New Features:
-
Added 3 new Wyzie subtitle sources (Kitsunekko, Jimaku, YIFY): The Wyzie API added 3 new subtitle providers — Kitsunekko (anime subtitles), Jimaku (Japanese anime subtitles), and YIFY (movie subtitles) — bringing the total from 6 to 9 available sources. Added the new providers across the full config stack: backend
allSourcesarray inwyzieSubs.js, config defaults and migration/backwards-compatibility fallback inconfig.js, form population and collection logic inconfig.js, new checkboxes in the main settings page (main.html) and Quick Setup wizard (quick-setup.html), and state defaults/reading/restore inquick-setup.js. New sources default to disabled for new users; existing Wyzie users who had sources enabled get the new providers auto-enabled via the migration fallback. All 3 providers use the same Wyzie API response format — no backend logic changes were needed. -
"ASS/SSA Passthrough" setting exposed to all users: The
convertAssToVtttoggle was previously hidden. It is now always visible in the Settings section for all users, since ASS passthrough is production-ready with the new translation pipeline. Removeddisplay:nonefrommain.html, and removed devMode gating inconfig.js(toggleConvertAssToVttGroup()function and config-load visibility). -
ASS/SSA subtitles: ASS/SSA subtitles can now be fetched and translated while preserving their original format — styles, fonts, colors, positioning tags (
{\an8},{\pos()},{\b1}, etc.), and document structure ([Script Info],[V4+ Styles],[Events]). Previously, all ASS/SSA subtitles were unconditionally converted to SRT before translation. The new pipeline uses a wrapper approach: (1) parse ASS dialogue entries, separating override tags from translatable text, (2) build a temporary SRT from the extracted clean text, (3) translate via the existingTranslationEngine(unchanged), (4) re-inject translated text back into the original ASS structure with tags restored at proportionally-mapped positions. NewassTranslationHelper.jsutility (~280 lines) withparseASSForTranslation(),buildSRTFromASSDialogue(), andreassembleASS()functions. Modifiedsubtitles.jsto passskipAssConversionto all 14 provider download calls in both pre-flight and main translation paths, detect ASS format via content inspection ([Script Info]+[V4+ Styles]), conditionally bypassensureSRTForTranslation(), and reassemble ASS after translation completes. Added_asscache key variant incacheKeys.jsto prevent format collisions between ASS and SRT translations in the permanent (shared) cache. Updatedindex.jstranslation route to detect payload format viadetectSubtitlePayloadFormat()and serve correctContent-Type(text/x-ssa) and file extension (.ass/.ssa) in all 3 response paths (bypass cache, permanent cache, main response). Learn Mode compatibility preserved: translated ASS content is converted to SRT viaensureSRTForTranslation()beforesrtPairToWebVTT()pairing. -
ASS/SSA format-aware subtitle file selection in archives:
findSubtitleFileinarchiveExtractor.jsnow respects the ASS passthrough setting when selecting subtitle files from archives. WhenskipAssConversionis enabled, the function prioritizes ASS/SSA files over SRT files (using aprimaryFiles/secondaryFilesabstraction) for both season-pack episode matching (anime + TV patterns) and single-file selection. Previously, the function always searched SRT files first regardless of passthrough setting — ASS files were only found via the all-entries fallback, which worked by accident but would select an SRT file over an ASS file when both existed in the same archive. -
Automatic URL extension enforcement for ASS passthrough: When ASS passthrough is enabled (
convertAssToVtt: false), the addon now automatically forcesurlExtensionTestto'none'(no file extension in subtitle URLs). Previously, subtitle URLs were built with a.srtextension by default, causing an extension/payload mismatch when the actual content was ASS/SSA — Stremio would request a.srtURL but receive ASS content. The enforcement is applied in three places: backend config normalization (src/utils/config.js), frontendcollectConfig()(public/config.js), and the existing URL builder in the subtitle handler (subtitles.js:2860-2862) already handles the'none'value correctly by setting bothurlExtensionandtranslationUrlExtensionto empty strings. The serving route atindex.jsdetects the payload format dynamically and sets the correctContent-Typeheader (text/x-ssa; charset=utf-8) regardless of URL extension.
PS: Partial delivery during translation remains SRT (from the translation engine), with final ASS reassembly applied only to the completed output. Activated by unchecking "Convert ASS/SSA to VTT" in settings — when disabled and the source is ASS/SSA, the translation pipeline automatically uses the new ASS-aware path.
Bug Fixes:
-
Fixed phantom "New stream detected" notifications for streams watched hours/days ago: The stream activity LRU cache had
updateAgeOnGet: true, which reset the 6-hour TTL on every SSE connect, heartbeat, or page refresh — old entries essentially never expired as long as a toolbox tab was open. When opening a new toolbox/SMDB page,subscribe()would immediately send the cached entry as anepisodeevent, triggering a toast for a stream that was watched days ago. RemovedupdateAgeOnGetso entries actually expire after 6 hours. Added afirstSeenAttimestamp to stream activity entries (preserved across heartbeat/enrichment updates) so clients can distinguish "stream started 3 hours ago" from "entry refreshed 10 seconds ago." Added a 30-minute age check (configurable viaSTREAM_ACTIVITY_SNAPSHOT_MAX_AGE_MS) insubscribe()that suppresses stale cached entries from being sent to new SSE subscribers. Added a matching client-side staleness guard inquickNav.jsthat checksfirstSeenAtagainstwindow.__SUBMAKER_PAGE_LOAD_TIMEbefore showing a toast, catching stale data from the polling fallback path that bypasses the server-side SSE check. -
Fixed ghost linked streams with non-matching hashes on the same title: When Stremio sent multiple subtitle requests for the same
videoIdwith slightly different filenames (different stream sources, URL encoding differences, different addons),deriveVideoHash()produced different MD5 hashes, and the merge logic inrecordStreamActivity()allowed the hash to flip — triggering a "new stream" notification for what was really the same stream with a mangled hash. Strengthened the hash drift guard: when the previous entry already has an authoritative filename+hash pair for the samevideoId, the incoming hash is rejected regardless of whether the incoming payload has a filename or not. Additionally, notifications now only fire when thevideoIdactually changes — field enrichment for the samevideoId(e.g., empty hash → real hash) is stored silently without broadcasting. The SMDB page'shandleStreamLinked()also gained a duplicate-videoId guard and usesfirstSeenAtfor its staleness check. The_handleRemoteStreamEventRedis pub/sub handler now applies the same change-detection guards instead of blindly setting LRU entries and notifying listeners. -
Fixed Google Translate failing on larger subtitle files: The
GoogleTranslateProvider.callTranslate()method usedaxios.get()withparams, placing the entire joined subtitle text into the URL query string. Google's endpoint returns HTTP 400 (Bad Request) when the URL-encoded payload exceeds ~15K characters, which commonly happens with batches of 300+ subtitle entries. SwitchedcallTranslate()from GET to POST withapplication/x-www-form-urlencodedbody, eliminating the URL length ceiling entirely. Additionally, added internal chunking (chunkTexts()+_translateChunked()) that splits large payloads into ~6K-character chunks at entry boundaries when the joined text exceedsMAX_CHARS_PER_REQUEST. Each chunk is translated in a separate sequential API call and results are reassembled in correct order. A graceful fallback (_isPayloadTooLargeError()) detects 400/413/414 status codes and automatically switches from single-request to chunked mode, providing resilience even if Google changes their exact limit. All changes are self-contained ingoogleTranslate.js— no modifications needed to theTranslationEngineor handler code. The fix applies regardless of whether Google Translate is configured as the main or secondary/fallback provider, since all code paths callGoogleTranslateProvider.translateSubtitle().
xSync Extension:
-
OPFS-based stream download for large files (fixes "Array buffer allocation failed"): The xSync extension previously pre-allocated a single
Uint8Array(totalBytes)in RAM when downloading full streams in Complete mode — any file exceeding the browser's ~2-4 GB memory ceiling would crash with "Array buffer allocation failed." Added a newfetchFullStreamBufferOPFS()function that streams download chunks directly to the Origin Private File System (OPFS) viaFileSystemWritableFileStream, avoiding the upfront memory allocation entirely. After download completes, the file is read back as anArrayBufferusing the browser's memory-mapped file I/O (significantly more efficient than pre-allocating a contiguous buffer). AfetchFullStream()dispatcher reads the user's preferred mode and delegates to either the OPFS or legacy RAM path. If OPFS is unavailable (e.g. older browser), it silently falls back to RAM. All 5 call sites across embedded-subs extraction (extractEmbeddedSubtitles), autosync audio extraction (extractAudioFromStream— 3 sites), and Assembly AI autosubs (handleAssemblyAutoSubRequest) now route through the dispatcher. Return shape is unchanged ({ buffer, totalBytes, contentType }), so all downstream consumers (FFmpeg offscreen page, subtitle extraction, audio decoding) work without modification. -
Stream Buffer Mode setting: Added a new "Stream Buffer Mode" dropdown to the xSync options page (General → Global Behaviour section) with two options: Disk (default) — uses OPFS, safe for large files; RAM — legacy in-memory path, faster but may crash on files >2 GB. The setting is stored in
chrome.storage.syncunderxsync-settingsand read by the background service worker before each full-stream download. Default is Disk for safety. -
Fixed FFmpeg subtitle extraction silently doing nothing (bare core no-op): The offscreen FFmpeg bare core's
run()function checked formodule.execthenmodule.callMainto invoke FFmpeg, butffmpeg-core.jsexports neither symbol. The fallthrough?: 0silently returned success without ever executing FFmpeg, causing all subtitle extractions via the bare core path to report "No subtitle streams found" regardless of whether the file contained subtitles. Combined withforceBareCore = true(which bypassed the working ffmpeg.wasm wrapper), the offscreen demux path was completely broken. Fixed by settingforceBareCore = falseto restore the ffmpeg.wasm wrapper path, and replaced the silent fallthrough with an explicitthrowso a missing entry point is immediately surfaced as an error. Applied the same safety fix tobackground.full.js.