🐟Webpage Screenshot

This node is exactly what its name says: it drives a real headless Chrome to a URL, screenshots it, and hands you the result as a standard ComfyUI IMAGE. At 3 impressions it's basically undiscovered, which is a shame, because "grab actual web content as an image" is one of those things that's fiddly to do any other way in ComfyUI.

Where it earns its keep: you want a real-world reference image to img2img against, you want to screenshot your own dashboard or a page you're designing to feed into a vision model, or you want to turn a live web page into a controlnet-friendly input. It also pairs naturally with the rest of this pack - same browser machinery that renders the HTML templates, pointed at the internet instead.

How it works

It's the pack's shared Selenium base doing the work. Chrome opens a window at your width and height, then the node waits for the page's document.readyState to hit "complete" and for every image to finish loading, plus your extra wait_time_seconds on top. Then it either grabs the whole viewport, expands to the full page, or zooms in on a single element and screenshots that.

The inputs that matter

• url - what to shoot. Defaults to https://example.com; put your real target here.
• width / height - viewport size, 64–4096 (default 1280×720).
• wait_time_seconds - extra settle time before the shot. Default 0.5; bump it for slow or JS-heavy pages.
• full_page - boolean. When on, it scrolls to the full document height and ignores height. Great for long articles, but pages with sticky/fixed headers will show the same element repeated.
• css_selector - optional. Give it something like .content or #main-header and it captures just that element instead of the whole page. Useful for grabbing one chart or one card out of a cluttered layout.

Outputs

A single IMAGE output. That's it - wire it into a Save Image, an img2img sampler, or a vision-model node.

Installing

ComfyUI Manager → search "ComfyUI-Html2Image". Or:

cd ComfyUI/custom_nodes
git clone https://github.com/liuqianhonga/ComfyUI-Html2Image.git
pip install -r requirements.txt

Then restart. Dependencies are selenium, webdriver_manager, jinja2, Pillow. The silent requirement: Google Chrome must be installed - Selenium launches a real Chrome binary; webdriver_manager only auto-downloads the matching chromedriver, not the browser itself. On a machine without Chrome this node dies instantly with a "no such binary" style error, and the README won't warn you.

Troubleshooting

• Blank or partial screenshot - the site probably needs login, blocks bots, or loads content after your wait time. The README's own FAQ points at exactly this. Increase wait_time_seconds first; accept that login-gated pages will just come back blank.
• Selector not found - the node raises a clear error for that. Usually the element is rendered dynamically after load, so bump the wait time rather than second-guessing your selector syntax.
• Full-page capture looks broken - fixed-position elements are the usual culprit; the page height calculation also goes sideways on infinite-scroll sites.
• First run is slow - chromedriver downloads once, needs internet.

For a node nobody's talking about, this one is refreshingly honest about what it does. Just remember it renders pages in a real browser, so treat it like a browser: no login, no bot protection, no infinite scroll.