Releases: amariichi/ScreenshotTranslator2
Releases · amariichi/ScreenshotTranslator2
v7.0.0
Summary
- make Gemma 4 26B-A4B the default model for OCR and translation
- align the Web UI OCR path with the shortcut client for more consistent results
- document safe mmproj coexistence when keeping Gemma 4 E4B alongside the 26B default
Notes
- default projector path models/mmproj-F16.gguf now assumes the 26B-A4B projector
- if you also keep E4B locally, save its projector under a different filename and point LLAMA_MMPROJ at it explicitly
v6.0.1
Summary
- clarify the Qwen3.5 mmproj naming in the README
- avoid documenting a local-only renamed filename as if it were canonical
- note that Qwen3.5 requires its own mmproj and that local renaming is acceptable when filenames would conflict
Scope
- documentation-only patch release
- no runtime behavior changes from v6.0.0
v6.0.0
Summary
- switch the default local model from Qwen3.5 to Gemma 4 E4B
- use mmproj-F16 and --parallel 1 as the new single-user defaults
- add a minimal GitHub Actions CI workflow for Python and shell syntax checks
- update README, UI labels, and app metadata to match the new default model
Compatibility
- Qwen3.5 remains available, but it is no longer the default
- Qwen3.5 requires its own mmproj; do not reuse the Gemma 4 mmproj
- if filenames would conflict locally, you can save the Qwen mmproj under another name and point LLAMA_MMPROJ to that path
Validation
- CI / syntax-smoke passed on PR #1
- local checks: python3 -m compileall app and bash -n start.sh
v5.0.0 release
v5.0.0 Release Notes
Highlights
- デフォルトモデルを Qwen3.5-35B-A3B に切り替え
- Qwen3.5 用 chat template を同梱し、
enable_thinking=falseを既定化 --reasoning-budget 0を既定適用し、OCR/翻訳レスポンスの遅延を抑制
Changes
start.shを更新し、Qwen3.5 /mmproj-F32.ggufをデフォルト起動に統一LLAMA_CHAT_TEMPLATE_FILE/LLAMA_THINK_BUDGETを正式サポートLLAMA_ARG_CHAT_TEMPLATE_FILE/LLAMA_ARG_THINK_BUDGET(および typo 互換)を受け付け- README を Qwen3.5 運用前提に整理し、モデル配布元リンクを追加
Breaking Changes
- 旧
Qwen3-VLへの自動フォールバックを廃止 mmprojはmodels/mmproj-F32.gguf前提- バージョンを 5.0.0 に更新
Notes
- 既存の API エンドポイントと UI 操作フローは維持
- 適用後は
./start.shで従来どおり起動可能
v4.2 release
v4.2 Release Notes
Highlights
- GNOME拡張に TTS Once Mode(1回読み上げ) を追加
- 長文の読み上げを文単位で逐次再生し、体感レイテンシを改善
- 監視モードの重複読み上げ抑制を強化(OCR揺らぎ耐性)
Changes
- 新API: POST /api/v1/ocr_translate_tts_once
- Kokoro TTS の 生成/再生を分離し、先読み生成で文間の沈黙を短縮
- 長文の連続再読を抑えるロジックを追加
Notes
- 既存のモード(Overlay/Monitor)はそのまま
- GNOME拡張は 再コピー+再ログインで反映してください
v4.1.1 release
Bind servers to localhost by default