☁️ 加载图片

Sometimes the image you want isn't on your disk - it's sitting on a server somewhere and you'd rather ComfyUI just go get it. That's all this node does: paste a URL, get an IMAGE tensor out the other side. In a pack full of cloud-upload utilities (comfyUI_LLM), this is the natural complement - the author's workflow uploads images to a CDN and then pulls them back down by URL, and this node is the "pull" half.

How it works

Not much to it, mechanically. The node does a requests.get(url, stream=True, timeout=10), opens the response with Pillow, applies exif_transpose (so phone photos come in the right way up instead of sideways), and converts to a float32 tensor in [0,1] with shape (1, H, W, C). Grayscale images get an extra dimension so they match the 4D shape ComfyUI expects. If the URL 404s or times out, you get a raised error in the console - nothing graceful, no retry.

Inputs and output

Just the one input, url (STRING) - paste or wire in the URL. The output is image (IMAGE), a single-image batch you can plug into anything that takes an IMAGE: VAE encode, a ControlNet preprocessor, an image-to-video model, or the LLM-vision-style captioning flows this pack's author tends to build.

One thing to know: the default value in the node is a URL on the author's own Qiniu CDN (swqqsa5wv.hb-bkt.clouddn.com). It's a working example, but it's their image - swap it for your own URL before you run anything serious.

Installing it

Same drill as every node in this pack - ComfyUI Manager → search comfyUI_LLM, or:

cd ComfyUI/custom_nodes
git clone https://github.com/XieJunchen/comfyUI_LLM
# restart ComfyUI

The requests dependency comes in via the pack's requirements.txt, so a normal install covers it.

When it bites you

• No auth, no headers, no cookies. If the image sits behind a signed URL, a login, or a referrer check, this node can't fetch it. Download it first and use the normal Load Image node.
• 10-second timeout, no retry. Big images on slow CDNs can just fail. Retry the queue.
• It's a pull, not a preview. There's no widget thumbnail here - you don't get the nice preview pane like ComfyUI's built-in loader.

Honestly? If you mostly work locally, the built-in Load Image node does everything you need and adds a picker. Reach for this one when your workflow is genuinely URL-driven - e.g. you're pulling uploaded results back out of cloud storage as part of an automated pipeline, which is exactly the niche it was built for. For that case it's dead simple and does the job.