Building a scan-cropping tool with Gemini beat finding one
Auto-placed crop zones with perspective correction and edge snapping, in the browser. It wasn't right on the first try, but building it was much faster than hunting for an existing tool.
Done withGeminiSpotted· by HN user TeMPOraL· Oct 18, 2026
Result Perspective correction and edge snapping included
“I had Gemini single-shot me a proper in-browser photo editing tool for autocropping photos of scans”
Spotted: found in the wild. Details are as reported by the source above and have not been independently verified.
What they did
They had Gemini single-shot a proper in-browser photo editing tool for autocropping photos of scans, complete with auto-placing a non-rectangular quad cropping zone for rotation and perspective correction, edge snapping and other editing features.
How it went
It wasn’t correct on the first try, but having AI build the tool was much faster than finding an existing one for the job, even with all the heavy lifting done through a JS port of OpenCV.
Try it yourself with Gemini
When you can't find an existing tool for a specific editing or processing job, describe exactly what you need to [your AI assistant] (e.g. a cropping zone, specific corrections, edge snapping) and ask it to build a single-page, in-browser tool for it. Iterate on the first draft rather than expecting it correct on the first try.
A starting prompt written by Agents in the Wild. Fill in the [brackets], and check what the agent does before it sends, pays for or deletes anything.
Tired of typing new Date().toISOString() and doing timezone math by hand, they built a small tool with Gemini, which added obscure date formats and random historical events along the way.
To keep up with industry and tech news, the writer built a free-tier Google Cloud Run job pulling Google News RSS feeds. Gemini picks which articles match their criteria and triggers a summary email.
Escape the algorithm's repeats. ChatGPT and Gemini helped a former web designer build a generator that pulls 1 or 2 lesser-known songs from each of their top artists, in a couple of evenings.
Six AI workers, two each on Claude, Codex and Gemini, were set loose to build a SQLite-like engine in Rust. It passed 64 queries of a 1,000+ query suite, with slop like redundant buffer copies and no concurrency.
Amazon support couldn't figure out the URL for an account question and wanted to escalate with a phone call instead. Gemini gave the URL and explained what was needed in one reply. ChatGPT got it right too.
A T-shirt like the Asket tee at its old $50 price, and shorts to match an old pair. Grok invented a few review claims on the way, but both picks turned out right.