v0.9.3「东西存哪儿、出多大,你说了算」/ "You decide where things live and how big they get"
来自 #3 的用户建议:新增设置页(页头「设置」进入)。核心是把几件「藏在代码里、用户找不到」的事摆到台面上——尤其是那个一直没人知道存在哪儿的 1GB 语音模型。
🗂️ 模型存放位置:看得见,也搬得走
用户原话:「模型存放位置,对于我这种比较想知道他存放在哪里的人来说比较有用。我确实也找过,但没有找到,因为他文件比较大,大文件我都会比较关注。」
- 完整路径直接写在设置里,一键「打开文件夹」
- 九个模型逐条列出:用途(语音识别·快/准/方言、标点断句、多人对谈分离、人脸跟随、表情峰值、镜头切点)、模型 id、已装的实际占用、未装的预计下载体积,外加合计占用
- 可整体搬到别的盘:同盘秒搬;跨盘走「先完整复制 → 校验字节 → 才删原件」。搬家失败时原目录分毫不动,并把真实错误显示出来——模型重下一次要一小时,搬丢了比搬不动严重得多。目标目录非空、或选到当前目录的子目录,会直接拒绝
- 搬完之后,后续下载、CLI 与 MCP 也都认新位置(桌面/命令行共用一份配置)
🎚️ 导出画质三档
- 高画质(默认,与此前成片完全一致)/ 标准 / 省空间
- 实测同一素材:高画质 4319 KB → 标准 2762 KB → 省空间 1460 KB(小 66%),三档成片时长与剪辑内容完全相同——画质档只动清晰度和体积,绝不改变切点、字幕或任何剪辑决策
💬 默认字幕样式 + 导出位置
- 默认字幕样式现在可以直接选(六种样式)。其实它一直会记住你上次的选择,只是从没在界面上露过面
- 导出位置也一并收进设置页(出片条上的显示与「更改」照旧保留)
⬆️ 升级方式
到 Releases 下载新安装包覆盖安装即可。已下载的模型、词表、偏好原位保留,升级后不会重新下载任何东西。
A user suggestion from #3: a new Settings page (via "Settings" in the header). It surfaces the things that were previously buried in code where nobody could find them — above all the 1GB speech model whose location was never shown anywhere.
🗂️ Model storage: visible, and movable
In the reporter's words: "I'd really like to know where the models are stored. I did look for them, but couldn't find them — they're large files, and I pay attention to large files."
- The full path is right there in Settings, with one-click "open folder"
- All nine models itemized: purpose (speech fast/accurate/dialects, punctuation, speaker separation, face tracking, emotion peaks, shot detection), model id, actual disk usage when installed, expected download size when not — plus a running total
- Move them to another drive: instant on the same disk; across drives it copies in full, verifies the byte count, and only then removes the original. A failed move leaves the original untouched and shows the real error — re-downloading a model takes an hour, so losing one in transit is far worse than not moving it. Non-empty targets, or a target inside the current folder, are refused outright
- After a move, subsequent downloads plus the CLI and MCP server all use the new location (desktop and command line share one config)
🎚️ Three export quality tiers
- High (default, byte-for-byte the same encode as before) / Standard / Compact
- Measured on identical source: High 4319 KB → Standard 2762 KB → Compact 1460 KB (66% smaller), with identical duration and identical edits across all three — quality tiers touch sharpness and size only, never cut points, captions, or any editing decision
💬 Default caption style + export location
- Default caption style is now directly selectable (six styles). It always remembered your last choice — it just never had a visible home
- Export location moved in here too (the display and "Change" on the export bar stay where they were)
⬆️ Upgrading
Download the new installer from Releases and install over the top. Downloaded models, glossaries and preferences stay exactly where they are — nothing re-downloads.