🖼️ Download Image from URL

The simplest possible bridge between "an image exists somewhere on the web" and "an image exists inside my ComfyUI graph." Give it a URL, it fetches the file, decodes it, and hands you back an IMAGE tensor plus its width and height - no manual download-then-upload round trip, no leaving ComfyUI at all.

The real use case is anywhere you'd otherwise be manually saving a reference image to disk just to re-upload it a second later: pulling in a reference for img2img or IPAdapter-style workflows, grabbing images from URLs that came out of another node upstream (an API response, a scraped list, a spreadsheet import), or building a workflow that processes a batch of external image links without any manual step in the middle.

Inputs: image_url is the only required field - a plain URL to the image. save_file_name_override lets you name the local copy something specific instead of whatever the node picks by default. save_path controls where that local copy lands; leave it blank and the tooltip says it defaults to the current directory.

That local-copy behavior is worth noticing: this node doesn't just hand you an in-memory tensor, it writes a copy to disk as it goes, which makes it double as a lightweight archiver for anything you download through it - useful if you're building a reference library from a list of URLs rather than a one-off fetch.

Outputs: image is the decoded picture, ready to wire into anything downstream that expects an IMAGE. width and height come along for free, handy if a downstream node needs to know the source dimensions without a separate sizing step.

Once the file lands on disk, it's a normal ComfyUI input like anything you uploaded by hand - which means this pack's own Get File Path node can turn around and give you a path string back for it, if some later step in your workflow needs the location rather than the tensor.

Installing it: ComfyUI Manager, search "ComfyUI-mnemic-nodes," or git clone https://github.com/MNeMoNiCuZ/ComfyUI-mnemic-nodes into custom_nodes, then restart. No models, no API key - just an HTTP fetch.

What actually fails: URLs behind hotlink protection or that expire after a short window (a lot of CDN and chat-app image links fall into this category) will fail here the same way they'd fail in any plain HTTP client - this node isn't doing anything special to get around access controls, and it can't. Pointing it at a webpage URL instead of a direct link to the image file itself will also fail, since there's nothing to decode as an image. And there's no timeout or retry setting exposed anywhere in this node's inputs, unlike the pack's Groq nodes which do expose a max_retries field - so a slow or dead URL will just sit there rather than failing fast, which is worth knowing if you're batch-processing a long list of links and one of them is bad.