Auto Download ALL WAN Models fromalibaba-pai/Wan2.1-Fun-1.3B-Control

This is the smallest, least-known, and honestly most charming node in the pack: a four-checkbox downloader for alibaba-pai/Wan2.1-Fun-1.3B-Control. That's Alibaba's 1.3-billion-parameter Wan with the Fun-Control adapter - a light, control-aware variant that runs on modest hardware and pairs with a ControlNet-style input in the Fun pipeline. If you're on a 6–8 GB card, this is the Wan you can actually play with, and this node exists so you don't have to piece together four separate files from a repo page by hand.

The display name is a dead giveaway about this pack's polish level - "fromalibaba-pai/Wan2.1-Fun-1.3B-Control", missing the space, and the underlying description is a copy-paste from the Kijai node ("This node downloads ALL Kijay's WAN models"). It's cosmetic noise, but it tells you the author's testing bar: functional, not fancy. Worth knowing before you expect support tickets answered.

The four files, and why you need all of them

A Wan setup is a stack, not a single file, and this node enumerates the whole stack:

• diffusion_pytorch_model → the 1.3B diffusion transformer itself, into /workspace/ComfyUI/models/unet/
• Wan2.1_VAE_pth → the Wan VAE, into vae/
• models_t5_umt5_xxl_enc_bf16_pth → the UMT5-XXL text encoder, into clip/
• models_clip_open_clip_xlm_roberta_large_vit_huge_14_pth → the CLIP vision encoder, into clip/

Tick all four and you've got a complete, runnable model in one execution. The two clip/ files in particular are easy to forget - without the text encoder the model won't even prompt - so the "one checkbox per piece" layout genuinely saves you a round of "which of these 40 files do I need."

Every checkbox is a plain hf_hub_download against the alibaba-pai repo, so downloads resume if interrupted and get the hf_transfer speed boost the pack installs. Unchecked boxes download nothing.

Install and the unavoidable caveat

Pack-wide install, unchanged: ComfyUI Manager → search ComfyUI_AutoDownloadModels → install, or

cd ComfyUI/custom_nodes
git clone https://github.com/AIExplorer25/ComfyUI_AutoDownloadModels
cd ComfyUI_AutoDownloadModels
pip install -r requirements.txt

restart, done.

And the caveat that applies to every node in this pack except the plain AutoDownloadModels: the destinations are hardcoded to /workspace/ComfyUI/models/..., the RunPod/cloud layout. On a local machine, either run it in a cloud box or make the path real:

ln -s /path/to/ComfyUI/models /workspace/ComfyUI/models

The output is a path string (the last file's path, effectively a status echo), and downloads block the queue until finished - for a 1.3B stack that's mercifully short compared to the 14B siblings.

Verdict

Three impressions a month of search traffic says almost nobody uses this node, which is a shame - for 1.3B Fun-Control specifically, the "download the whole stack in one run" idea is at its most useful here because the stack is small enough to actually finish. If you're chasing a 14B Wan, the Kijai and Comfy-Org sibling nodes are the ones you want. If you're on a little GPU and want a controllable Wan without the multi-hour download, this is the quiet best pick in the pack.