Instagram blocks automation, yet a Hacker News post shows how to bypass its fragile DOM with pure computer vision. Instead of scraping random class names that change on every deployment, the author captures a screenshot, locates the heart icon visually, moves the cursor, and clicks, and can be scripted in Python. This pixel‑based method works on any rendered interface, from web pages to native apps.

To avoid scanning the whole screen, the script first finds two invariant UI landmarks: the triple‑dot menu at the top‑right and the bottom action bar. Their positions define a narrow vertical strip where hearts always appear. Within this crop, a sliding‑window template match locates heart candidates, then a consensus‑X filter discards outliers, leaving reliable click coordinates, and runs in milliseconds.

The account was banned within days despite adding Bezier cursor paths, random delays, and idle fidgeting. Instagram’s bot‑detection team evidently flags any pattern that mimics human clicks. The author released the code as Oculix, proving that a camera‑pointed screen can be controlled solely by visual cues. The experiment highlights both the power and the risk of vision‑only automation for developers experimenting with UI bots.