Weekly Coding • 2025-07-28

Hello everyone. Time for a mini report.

Last week, I spent roughly three hours coding my personal website https://hinst.github.io/ and made a small optimization: image loading is now direct. Previously, images were packed in Base64 and JSON , which was a "legacy" from an older version. After the optimization, images load faster and are less prone to errors. The issues arose because some browsers struggled to properly handle links in format data:image/png;base64 .

Two hours were dedicated to code changes, and another hour was spent figuring out why my Docker builds broke. It seems a recent Docker update required specifying --platform at the command-line level, whereas previously setting it in a Dockerfile's FROM instruction was sufficient.

I installed and tested the new Stable Diffusion model called Flux Kontext , which allows you to alter an image using text. The model works remarkably well... but there are still kinks: sometimes the colors turn out too bright. For instance, if I input "change shirt color to yellow," the output is as if the shirt had been splashed with printer ink. Another issue is that the model sometimes doesn't understand the request and leaves the image unchanged. However, it can't notify you of this immediately; you have to wait for the entire process to complete to see the identical image.

Interestingly, while the article on Flux Kontext states a minimum requirement of 20 GB of video memory, in practice, only around 10-12 GB is needed.

I'd like to start working on a new hand-drawn illustration but when I'll find the time for it remains to be seen 😤

P.S. SmartProgress unexpectedly removed my old profile icon🗑️ I had to reinstall a new one.