Load Image from URL (Pro)

Core ComfyUI's LoadImage will only read files that already live in your ComfyUI/input folder. Want to use an image that's sitting on a website? Too bad - download it by hand, name it something you can find again, then wire it in. That chore is exactly what Load Image from URL (Pro) removes: you paste one or more URLs and it fetches the images over HTTP, right in the graph.

It's a slightly odd node to find inside comfyui-save-image-pro, a pack whose headline business is writing images to disk with fancy filenames. The URL-loading pair was bolted on in the October 2025 release, clearly borrowed from the comfyui-easyapi-nodes original and hardened with better error handling, a timeout, and logging. If you've ever wanted to batch-process a whole gallery of reference images, pull an image from an API, or test a workflow with https://picsum.photos links, this is the node that ends the download dance.

How it works

The urls input is a multiline text box: one URL per line, blank lines ignored. On each queue run the node goes through them in order, and for each one it:

• sends an HTTP GET (streamed, via requests) and checks the response status;
• decodes the bytes with Pillow and applies EXIF rotation automatically, so phone photos don't come in sideways;
• converts to RGB and returns a float image tensor of shape (1, H, W, 3) - exactly what LoadImage gives you, just sourced from the web.

A couple of things worth knowing before you wire it up. The outputs are lists: images (IMAGE) and masks (MASK). That's how one node hands you a whole batch - drop it into any input that takes an image list and downstream iterates. And if an image has an alpha channel, it's automatically extracted into a mask that's inverted (transparent pixels become white/1.0). Images without alpha just get an all-black mask, so don't assume the masks output is useful for a plain JPEG.

The inputs you'll actually touch: urls is the only required one. In the current build there's also a per-request timeout (30 seconds by default, up to 300) and a verify_ssl toggle that defaults to on - flip it off if your source uses a self-signed certificate. Failure handling is per-URL: a bad link is logged and skipped, not fatal.

Installing it

Search "ComfyUI Save Image Pro" in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/weekii/comfyui-save-image-pro.git

then restart ComfyUI. Dependencies are light - requests, pillow, piexif, pillow-avif-plugin - and requests is almost certainly already in your environment. No model downloads, no weights, no VRAM cost. (The README is mostly Chinese; don't let that put you off, the node UI is self-explanatory.)

Where people get burned

• Everything comes back black. If all URLs fail, the node quietly returns a 64×64 black image instead of erroring - which can look like "the node is broken" when it's really "every download failed." The console logs are your friend: look for Error reading image from URL.
• A 100-URL batch is slow. Fetches are serial, one at a time, up to 30s each. A slow host means the whole queue hangs on this node.
• Hotlink-protected hosts. Some image CDNs reject non-browser requests. If a URL works in your browser but not here, that's the host blocking, not the node - download it and use LoadImage instead.
• Name collision. If you also have comfyui-easyapi-nodes installed, its LoadImageFromURL and this LoadImageFromURLPro will sit next to each other in the search. They do the same job; the Pro one is the newer, better-logged sibling. The README itself calls out the distinction.

Wire the images output into anything that wants an IMAGE - an img2img pipeline, a reference loader, an IPAdapter input - and you've got a workflow that pulls its source material straight off the internet.