📥 HF Download (Rename)

Half of ComfyUI's setup is "go download this file from a HuggingFace repo and put it in the right models folder." This node automates that inside the graph, and it's the good kind of automation: it uses hf_transfer for multi-connection downloads (real speed gain on big checkpoints), supports resuming, lets you rename the file on the way in, and drops it into the exact models/ subfolder the loaders expect. The default values in the node are even pre-filled with a real example (Comfy-Org/z_image_turbo, the Z-Image turbo VAE path) so you can see the shape of a working call before you replace it.

Inputs that matter

• repo_id - the HF repository, e.g. black-forest-labs/flux1-dev or Comfy-Org/z_image_turbo.
• filename - the file within that repo, subdirectories included, e.g. split_files/vae/ae.safetensors. This is the input people get wrong most - you must name a real path inside the repo.
• save_dir - dropdown of your ComfyUI models/ subfolders. This is what makes the file actually usable: save a checkpoint to checkpoints, a VAE to vae, and the loaders see it immediately.
• custom_name - rename on download (the "(Rename)" in the display name). Leave empty to keep the original filename.
• hf_token - for gated models, paste your token. Optional, and a reminder that gated models need a logged-in HF account with granted access regardless.
• overwrite - default False; re-download only if you turn it on.
• save_dir_override - a raw path that bypasses the dropdown if your target folder isn't in the list.

Output: a status string reporting success or the error. The node needs huggingface_hub (plus optionally hf_transfer, which it tries to install itself for speed).

When you'd use it

Two good cases. First, template workflows: if you share a graph with friends, baking in the model download node means they run the graph and the files fetch themselves instead of you writing a four-line install instruction. Second, model-staging on a fresh machine or a RunPod-style instance where you want the setup in-graph. For everyday single downloads, honestly, huggingface-cli download in a terminal is equally fast - the node's advantage is repeatability and the folder routing.

Gotchas

• If hf_transfer isn't installed, the node tries to pip-install it on the fly. On a locked-down environment that fails gracefully and downloads at normal speed - fine, just slower.
• Watch the model's own requirements: downloading the safetensors is step one, but a repo often wants its config files next to it. The node fetches a single file; if the loader complains about missing config, you still need the repo's other files.
• Filenames with slashes in the repo need custom_name set if you want them flat in one folder, or use save_dir_override to keep the structure.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/amir84ferdos/ComfyUI-ArchAi3d-Qwen.git
cd ComfyUI-ArchAi3d-Qwen && pip install -r requirements.txt

Or ComfyUI Manager → search "ArchAi3d Qwen" → Install → restart. Category: ArchAi3d/Download. And the standing license reminder: free personal, paid commercial for the pack.