Feat/UI polish and english

.github/workflows/build-launcher.yml

@@ -1,14 +1,16 @@
 name: build-launcher
 
-# 一键启动器（tools/launcher.py → AnimaLoraStudio.exe）的构建。
+# Build for the one-click launcher (tools/launcher.py -> AnimaLoraStudio.exe).
 #
-# PyInstaller 产出的是**宿主平台**的可执行文件，没法交叉编译 —— Windows 的 exe
-# 只能在 windows runner 上出。Linux 的那份顺带一起出，方便本机 Linux 用户。
+# PyInstaller produces an executable for the **host platform** — it can't cross-compile,
+# so the Windows exe can only be built on a windows runner. The Linux one is built
+# alongside it for the convenience of local Linux users.
 #
-# 触发：
-#   - 改到启动器源码的 PR / push → 只构建，作为「exe 还能编出来」的守门
-#   - 打 tag（v*）→ 构建并把 exe 传成 release 附件
-#   - 手动 workflow_dispatch
+# Triggers:
+#   - PR / push touching the launcher source -> build only, as a gate that "the exe
+#     still builds"
+#   - tag push (v*) -> build and attach the exe as a release asset
+#   - manual workflow_dispatch
 on:
   pull_request:
     paths:
@@ -42,8 +44,9 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
         with:
-          # 启动器只用标准库，但**它自己**得能在用户机器上跑。用 3.10 构建
-          # 让产物的 glibc / API 需求落在我们声明支持的最低版本上。
+          # The launcher itself only uses the standard library, but **it** still has to
+          # run on the user's machine. Building with 3.10 keeps the produced binary's
+          # glibc/API requirements at the minimum version we claim to support.
           python-version: "3.10"
 
       - name: Install PyInstaller
@@ -52,8 +55,9 @@ jobs:
       - name: Build
         run: python tools/build_launcher.py
 
-      # 冒烟：产物必须能跑起来并认出自己在仓库里（--check 不装任何东西）。
-      # 这一步能挡住"打包成功但一运行就缺模块"这类只有跑一次才暴露的问题。
+      # Smoke test: the built binary must actually run and recognize itself inside the
+      # repo (--check installs nothing). This step catches "packaged fine but missing a
+      # module at runtime" issues that only surface when you actually run it once.
       - name: Smoke test
         run: ${{ matrix.binary }} --check
 

.github/workflows/publish-launcher.yml

@@ -1,22 +1,26 @@
 name: publish-launcher
 
-# 构建 Windows 版 `AnimaLoraStudio.exe` 并**提交回仓库**。
+# Builds the Windows `AnimaLoraStudio.exe` and **commits it back to the repo**.
 #
-# 为什么 exe 进 git（一般来说二进制不该进）：本 fork 的目标用户是「clone 完
-# 想双击就用」的本机用户。让他们先去 Releases 翻附件、或先装 PyInstaller 自己
-# 编，等于把要省的那一步又加了回来。exe 只有 7MB 且极少变（只在
-# tools/launcher.py 改动时需要重出），代价可以接受。
+# Why the exe goes into git (binaries usually shouldn't): this fork's target users are
+# people who "clone it and want to double-click it right away." Making them first dig
+# through Releases for the attachment, or install PyInstaller and build it themselves,
+# just adds back the step we're trying to save them. The exe is only ~7MB and rarely
+# changes (only needs rebuilding when tools/launcher.py changes), so the cost is
+# acceptable.
 #
-# 为什么必须走 CI 而不是本地编：PyInstaller 产出的是**宿主平台**的可执行文件，
-# 不能交叉编译 —— Windows 的 exe 只能在 Windows 上出。维护者用 Linux/macOS 时
-# 这个 workflow 是唯一的出路。
+# Why this has to go through CI instead of a local build: PyInstaller produces an
+# executable for the **host platform** — it can't cross-compile, so the Windows exe can
+# only be built on Windows. This workflow is the only path for maintainers on
+# Linux/macOS.
 #
-# 手动触发（Actions → publish-launcher → Run workflow）。刻意**不**挂到 push：
-# PyInstaller 的输出不是逐字节可复现的，每次 push 都重出会让仓库无谓地长胖，
-# 而且会和用户的 commit 抢同一分支。
+# Manually triggered (Actions -> publish-launcher -> Run workflow). Deliberately **not**
+# hooked to push: PyInstaller's output isn't byte-for-byte reproducible, so rebuilding on
+# every push would bloat the repo for no reason and would race with users' own commits on
+# the same branch.
 #
-# 常规的「exe 还编得出来吗」守门在 build-launcher.yml（每次改启动器源码时跑，
-# 不提交任何东西）。
+# The routine "does the exe still build" gate lives in build-launcher.yml (runs on every
+# change to the launcher source, commits nothing).
 on:
   workflow_dispatch:
     inputs:
@@ -38,7 +42,8 @@ jobs:
 
       - uses: actions/setup-python@v5
         with:
-          # 用支持的最低版本构建，产物的运行时要求才落在我们声明的下限上。
+          # Build with the lowest supported version so the binary's runtime requirements
+          # stay at the floor we claim to support.
           python-version: "3.10"
 
       - name: Install PyInstaller
@@ -47,9 +52,10 @@ jobs:
       - name: Build
         run: python tools/build_launcher.py
 
-      # 冒烟：产物必须真的能跑起来并认出自己在仓库里。--check 不装任何东西。
-      # 提交一个编得出来但一运行就报缺模块的 exe，比不提交更糟 —— 用户拿到的
-      # 是「双击没反应」，而仓库看起来一切正常。
+      # Smoke test: the built binary must actually run and recognize itself inside the
+      # repo. --check installs nothing. Committing an exe that builds fine but errors out
+      # on a missing module at runtime is worse than not committing at all — the user
+      # gets a "double-click does nothing" experience while the repo looks perfectly fine.
       - name: Smoke test
         run: |
           Copy-Item dist\AnimaLoraStudio.exe .\AnimaLoraStudio.exe -Force
@@ -61,9 +67,10 @@ jobs:
           git config user.name "github-actions[bot]"
           git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
           git add -f AnimaLoraStudio.exe
-          # PyInstaller 的输出每次都有字节差异（时间戳等），所以这里不能靠
-          # 「有 diff 就提交」判断有没有意义 —— 用 --quiet 只是为了跳过
-          # 「一个文件都没变」这种（理论上不会出现的）空提交。
+          # PyInstaller's output differs byte-for-byte every time (timestamps etc.), so
+          # "there's a diff" can't be used to decide whether committing is meaningful —
+          # --quiet here is only to skip the (theoretically impossible) case where
+          # literally nothing changed, to avoid an empty commit.
           if git diff --cached --quiet; then
             echo "nothing to commit"
             exit 0

.github/workflows/version-check.yml

@@ -1,12 +1,13 @@
 name: version-check
 
-# 跨文件 version drift gate。v0.8.1 release 漏 bump package-lock.json，结果
-# npm install 启动期把 lockfile 改脏 → ZIP 用户 self-update preflight 卡 dirty
-# tree。这个 workflow 防同类回归。
+# Cross-file version drift gate. The v0.8.1 release forgot to bump package-lock.json,
+# which meant npm install dirtied the lockfile at startup -> ZIP users' self-update
+# preflight got stuck on a dirty tree. This workflow guards against the same regression.
 #
-# 跑两件事：
-#   1. release_notes.yaml schema 校验（validate）
-#   2. __init__.py / package.json / package-lock.json 四处 version 一致（verify-versions）
+# Runs two checks:
+#   1. release_notes.yaml schema validation (validate)
+#   2. version consistency across __init__.py / package.json / package-lock.json
+#      (verify-versions)
 on:
   pull_request:
   push:

.gitignore

@@ -32,11 +32,12 @@ studio/web/node_modules/
 studio/web/dist/
 studio/web/.vite/
 
-# 顶层意外创建的 node_modules（vitest / 某些工具可能在 repo root 拉一份）
+# Stray top-level node_modules (vitest / some tools may pull one at the repo root)
 /node_modules/
 
-# Studio 走 preset 池（studio_data/presets/，已被 studio_data/ 总规则覆盖），
-# 旧的 config/ 目录已删，不再需要 user yaml 排除规则。
+# Studio now uses the preset pool (studio_data/presets/, already covered by the
+# studio_data/ blanket rule); the old config/ directory is gone, so the user
+# yaml exclusion rule is no longer needed.
 
 # OS / editor
 .DS_Store
@@ -72,37 +73,43 @@ dataset/
 nul
 # Root-level AI agent files — gitignored so contributors can put personal
 # config / @docs/AGENTS.md import here without conflicting with the repo.
-# Project 公约 lives at docs/AGENTS.md (NOT ignored — note the leading /).
+# The project conventions live at docs/AGENTS.md (NOT ignored — note the leading /).
 /AGENTS.md
 /CLAUDE.md
 
-# 本地解压的设计稿原始目录（已整理进 mockup/）
+# Local extracted design-mockup source directory (already organized into mockup/)
 1/
 
-# 工具生成的临时目录
+# Temporary directory generated by tooling
 .spec-workflow/
 
-# Self-update 进程间通信用的瞬时标志（server 写 → cli.py / wrapper 读后立即删）
+# Ephemeral flag used for self-update inter-process communication
+# (server writes it -> cli.py / wrapper reads it and deletes it immediately)
 tmp/
 
-# 文档本地草稿（个人 TODO / 未定稿设计 / 临时笔记）
+# Local documentation drafts (personal TODOs / unfinished designs / scratch notes)
 docs/_local/
 
 # Local UI snapshots before redesign (not for commit)
 _ui_backup/
 
-# 一键启动器的产物（tools/build_launcher.py 编出来的）。
+# Build output of the one-click launcher (produced by tools/build_launcher.py).
 #
-# `AnimaLoraStudio.exe`（Windows）**故意不在此列** —— 它随仓库一起提交，
-# 这样用户 `git clone` 下来就能直接双击，不用先去 Releases 找附件。
-# 由 .github/workflows/publish-launcher.yml 在 windows runner 上构建并回提。
+# `AnimaLoraStudio.exe` (Windows) is **deliberately not** in this list — it's
+# committed alongside the repo so that after `git clone` a user can just
+# double-click it, without first hunting for the attachment on Releases.
+# It's built and committed back by .github/workflows/publish-launcher.yml on
+# a windows runner.
 #
-# 非 Windows 的产物仍然忽略：Linux/macOS 用户直接 `./studio.sh` 就行，
-# 不需要引导程序，提交进去只是给每个人多下几 MB。
+# The non-Windows build output is still ignored: Linux/macOS users just run
+# `./studio.sh` directly and don't need a bootstrapper, so committing it would
+# only add a few extra MB for everyone to download.
 /AnimaLoraStudio
 /dist/
 
-# 第三方库缓存（pip / HuggingFace / torch / npm / ...）—— 见
-# studio/infrastructure/local_cache.py：默认收进项目文件夹，免得几个 GB
-# 悄悄堆在系统盘上。整个目录随时可删，删了只是下次重下。
+# Third-party library cache (pip / HuggingFace / torch / npm / ...) — see
+# studio/infrastructure/local_cache.py: it's collected into the project
+# folder by default so a few GB don't quietly pile up on the system drive.
+# The whole directory can be deleted at any time; deleting it just means
+# it gets re-downloaded next time.
 /.cache/