📥 UpscaleModel Download

The point of this node is to skip the download-and-drop dance. Normally, to use a new upscale model like 4x-UltraSharp, you go find the file, download it, drop it in models/upscale_models, restart, then load it. This node collapses all of that into a link. Paste a civitai or HuggingFace URL, and it downloads the model straight into the right folder and hands you a loaded UPSCALE_MODEL you can wire in immediately. First run fetches it; every run after, it's already there.

That makes it genuinely useful for portable workflows - a graph you're sharing, or one running on cloud/serverless where you didn't pre-load your model folder. The model travels with the workflow instead of being an unwritten prerequisite.

How it works

Under the hood it's a downloader plus a loader. Per the pack's docs, all of SDVN's download nodes take a direct model link from civitai or HuggingFace and fetch it, using aria2c for accelerated multi-connection downloads when it's available. It saves the file under the name you give it, then loads it as an upscale model and outputs it.

Worth being clear about what kind of upscaler this is. These are pixel upscalers - the ESRGAN family (4x-UltraSharp, Remacri, the anime variants, and so on). They add resolution, not content. They run in milliseconds, use almost no VRAM, and cannot hallucinate detail that wasn't there. That's the right tool when your image is already sharp and you just need it bigger. If your source is soft or damaged and you want a model to invent plausible detail, that's a different job entirely (SeedVR2, SUPIR, tile-ControlNet) and a different node.

The inputs and outputs that matter

• Download_url (STRING) - the direct download link to the model, from civitai or HuggingFace.
• Url_name (STRING, default model.safetensors) - the filename it saves as. Give it something descriptive with the right extension, like 4x-UltraSharp.pth, so you can find it later.

The output is UPSCALE_MODEL - wire it straight into an "Upscale Image (using Model)" node (the pack's own Upscale Image has a model slot for exactly this).

How to install it

Comes with the pack. ComfyUI Manager, search SDVN_Comfy_node; or:

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

from the ComfyUI root, then restart.

Common issues & troubleshooting

On Windows or macOS, install aria2c yourself. The pack's README calls this out directly: the auto-download nodes need aria2c, and it's not bundled on Windows/macOS. On Linux it's usually already there. Without it, downloads may fail or fall back to something slower.

The URL has to be a direct download link, not a model's web page. A civitai model page isn't the same as the download endpoint for the file - grab the actual file link.

Give it a sensible filename with an extension. Leaving it as model.safetensors works, but if you fetch several models you'll end up overwriting or confusing them. Name each one.

Don't expect it to add detail. If you point this at a blurry photo and it doesn't magically sharpen, that's not a bug - pixel upscalers don't restore, they enlarge. Use a generative upscaler for restoration and keep this for clean sources that just need more pixels.