Umbra v0.3.0

Five tools have been rebuilt this release, and the Bucket has been split into three.

Image to Text now works like Live Text. Your image stays on screen and the recognised text sits selectable directly on top of it, instead of being dumped into a separate box. Regions the engine wasn't sure about are marked on the image rather than reduced to a confidence score, and Cmd-F searches within them. Sideways photos work now — orientation was previously ignored, so a rotated phone photo full of text would come back as "no text found". Text also comes back in reading order rather than detection order, and detection runs at a higher resolution than before.

Image to Text, PDF and Images are three separate tools in the sidebar now. They were only ever grouped under "Bucket" because that is where the scaffolding put them.

Base64 is a single live view. Text converts as you type, data URIs are built, decoded and previewed in place, and binary output is reported honestly instead of guessed at. Encoded output can be line-wrapped.

UUID keeps v4 and v7 and adds format toggles for case, braces and hyphens, a short explanation of what each version is for, and bulk generation you can save straight to a file.

Hash lets you choose the algorithms instead of always returning a fixed set — SHA-256, SHA-512, SHA3-256, SHA3-512, MD5 and SHA-1. It hashes as you type, in hex or Base64, and a Verify panel checks a digest you paste against every selected algorithm at once. MD5 and SHA-1 are marked as weak where you would actually see them.

JWT decodes as you type. Header and payload each have their own copy button, time claims show both the relative and the absolute time, and a live clock re-evaluates expiry while you are looking at it. A token that is not signed is called out.

Cron replaces the free-text "describe the schedule in English" box with a guided builder that works identically in English and French. The cron-to-English direction stays and now shows a field-by-field breakdown alongside the sentence.