🏞️ Load Image

The one feature that makes this worth using over ComfyUI's built-in loader is right there in the inputs: it loads an image from a URL, not just from your input folder. Paste a web link, get the image, no downloading-and-uploading dance. That's the whole reason to reach for it, and it's a real time-saver when you're testing a workflow against some reference you found online.

It works as a two-mode loader. The Load_url boolean flips between them. With it on, the node reads the Url field and pulls the image straight off the web - and per the pack's docs this includes smart handling of Pinterest links, where it'll go find the highest-quality version rather than the thumbnail. With Load_url off, it falls back to the familiar image dropdown that lists what's already sitting in your ComfyUI input folder, subfolders included. Either way you get the same three outputs.

Those outputs are image, mask, and img_path. The image is the picture itself, ready to wire into a VAE encode, a preprocessor, a resize, whatever's next. The mask is the alpha channel - if you loaded a transparent PNG, you get its cutout mask for free here, which is a tidy way to recover an existing cutout without re-running a background remover. And img_path is the resolved path or location as a string, useful when a downstream node needs to know where the file actually lives, or when you're logging what got loaded.

The inputs that matter are really just the mode switch and the source: flip Load_url on and fill Url to grab from the web, or leave it off and pick from the image dropdown for something already uploaded. That's it. There's no resize baked into this particular loader - it hands you the image at its native size - so if you need it at a specific resolution, put one of the pack's upscale/resize nodes right after it.

Why you'd want it: the URL path is the killer feature for quick iteration. Instead of right-click-save-then-upload every reference you want to try, you paste the link and go. For a workflow you're building and re-running against different source images, that's genuinely faster, and the Pinterest quality-hunting is a nice touch if that's where your references come from. Against the native ComfyUI loader the trade is dead simple - the built-in is fine once a file is uploaded, but the moment you want a URL, this saves you the round trip.

Install comes with the pack. ComfyUI Manager, search SDVN_Comfy_node; or cd ComfyUI/custom_nodes && git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node, then pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt from your ComfyUI root and restart. No weights involved - it's pure I/O.

Troubleshooting is mostly about the URL path. If a link fails to load, the usual culprits are a page that needs a login (the plain URL fetch can't see anything behind auth) or a site that blocks hotlinking - for scraping galleries and cookie-gated sources you'd want the pack's Image Scraper node instead, which is built for that. On the download side, the auto-fetch leans on aria2c for speed, and on Windows or macOS you have to install aria2c yourself for the download-capable nodes to behave; on Linux it's usually already there. And if you loaded a transparent PNG but the mask comes out empty, the file didn't actually carry an alpha channel - a lot of "transparent-looking" images are just white backgrounds. Beyond that it's about as low-drama as a loader gets.