[harran]

Hi HN,

I wanted to share this because the results genuinely surprised me.

The Backstory: I originally wanted to build this background remover on the server side. However, I realized my server specs weren't powerful enough to handle the model inference for multiple users. I almost gave up, but decided to try shifting the entire process to the client's browser as a "last resort."

I had serious doubts about whether it would work or if the quality would be usable. To my surprise, the browser-native performance was excellent and the edge-detection quality was high.

Technical Details:

Inference: Runs 100% in your browser using the user's local GPU/CPU.

Privacy: No images are uploaded to any server. What started as a hardware limitation became a privacy feature.

Performance: Once the model is cached, the tool provides results instantly and works offline.

This tool is part of my larger project, AI Shortcut Lab, where I provide ready lists of AI tools and automated workflows.

I’d love to know: How does the processing speed feel on your machine?