Documentation images fail in two directions at once. Too heavy and every page of a docs site drags, because a help article with eight full-width retina screenshots can easily ship 6 MB of images for two paragraphs of text. Too aggressive and the screenshot stops doing its job, since compressed UI text turns into grey mush and the reader cannot tell which button the article is pointing at. The distinction that resolves it is what the image contains: photographic content and gradients survive heavy compression, while crisp UI text, thin borders and syntax-highlighted code do not. That means a screenshot of a settings panel should be resized to the width it renders at and compressed with a mild setting, whereas a photo of a physical device can take much more. A second practical rule is to favour WebP or PNG for screenshots with flat UI and text and JPEG for photographic content, because JPEG artefacts land exactly on the letterforms. Skwish compresses and converts PNG, JPG and WebP in the browser with a per-file before/after readout and a ZIP download for a batch, so you can compare settings on the actual screenshot instead of trusting a default. Nothing is uploaded.