntu 22.04 container

Prerequisites

Before running the notebook on Kaggle, there are models and files you may need to install. The notebooks will automatically symlink the model files locations to the ComfyUI kaggle/working/ComfyUI directory, but you'll need to ensure that the models are placed in the correct kaggle input directory with the exact name mentioned here. Below are the specifications and the links to the models:

Models

Note that wan-2-2-animate-comfyui-gradio.ipynb uses q6_k, and wan-2-2-animate-comfyui.ipynb notebook uses q3_k_m by default. If you prefer to use a different model for a notebook, you have to change the model name in the link_map found within code.

GGUF Model /kaggle/input/wan-animate-q3_k_m-gguf or /kaggle/input/wan-animate-q6_k-gguf

• Wan 2.2 Animate 14B Q3_K_M GGUF — https://huggingface.co/QuantStack/Wan2.2-Animate-14B-GGUF/blob/main/Wan2.2-Animate-14B-Q3_K_M.gguf
• Wan 2.2 Animate 14B Q6_K GGUF — https://huggingface.co/QuantStack/Wan2.2-Animate-14B-GGUF/blob/main/Wan2.2-Animate-14B-Q6_K.gguf

Text Encoder /kaggle/input/wan-animate-text-encoder

• UMT5 XXL FP16 Safetensor — https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/blob/main/split_files/text_encoders/umt5_xxl_fp16.safetensors

VAE /kaggle/input/wan-animate-vae

• Wan 2.1 VAE BF16 Safetensor — https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan2_1_VAE_bf16.safetensors

LoRAs /kaggle/input/wan-animate-loras

• Wan Animate Relight LoRA FP16 Safetensor — https://huggingface.co/Kijai/WanVideo_comfy/blob/main/LoRAs/Wan22_relight/WanAnimate_relight_lora_fp16.safetensors
• LightX2V I2V 14B 480p — https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors

Clip Vision /kaggle/input/wan-animate-clip-vision

• Clip Vision Safetensor — https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files/clip_vision

Custom Nodes

Manual installation of custom nodes is not needed, but it can skip the git pull and installation of them on Kaggle notebook if you upload them to the kaggle input.

Core Plugins /kaggle/input/wan-animate-custom-nodes/custom_nodes

• ComfyUI-Manager — https://github.com/ltdrdata/ComfyUI-Manager.git
• ComfyUI-GGUF — https://github.com/city96/ComfyUI-GGUF.git
• ComfyUI-WanVideoWrapper — https://github.com/kijai/ComfyUI-WanVideoWrapper.git

Kijai Workflow Requirements

• ComfyUI-KJNodes — https://github.com/kijai/ComfyUI-KJNodes.git
• comfyui_controlnet_aux — https://github.com/Fannovel16/comfyui_controlnet_aux.git
• ComfyUI-VideoHelperSuite — https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite.git
• ComfyUI-segment-anything-2 — https://github.com/kijai/ComfyUI-segment-anything-2.git
• ComfyUI-Advanced-ControlNet — https://github.com/Kosinkadink/ComfyUI-Advanced-ControlNet.git
• comfyui-tooling-nodes — https://github.com/Acly/comfyui-tooling-nodes.git
• ComfyUI_essentials — https://github.com/cubiq/ComfyUI_essentials.git

Additional Required Nodes

• cg-use-everywhere — https://github.com/chrisgoringe/cg-use-everywhere.git
• was-node-suite-comfyui — https://github.com/WASasquatch/was-node-suite-comfyui.git
• ComfyUI-Custom-Scripts — https://github.com/pythongosssss/ComfyUI-Custom-Scripts.git
• ComfyUI-Frame-Interpolation — https://github.com/Fannovel16/ComfyUI-Frame-Interpolation.git
• rgthree-comfy — https://github.com/rgthree/rgthree-comfy.git
• ComfyMath — https://github.com/evanspearman/ComfyMath.git
• ComfyUI-Easy-Use — https://github.com/yolain/ComfyUI-Easy-Use.git
• comfyui-mixlab-nodes — https://github.com/MixLabPro/comfyui-mixlab-nodes.git

Setup

1. Download your preferred notebook and workflow from this repository.
2. Download the models from the respective links as specified above.
3. Log in to Kaggle and import your notebook. Make sure that the internet and accelerator GPU 2 X T4 is set for your notebook
4. Upload those models as datasets into the respective directories in kaggle input. Make sure the names of the uploads are exactly the same as in their directories specified above
5. Run the notebook (either in interactive session or Save Version)