v1.5.5

Summaries now contain only characters you could type on a keyboard. Em dashes, curly quotes and the ellipsis character are gone: the default prompt forbids them in all 11 languages, and a cleanup pass (sanitizeTypography) normalizes every Gemini reply regardless of prompt.

Prompt migration: PROMPT_GENERATION 4; stored prompts that exactly match a 1.5.2-1.5.4 default are migrated to the new default. Customized prompts are never touched.

Also fixed: the "improved default prompt" notice appeared for prompts that were customized after the update. Saving the settings now marks the notice as seen.

v1.5.4

Completes the Android repair started in v1.5.3 — that release fixed comment injection but transcript fetching still failed on a real device; it never shipped to AMO.

Transcript fetching on Firefox for Android

• The player only exposes caption track URLs (with the required proof-of-origin token) once playback has started — and mobile blocks autoplay, so the transcript engine came up empty on devices. The add-on now briefly starts playback itself (the button tap counts as the required user gesture) and pauses it again afterwards; a video that is already playing is left alone.
• If no transcript is found on mobile, the error now says to play the video for a moment — instead of pointing to a transcript panel that does not exist on m.youtube.com.

Comment field

• After injection the comment box grows to 40% of the screen height. The default is 84px tall and its resize handle is nearly impossible to grab on a touchscreen.

Full end-to-end run verified on a physical device (Galaxy A51, Firefox Nightly via web-ext).

v1.5.3

Firefox for Android: comment injection fixed

• v1.5.2 opened the comments panel but failed to reach the input field. The visible composer on m.youtube.com is a button that only turns into a real text field when clicked; the field carries no aria-label, so the old selector never matched. Both steps are fixed and were verified on a physical device (Galaxy A51).
• Text is now written through the native textarea setter, so YouTube's framework registers the input and enables the submit button.

Never lose a summary again

• If comment injection fails for any reason, the finished summary is copied to the clipboard and the error message says so — paste it manually. All 11 languages.

v1.5.2

Default prompt improvements

• New "numbers and evidence" section: summaries now carry concrete numbers, data and study references from the video — and state it when a central claim comes without them. All 11 languages; abridged translations brought to full prompt depth.
• Fix: saving the settings permanently froze the then-current default prompt, so prompt improvements never reached existing installations. Stored prompts identical to an old default are migrated automatically; customized prompts stay untouched and get a dismissible notice in the settings.

Firefox for Android

• The Distiller button no longer appears multiple times (concurrent injection runs are now locked out).
• Comment posting no longer fails with "comments section not found": the add-on opens the comments panel itself. Selectors verified against the real m.youtube.com DOM; the logged-in comment composer is still pending a test on a physical device.

v1.5.1 – Model dropdown & smarter defaults

What's new

• Gemini model dropdown — the free-text model field on the settings page is now a dropdown with Google's three alias IDs: Gemini Flash-Lite (gemini-flash-lite-latest, the new default), Gemini Flash (gemini-flash-latest) and Gemini Pro (gemini-pro-latest). Custom model ID… still accepts any model ID. The section now sits directly below the API key.
• New default: gemini-flash-lite-latest — the alias always points at the newest Flash-Lite release, so the default no longer breaks when Google retires a model version. Flash-Lite also has the most generous free-tier quota of the three; gemini-3.5-flash allows only 20 requests per day on the free tier.
• Actionable error messages — when Gemini answers 404 (model retired), 429 (daily quota used up) or 503 (overloaded), the error now tells you the fix: pick a different model in the add-on settings. Localized in all 11 languages.

Existing users keep their stored model; the dropdown shows it under Custom model ID…. Switching to Flash-Lite is one click.

v1.5.0 – Silent player-based transcripts, description-aware summaries

What's new

New transcript engine. The add-on now reads the caption track URL straight from the YouTube player (including its proof-of-origin token) and fetches the transcript silently — no transcript panel opens, no description needs expanding, no scrolling. The previous methods (internal API, DOM scrape) remain as fallbacks.

Summaries use the video description. Title, channel and full description now come from the player response instead of DOM scraping, so the description reaches Gemini complete and on every layout.

Gemini model fix (from unreleased v1.4.1). Google retired gemini-2.5-flash (the API returns 404 for it). The default model is now gemini-3.5-flash, and the model ID is configurable on the add-on settings page — future model retirements no longer require an update. If a model answers 503 "high demand", gemini-3.1-flash-lite is a working fallback.

Groundwork for Firefox on Android. The content script now also runs on m.youtube.com, with mobile injection points and a mobile comment flow. This path is still being tested on real devices — on desktop everything is verified.

UI. The gear tooltip is now "Distiller Settings"; the model field lives on the settings page only.

v1.4.1 – Configurable Gemini model

What's new in v1.4.1

Fix

• Google retired gemini-2.5-flash — the API now returns 404 "no longer available". The default model is gemini-3.5-flash.

New

• The Gemini model is user-configurable: new Gemini Model field in the settings modal on YouTube and on the add-on options page. Any model ID that supports generateContent works (e.g. gemini-flash-latest, gemini-3.1-flash-lite), so the next model retirement needs no add-on update.

i18n

• New settings strings translated in all 11 locales.

Repository

• The v1.4.0 source (mood/style prompt lists, video description included in the Gemini context) is now in the repository; it was previously only on AMO.

v1.3.1 – Privacy & UX improvements

What's new in v1.3.1

Privacy

• Added telemetry opt-out checkbox in extension settings
• Anonymous usage statistics are enabled by default and can be disabled in the extension settings

UX

• API key field in the settings modal is now hidden once a valid key is stored — less clutter
• If the API key is rejected by Gemini (401/403), the key field reappears automatically with a warning
• Quota exceeded: button now shows a live countdown instead of just an error message

Compatibility

• Raised minimum Firefox version to 140 (required for data_collection_permissions)
• Raised minimum Firefox for Android version to 142

Internals

• Settings modal rebuilt using DOM methods instead of innerHTML (AMO linter compliance)
• Footer text now matches the chosen response language, not the UI language

v1.3.0 – Initial Release

First public release of YouTube Transcript Distiller.

• AI-powered YouTube transcript summarization via Google Gemini
• Posts summary directly into the YouTube comment field
• 11 languages supported
• Customizable prompt
• Usage statistics via marsgasse.com