This morning someone asked in our WhatsApp group if there was any off-the-shelf software that could detect a face in an image and crop it to a fixed size while keeping the face centered.
Anyone seen anything off the shelf for this with AI. I need software that can identify faces in an image and crop the image to a specific size ensuring the face stays in the middle of the frame.

I pasted the request verbatim into Claude Code and let it generate a solution with full local execution enabled (filesystem + shell access).

~7 minutes later I had a small CLI binary called facecrop that:
  • detects the largest face in an image
  • recenters and crops to a target resolution
  • works on a directory of images

It worked perfectly! đźŽ‰

Under the hood it’s just using Apple’s Vision framework for face detection. There is no custom ML, or model training. The interesting part wasn’t the algorithm, but how quickly it went from a vague requirement to a runnable tool that correctly wrapped a native API.

This isn’t a claim that LLMs replace careful engineering, but for small, well-scoped utilities they can generate useful primitives in a single shot, producing a runnable artefact that correctly wires system frameworks with minimal human intervention.
Sometimes Claude Code + Dangerous Mode is all you need.

Code here if anyone wants to inspect or improve it: https://files.littlebird.com.au/facecrop.zip

Curious:
  • Is there an existing tool I should’ve used instead?
  • How comfortable would you be letting an LLM generate code against system frameworks like Vision?