Changelog

Notable changes to MagicLens, newest first. A build is identified by the version in package.json; the date is the day it was cut. Every tagged build also has its packages on the GitHub releases page.

0.2.2 — 17 October 2026

Changed

• Firefox’s manifest now declares authenticationInfo alongside websiteContent in its data collection permissions, and the options page requests both when a cloud AI provider is chosen. The install consent now matches what the extension actually sends; AMO requires the declaration for new listings.

0.2.1 — 17 October 2026

Changed

• The 128px extension icon now follows the Chrome Web Store layout rule: a 96×96 icon centered in a 128×128 PNG with 16px of transparent padding on every side. It still reads on light and dark backgrounds, and the other icon sizes are unchanged.

0.2.0 — 17 October 2026

Added

• Links in the recognized text. URLs, www. hosts, bare domains (example.com/path), email addresses and IPv4 addresses now become real links laid over the image. Hovering highlights the clickable words and previews the destination; clicking opens it in a new tab, and mailto: hands off to the mail application. Trailing sentence punctuation and unmatched brackets stay outside the link, and file names such as main.ts, setup.py, build.sh or README.md are not mistaken for host names.
• The panel reports how many links a read found next to the line count.
• A changelog page on the site, /changelog/, and this section on the landing page.

0.1.0 — 10 October 2026

First packaged build.

Added

• In-place text layer over images: every word is positioned on its baseline and stretched to the width of the original glyphs, so it can be selected and copied with clean spacing.
• Targets: <img> (including srcset, object-fit, object-position, padding, borders and rounded corners), <video> frames, <canvas>, inline SVG, CSS background images, or any element picked on the page. Elements the engine cannot download fall back to a screenshot.
• Local OCR with Tesseract (30 languages, English built in) and PaddleOCR PP-OCRv5, both running inside the browser with no network access.
• Optional AI: Hybrid mode corrects local OCR text while keeping exact word boxes, or reads the image alone. Translate in place, explain, summarize, or ask questions about the image. OpenAI, Claude, Gemini, OpenRouter, Groq, Mistral, xAI, Together, and Ollama / LM Studio / any OpenAI-compatible server.
• Triggers: hover button, right-click menu, Alt+Shift+L for the main image, Alt+Shift+K to pick an element.
• Processing effects that animate only transform and opacity, so the page stays responsive while a read runs.
• Downloaded updates are applied as soon as no read is running.
• Privacy policy and terms pages, a store kit, and the landing page with a live demo.