Hypereel Product Scrape (URL to references)

The Hypereel pipeline in symbiotica-ai/comfyui-nodes is built for UGC-style ads - a product gets a script, a facecam, and a gameplay-style background. Which means every reel starts with the same chore: gather the product's logo and screenshots, and write a summary of what it is. Hypereel Product Scrape does that from a single URL. Paste a product page, app page, or app-store listing, and it hands back a text summary plus the logo and up to three screenshots as IMAGE outputs.

That "from a URL" is doing a lot of work. The scrape isn't a dumb screenshot grab: it follows the first app-store link for the curated promo screens, promotes the AppIcon to the logo, prefers the biggest image variants, drops badges and template URLs, and flattens transparency onto white (so dark brand marks on transparent backgrounds survive as dark lettering, not black blobs). The summary output is written for the script LLM - it drives the CTA rule from the platform you picked.

The inputs that matter

• url - product page, app page, or app-store listing. The only input you must actually fill in.
• platform - mobile app / desktop app / physical product. Doesn't change what's scraped; it changes the CTA rule baked into the summary (a mobile app wants "download", a physical product wants "buy").
• include_details - optional, off by default. Appends a page-text DETAILS digest to the summary. Off matches the platform's engine exactly; turn it on when you want more raw material for the script LLM.

Outputs

• summary (STRING) → the PRODUCT block for the script LLM.
• logo (IMAGE) → feeds the video node's reference slot (avatar first, then logo as image_2 per the Hypereel wiring).
• screenshot_1..3 (IMAGE) → the remaining reference slots.
• found (INT) - how many image assets actually came back, so you can tell at a glance whether the scrape found a real gallery or came up empty.

The part most people don't expect: it's SSRF-guarded

Fetching URLs on the user's behalf is exactly the shape of thing that gets abused to probe internal networks, and the node takes that seriously: each request goes through a guard that resolves the host before trusting it, and follows redirects one hop at a time re-checking each target, so a 3xx redirect to a private or metadata address can't slip past. Non-public targets are refused. That's a genuinely thoughtful bit of engineering for a scrape node.

Install

Part of symbiotica-ai/comfyui-nodes (ComfyUI Manager → search "Symbiotica", or clone + pip install -r requirements.txt). No API key - it's plain HTTP fetching. Restart ComfyUI after installing.

Gotchas

Scraping is scraping: sites can block, the structure can change, and found is your early-warning system for "it returned nothing." If found is 0, check the URL is a page the scraper can actually reach (and that it's public). And the node follows one app-store link for the curated screens - if the page's layout differs from what the platform engine expects, you may get the logo but not all three screenshots, which is why screenshot_2/screenshot_3 are separate outputs rather than one batch. Wire whatever exists; the Hypereel flow handles missing slots gracefully.