Daply Image Loader URL

Every ComfyUI user eventually hits the same wall. You want to run a reference through ControlNet or an IP-Adapter, or img2img a photo, and the image lives at a URL - a CivitAI preview, a Discord link, a gallery you're scraping by hand. Core LoadImage only shows you what's already sitting in input/, so the ritual is: download the file, drag it into the input folder, refresh the dropdown, load it again. Daply Image Loader URL is the fix for exactly that chore. Paste one URL - or ten - into a text box, hit run, and it fetches them all and hands you a batched IMAGE tensor ready to wire into the rest of the graph. No download step, no input/ folder management.

The name is honest, which is refreshing: it calls no API, needs no key, and wants no models. It's a plain HTTP fetch wrapped in a ComfyUI node, from a one-node pack by an author with essentially zero community footprint. That's fine. For a utility this simple, "no reputation" just means "nothing to be scared of either."

How it works

The mechanism is refreshingly transparent (you can read the whole node in one file). The urls_or_paths box is split on line breaks, whitespace-stripped, and each line is treated as a source: if it starts with http:// or https:// it's fetched with a requests.get call (browser user-agent, 30-second timeout); otherwise it's opened as a local file. Each image is loaded with Pillow, animated formats are walked frame by frame, and everything gets normalized to 0–1 float tensors and concatenated into a single batch. The code pulls an EXIF transpose in there too, so phone photos come in the right way up.

The inputs that matter

There are only three, and honestly only two you'll touch:

• urls_or_paths - the whole node. One source per line, either an https:// URL or an absolute local path like /home/you/refs/photo.png. Relative paths do not resolve against your input/ folder, which trips people up on day one.
• output_format - RGB by default. Switch to RGBA if you need the alpha channel downstream, or Grayscale if you're feeding mask/luma work.
• convert_webp_to_png - defaults on. Worth knowing: the pipeline re-encodes WebP frames regardless; this flag mostly controls whether the reported filename gets a .png extension.

The outputs

• image - the batched IMAGE tensor, all frames from all sources stacked along the batch dimension.
• mask - computed as 1 − alpha where the source has an alpha channel (the same convention core LoadImage uses); a 64×64 zeros tensor otherwise.
• count - total frame count across everything you loaded. One still per line, so this equals your number of lines.
• filenames - a single string, names joined by newlines, one per frame, with .webp rewritten to .png when converted.

Where people get burned

The source makes the failure modes easy to predict, and they're worth knowing before you wire this into a workflow:

1. Mixed image sizes crash the batch. The node does a torch.cat across all sources, so every image must share the same height and width. Feed a 1024 and a 512 in one list and you get an error, not a resize. Load same-size images, or use one node instance per size.
2. Absolute paths only. A relative path or a ~/ shortcut won't expand - the check runs against the ComfyUI working directory, not your input/ folder. Give the full path.
3. The mask is only real when there's alpha. Opaque sources (most JPEGs, most WebP) produce a useless 64×64 zero tensor. Wire mask into a mask-consuming node only if your sources genuinely carry transparency.
4. 30-second timeout, no auth, hotlink-prone. Slow CDNs will fail the fetch, sites with hotlink protection 403 it, and private/authenticated URLs just won't load - the request sends a browser user-agent and nothing else.

Installing it

It's one of the easiest installs in the ecosystem, because it has no model files and no heavy dependencies - just requests, Pillow, and numpy (torch comes from ComfyUI itself).

ComfyUI Manager → Install Custom Nodes → search Comfy UI Daply Nodes → Install, then restart.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Daply-AI/comfy_ui_daply_nodes.git
pip install -r comfy_ui_daply_nodes/requirements.txt

Restart ComfyUI and the node appears under the image category.

Bottom line

Is this the node you reach for daily? Probably not - if your references already live in input/, core LoadImage does the job and always will. But the day you need to pull several remote images into one batch without the download-and-drag routine, this is exactly the right tool, and it gets out of the way. Just keep your sources the same size and your paths absolute.