Wan 2.2 ComfyUI Models Folder: Every File and Its Path

Quick answer

Wan 2.2 in ComfyUI uses four folders under ComfyUI/models/. The video model goes in diffusion_models/, the text encoder in text_encoders/, the VAE in vae/ and the 4-step LoRAs in loras/. The one trap is the VAE: the 5B needs wan2.2_vae.safetensors, and every 14B workflow uses the older wan_2.1_vae.safetensors.

For the 5B text-and-image-to-video template, video_wan2_2_5B_ti2v.json:

ComfyUI/
└── models/
    ├── diffusion_models/
    │   └── wan2.2_ti2v_5B_fp16.safetensors          10.00 GB
    ├── text_encoders/
    │   └── umt5_xxl_fp8_e4m3fn_scaled.safetensors    6.74 GB
    └── vae/
        └── wan2.2_vae.safetensors                    1.41 GB

For the 14B text-to-video template, video_wan2_2_14B_t2v.json:

ComfyUI/
└── models/
    ├── diffusion_models/
    │   ├── wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors        14.29 GB
    │   └── wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors         14.29 GB
    ├── loras/
    │   ├── wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors   1.23 GB
    │   └── wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors    1.23 GB
    ├── text_encoders/
    │   └── umt5_xxl_fp8_e4m3fn_scaled.safetensors                   6.74 GB
    └── vae/
        └── wan_2.1_vae.safetensors                                  0.25 GB

Every file comes from the Comfy-Org/Wan_2.2_ComfyUI_Repackaged repository on Hugging Face. Inside it, each file sits under split_files/<folder>/, and that folder name is the one it goes in on your side.

We read the folder paths from ComfyUI's documentation and its source, and the file lists from the shipped template files, on 2026-10-09. Sizes are from the Hugging Face API, in decimal gigabytes.

What each template loads

The 11 Wan 2.2 templates bundled with ComfyUI, and the files each one asks for. The text_encoders/ column is the same for all of them: umt5_xxl_fp8_e4m3fn_scaled.safetensors.

Two templates need a fifth folder:

• S2V loads wav2vec2_large_english_fp16.safetensors (0.63 GB) from audio_encoders/. ComfyUI's tutorial says to create the folder if you cannot find it.
• Animate loads clip_vision_h.safetensors (1.26 GB) from clip_vision/. That file is not in the Wan 2.2 repack; it is in Comfy-Org/Wan_2.1_ComfyUI_repackaged.

The 14B models are always two files of the same size, a high-noise and a low-noise expert, and both go in diffusion_models/. The template loads each with its own Load Diffusion Model node. The full size list, with FP16 versions, is on our Wan 2.2 VRAM requirements page.

GGUF files go in unet/

The ComfyUI-GGUF custom node's README says: place the .gguf model files in ComfyUI/models/unet, and load them with its Unet Loader (GGUF) node instead of Load Diffusion Model. For the 14B you still need two GGUF files, one per expert, of the same quant.

The text encoder, VAE and LoRAs stay where they were. A GGUF workflow only replaces the video model files.

ComfyUI itself treats unet/ as an old name for diffusion_models/: its folder_paths.py registers both paths under the same key, so Load Diffusion Model lists .safetensors files from either folder.

Four places the official docs and templates disagree

These are worth knowing before you download 38 GB.

1. The I2V download cards list FP16 files, the template loads FP8. ComfyUI's Wan 2.2 tutorial links the 28.58 GB FP16 I2V experts, but video_wan2_2_14B_i2v.json asks for the 14.29 GB fp8_scaled ones. Download the FP8 pair, or switch the loader to the FP16 file names yourself. The same section's step text names T2V files.
2. The tutorial's download lists leave out the 4-step LoRAs. The 14B T2V and I2V templates load them anyway, so a workflow downloaded from the list alone stops at a missing LoRA. They are in the repack under split_files/loras/.
3. The tutorial's lightx2v link returns 404. It points at a folder name without -Seko-; the folders in lightx2v/Wan2.2-Lightning today all carry it. The repack's copies are the same bytes as the Seko V1 and V1.1 files, so the repack is the simpler source.
4. The Animate tutorial's folder tree says clip_visions/. ComfyUI's real folder is clip_vision/, singular, per folder_paths.py and the template's own metadata. A file in clip_visions/ is not seen.

The text encoder link in every template points at the Wan 2.1 repack, not the Wan 2.2 one. That is not an error: the two copies are byte-identical, so either works and you only need one.

When ComfyUI does not see a file

• Restart, or refresh the model list. ComfyUI reads the folders at start-up; a file added while it runs needs a refresh before it appears in a loader's dropdown.
• Check the extension and the folder. .safetensors in diffusion_models/ for Load Diffusion Model, .gguf in unet/ for the GGUF loader.
• Check you are in the right install. The desktop app and the portable build keep their models/ folders in different places, and models can also live outside both through extra_model_paths.yaml. Our ComfyUI download page covers where each install keeps its files.
• Do not put these in checkpoints/. None of the Wan 2.2 files is a checkpoint; each is one part, and Load Checkpoint cannot read it.

Licence and downloads

Wan 2.2 is released under the Apache 2.0 licence, and the repacks, GGUF builds and lightx2v LoRAs named here carry the same tag. Two exceptions: the Animate template's Kijai files are separate uploads, and the Wan 2.2 repack also carries NVIDIA ChronoEdit files, which are not part of Wan 2.2. Check their own pages.

We do not host any of these files. Download them from the repositories named below.

To check whether your GPU and system RAM can hold a set like this, use the system requirements checker. Wan 2.2 is not one of its presets yet.

GenVidKit is an independent guide. It is not affiliated with Alibaba, the Wan team, Comfy Org, Kijai, QuantStack, lightx2v or Hugging Face.

Sources

All read on 2026-10-09.

• Wan2.2 native workflow, ComfyUI documentation — folder paths, download cards, the I2V and lightx2v discrepancies.
• Wan2.2 S2V and Wan2.2 Animate, ComfyUI documentation — the extra folders.
• ComfyUI_examples: Wan 2.2 — the VAE split between 5B and 14B.
• Comfy-Org/workflow_templates — what each video_wan2_2_*.json loads.
• Comfy-Org/Wan_2.2_ComfyUI_Repackaged and Comfy-Org/Wan_2.1_ComfyUI_repackaged — files and byte counts.
• lightx2v/Wan2.2-Lightning — the LoRA folders.
• ComfyUI folder_paths.py — folder names and the unet alias.
• city96/ComfyUI-GGUF — the GGUF folder and loader.