Silver Load Largest Web Image

The URL Image Loader wants a direct image link. This one is the scrappier cousin: give it a webpage - a blog post, a portfolio, a wiki article - and it finds the biggest content image on the page and downloads it. The workflow it's designed for: page URLs that serve a random featured image (the README literally targets /random/ endpoints), where you can't know the image URL ahead of time. Point it at the page, get a fresh reference photo every run.

How it works

It fetches the page HTML with a browser-ish User-Agent and parses it with BeautifulSoup. Every <img> element gets its size estimated from the width/height attributes, inline CSS, or data-* attributes (the lazy-load fields like data-original-width), and images are ranked by that area. A keyword blacklist - you control it via exclude, and the default is a long list of logo, icon, banner, thumb, advert, sponsor, widget, related and friends, plus sprite and thumbs always - filters out the junk. The node then tries the biggest candidate first, sending a Referer header to get past hotlink protection, verifies the content-type is actually an image, and content-hashes the bytes so it won't hand you the same image twice.

Inputs

• url - the webpage URL, not an image URL.
• exclude - comma-separated keywords. The default is surprisingly good; shorten it if the node is filtering out the image you actually want.
• seed - set above 0 for consistent selection behavior.
• timeout - optional, seconds, default 10.

Outputs

• IMAGE - the downloaded image.
• MASK - a solid 1.0 mask, present for wiring convenience; nothing meaningful in it.

Gotchas

The big honesty note, straight from the README: "Largest" is determined by HTML attributes, not actual download size. If a page lies about its image dimensions, or lazy-loads images with no width/height anywhere, the node can pick the wrong image - or skip the real hero image entirely because it reported zero area. It's a best-effort heuristic, not a guarantee.

Scraping also means you're at the mercy of the site. Cloudflare-fronted pages and aggressive bot-detection will 403 despite the retries (five attempts with backoff), and JavaScript-rendered pages where images only appear after scripts run won't have them in the raw HTML at all - no headless browser here, just the source. And while it politely respects content-type checks and dedup, this is still crawling someone else's site; keep it to pages you're allowed to fetch, don't hammer, and don't be surprised when a site blocks you.

Install

ComfyUI Manager → search "ComfyUI Silver Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/SilverAndJade/comfyui-silver-nodes

Restart after cloning. This is the one node in the pack that genuinely needs a dependency beyond stock ComfyUI: BeautifulSoup for HTML parsing. It's in the pack's requirements.txt; if the node fails on import, run pip install -r requirements.txt in your ComfyUI venv and restart.