AI_web_tool

This is the generalized, browser-automation half of the Doubao workaround the pack ships - AI_DoubaoWebPreview is its narrower sibling. Where that node just waits and retrieves, AI_web_tool is the one that actually drives a browser: it opens a page, does whatever it's built to do there, and pulls the resulting images back into your graph. The default page_url gives away exactly what it was built and tested against: https://www.doubao.com/, ByteDance's free consumer AI web app.

There's no required input at all - every field is optional, which tells you this node has sensible defaults for exactly the use case it was designed around (doubao.com) and is technically flexible enough to point elsewhere if you're willing to experiment. page_url (default the Doubao homepage) is where it navigates. image_address lets you pass in an image reference - likely a URL or file path the automated session uses as input to whatever the web app is doing, though the exact mechanics of how the site consumes it depend on that page's own UI, which this node has to interact with blind. image_count (1–32) caps how many images it collects, and download_dir sets where they land on disk. The output is image, a list of IMAGEs.

The reason a node like this exists at all comes back to the same access problem as the rest of the pack's Doubao tooling: ByteDance's Seedream API is sold through Volcano Engine, a Chinese-market-first cloud console that commonly expects a mainland China phone number to verify an account. The public web app doesn't have that barrier - anyone can open it in a browser - so automating the browser sidesteps paying for and signing up to an API entirely. It's a genuinely clever workaround, and also, structurally, the most fragile node in this whole pack.

Installing it: search "ComfyUI-Apt_Preset" in ComfyUI Manager, or cd ComfyUI/custom_nodes && git clone https://github.com/cardenluo/ComfyUI-Apt_Preset followed by install.bat and a restart. Expect this node specifically to depend on a headless browser automation library (Playwright/Selenium-class tooling plus a Chromium install) - if it fails to import after the standard install, that dependency is the first thing to check by hand rather than assuming the whole pack is broken.

Troubleshooting: because this node works by scripting a real website rather than calling a documented API, it inherits every problem that comes with that - layout changes, new login walls, CAPTCHAs, regional blocks, and rate limiting can all break it without any error message that points at the actual cause. If it returns nothing, start by opening the same page_url manually in a normal browser from the same machine and confirming the page loads and behaves the way you'd expect a fresh session to - if you hit a login prompt or a verification step by hand, the automated node is hitting the same wall silently. And don't expect this to work reliably against arbitrary AI web apps just because page_url is editable - it was built and tuned against doubao.com's specific page structure, so pointing it somewhere else is unsupported territory, not a documented feature.