1.15.2

• Auto-retry on cuBLAS errors: When a compute type isn't supported by your GPU, the GUI now automatically falls back to a compatible type (float16 → float32)
• Bug Fixes

1.15.1

• Fixed compact console parsing so timestamp lines with hour format (h:mm:ss / hh:mm:ss) are shown correctly, preventing console output progress display from appearing to stop after 59 minutes. (#17)
• Fixed compact console formatting to show Subtitles are written to ... cleanly on its own line block.
• Fixed console auto-scroll behavior. Auto-follows only when the scrollbar is already at the bottom. Scrollbar preserves user position when scrolled to the middle/top.
• Fixed scroll bar forced jump on new line of console output.
• Added explicit horizontal and vertical scrollbar styling for Light, Dark, and AMOLED themes to eliminate
  checkerboard/fallback track artifacts.

1.15.0

• Added Global Settings options for output location (Use source folder) and output naming (Use input filename for outputs)
• Routed output directory resolution per input file and fixed output renaming path logic
• Added yt-dlp 403 guidance with actionable in-app hint
• Replaced native top menu with toolbar menus and adaptive font scaling for 100%/150% text size
• Expanded VAD tab with diarization controls, review prompt/button, and speaker-label post-processing helpers

1.14.1

• Add Verify/Repair Bundle controls and clearer EXE/source conversion guidance in Manage Models.
• Auto‑repair invalid converter bundles and show accurate download/extraction/conversion progress.
• Improve duplicate/partial model recovery for HF and local imports.

1.14.0

• Added HF Transformers model detection and CT2 conversion flow (model.safetensors → model.bin).
• Added on‑demand converter bundle download with EXE support and status display.
• Improved conversion UX: progress/ETA, success modal, and Manage Models refresh.

1.13.0

• Add Manage Models dialog (Hugging Face download, local import, enable/disable, rescan).
• Add Verify Models workflow in Manage Models.
• Add support for adding custom models by Hugging Face repo URL/ID and download all model files automatically.
• Add explicit GitHub API User‑Agent to reduce update check failures.

Converter Bundle

Converter bundle for Faster-Whisper-XXL-GUI.
Includes Python + CTranslate2 + Transformers dependencies.

1.12.0

• Fixed cases where yt‑dlp downloads were completed but the app didn’t detect the output file.
• Prevented duplicate transcriptions when using yt‑dlp.
• Better handling of VAD GPU memory errors (auto fallback to CPU).
• Failure logs are now clearer and privacy‑safe, and include model type.
• Re‑running the same file now overwrites outputs instead of creating extra copies.

1.11.1

• Fixed model directory detection so you can point to either the parent _models folder or a specific model folder.
• Updated large-v3-turbo downloads to use a working CTranslate2 conversion repo.
• Distil model downloads now use the correct faster-distil-whisper-* repo naming.
• Added a tooltip/note clarifying CT2‑only models and Model Directory usage.
• Moved yt‑dlp update debug logs into the logs/ folder
• Improved model download dialog readability in across themes (fixed label contrast and background blocks).

1.11.0

• Added Paths and Overrides tab: consolidated EXE path overrides (Whisper/yt‑dlp/ffmpeg) plus Extra CLI Args.
• Added EXE‑aware yt‑dlp update flow: version check uses the selected EXE and Software Info reflects EXE/manual mode.
• Added Model Directory test table: lists model name, path, and size with a clear status summary.