Fix Bilibili 412 error: inject dm_img risk-control params into yt-dlp wbi/playurl requests

将 README 社区区块中 5 个会过期的微信群二维码，替换为公众号二维码，
关注后回复「交流群」获取最新群码，避免群码失效。与关于页保持一致。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

BillNote_extension/src/background/main.ts

@@ -3,6 +3,7 @@ import type { Settings, TaskRecord } from '~/logic/types'
 import { DEFAULT_SETTINGS, MAX_TASKS, SETTINGS_KEY, TASKS_KEY } from '~/logic/constants'
 import { detectPlatform } from '~/logic/platform'
 import { fetchBilibiliSubtitle } from '~/logic/bilibili-subtitle'
+import { normalizeVideoTitle } from '~/logic/task-display'
 
 // only on dev mode
 if (import.meta.hot) {
@@ -58,6 +59,7 @@ async function upsertTask(record: TaskRecord) {
 
 async function startTask(url: string, title?: string): Promise<{ ok: boolean, taskId?: string, error?: string }> {
   const platform = detectPlatform(url)
+  const displayTitle = normalizeVideoTitle(title)
   if (!platform)
     return { ok: false, error: '当前链接不是支持的视频平台' }
 
@@ -107,7 +109,7 @@ async function startTask(url: string, title?: string): Promise<{ ok: boolean, ta
       message: '已提交',
       createdAt: Date.now(),
       updatedAt: Date.now(),
-      title,
+      title: displayTitle,
     })
     return { ok: true, taskId: body.data.task_id }
   }

BillNote_extension/src/components/MindMap.vue

@@ -1,32 +1,181 @@
 <script setup lang="ts">
-import { onMounted, ref, watch } from 'vue'
+import { nextTick, onMounted, onUnmounted, ref, watch } from 'vue'
 import { Transformer } from 'markmap-lib'
 import { Markmap } from 'markmap-view'
 import { absolutizeMarkdownImages, stripSourceLink } from '~/logic/api'
 
 const props = defineProps<{ markdown: string }>()
 
+const wrapRef = ref<HTMLDivElement | null>(null)
 const svgRef = ref<SVGSVGElement | null>(null)
 let mm: Markmap | null = null
+let resizeObserver: ResizeObserver | null = null
 const transformer = new Transformer()
+const MIN_EXPORT_FONT_PX = 256
+const MIN_EXPORT_WIDTH = 12800
+const MAX_EXPORT_SCALE = 24
+const MAX_CANVAS_SIDE = 32767
 
-function render() {
+function canvasToBlob(canvas: HTMLCanvasElement): Promise<Blob> {
-  if (!svgRef.value)
+  return new Promise((resolve, reject) => {
-    return
+    canvas.toBlob((blob) => {
-  const md = absolutizeMarkdownImages(stripSourceLink(props.markdown || ''))
+      if (blob)
-  const { root } = transformer.transform(md)
+        resolve(blob)
-  if (!mm)
+      else
-    mm = Markmap.create(svgRef.value, undefined, root)
+        reject(new Error('导出思维导图图片失败'))
-  else
+    }, 'image/png')
-    mm.setData(root).then(() => mm?.fit())
+  })
 }
 
-onMounted(render)
+function createSvgElement<K extends keyof SVGElementTagNameMap>(tag: K): SVGElementTagNameMap[K] {
+  return document.createElementNS('http://www.w3.org/2000/svg', tag)
+}
+
+function sanitizeSvgForCanvas(svg: SVGSVGElement): SVGSVGElement {
+  const cloned = svg.cloneNode(true) as SVGSVGElement
+
+  cloned.querySelectorAll('image').forEach(el => el.remove())
+  cloned.querySelectorAll('foreignObject').forEach((foreignObject) => {
+    const textContent = foreignObject.textContent?.replace(/\s+/g, ' ').trim()
+    if (!textContent) {
+      foreignObject.remove()
+      return
+    }
+
+    const x = Number(foreignObject.getAttribute('x') || 0)
+    const y = Number(foreignObject.getAttribute('y') || 0)
+    const height = Number(foreignObject.getAttribute('height') || 20)
+    const text = createSvgElement('text')
+    text.setAttribute('x', String(x))
+    text.setAttribute('y', String(y + height / 2))
+    text.setAttribute('dominant-baseline', 'middle')
+    text.setAttribute('font-size', '14')
+    text.setAttribute('font-family', 'Arial, "Microsoft YaHei", sans-serif')
+    text.setAttribute('fill', '#333')
+    text.textContent = textContent
+    foreignObject.replaceWith(text)
+  })
+
+  return cloned
+}
+
+function getExportFontSize(svg: SVGSVGElement): number {
+  const text = svg.querySelector('text, foreignObject')
+  if (!text)
+    return 14
+
+  const fontSize = Number.parseFloat(getComputedStyle(text).fontSize || '')
+  if (Number.isFinite(fontSize) && fontSize > 0)
+    return fontSize
+
+  const attrSize = Number.parseFloat(text.getAttribute('font-size') || '')
+  return Number.isFinite(attrSize) && attrSize > 0 ? attrSize : 14
+}
+
+function stripMindmapNoise(md: string): string {
+  return absolutizeMarkdownImages(stripSourceLink(md || ''))
+    // 笔记里的截图/封面图片在思维导图中会被当作超大 SVG foreignObject，
+    // 容易把导图挤成截图里那种“只剩半框/一条竖线”的效果。导图只保留文字层级。
+    .replace(/!\[[^\]]*\]\([^)]*\)/g, '')
+    .replace(/<img\b[^>]*>/gi, '')
+}
+
+async function fit() {
+  await nextTick()
+  requestAnimationFrame(() => mm?.fit())
+}
+
+async function render() {
+  if (!svgRef.value)
+    return
+  const { root } = transformer.transform(stripMindmapNoise(props.markdown))
+  if (!mm)
+    mm = Markmap.create(svgRef.value, { autoFit: true }, root)
+  else
+    await mm.setData(root)
+  await fit()
+}
+
+async function toPngBlob(): Promise<Blob> {
+  await fit()
+  await nextTick()
+  if (!svgRef.value)
+    throw new Error('思维导图尚未渲染完成')
+
+  const svg = svgRef.value
+  const bbox = svg.getBBox()
+  const padding = 48
+  const x = Math.floor(bbox.x - padding)
+  const y = Math.floor(bbox.y - padding)
+  const width = Math.max(Math.ceil(bbox.width + padding * 2), 1)
+  const height = Math.max(Math.ceil(bbox.height + padding * 2), 1)
+  const cloned = sanitizeSvgForCanvas(svg)
+
+  cloned.setAttribute('xmlns', 'http://www.w3.org/2000/svg')
+  cloned.setAttribute('width', String(width))
+  cloned.setAttribute('height', String(height))
+  cloned.setAttribute('viewBox', `${x} ${y} ${width} ${height}`)
+  cloned.insertAdjacentHTML('afterbegin', `<rect width="100%" height="100%" fill="#fff"/>`)
+
+  const svgText = new XMLSerializer().serializeToString(cloned)
+  const url = URL.createObjectURL(new Blob([svgText], { type: 'image/svg+xml;charset=utf-8' }))
+
+  try {
+    const img = new Image()
+    img.decoding = 'async'
+    img.src = url
+    await img.decode()
+
+    // 不写死某个导出宽度：按导图内容和文字字号动态反推 PNG 倍率。
+    // 目标是让导出的正文至少有 MIN_EXPORT_FONT_PX 像素高，小图自动放大，
+    // 大图则按内容尺寸导出；同时限制最大边长，避免复杂导图撑爆内存。
+    const fontScale = MIN_EXPORT_FONT_PX / getExportFontSize(svg)
+    const widthScale = MIN_EXPORT_WIDTH / width
+    const rawScale = Math.max(window.devicePixelRatio || 1, fontScale, widthScale)
+    const sideLimitScale = Math.min(MAX_CANVAS_SIDE / width, MAX_CANVAS_SIDE / height)
+    const scale = Math.max(1, Math.min(rawScale, MAX_EXPORT_SCALE, sideLimitScale))
+    const canvas = document.createElement('canvas')
+    canvas.width = Math.ceil(width * scale)
+    canvas.height = Math.ceil(height * scale)
+    const ctx = canvas.getContext('2d')
+    if (!ctx)
+      throw new Error('当前浏览器不支持 Canvas 导出')
+    ctx.fillStyle = '#fff'
+    ctx.fillRect(0, 0, canvas.width, canvas.height)
+    ctx.scale(scale, scale)
+    ctx.drawImage(img, 0, 0, width, height)
+    return await canvasToBlob(canvas)
+  }
+  finally {
+    URL.revokeObjectURL(url)
+  }
+}
+
+defineExpose({
+  toPngBlob,
+})
+
+onMounted(() => {
+  render()
+  if (wrapRef.value) {
+    resizeObserver = new ResizeObserver(() => fit())
+    resizeObserver.observe(wrapRef.value)
+  }
+})
+
+onUnmounted(() => {
+  resizeObserver?.disconnect()
+  resizeObserver = null
+  mm?.destroy()
+  mm = null
+})
+
 watch(() => props.markdown, render)
+
 </script>
 
 <template>
-  <div class="w-full h-full bg-white rounded border overflow-hidden">
+  <div ref="wrapRef" class="w-full h-full min-h-[360px] bg-white rounded border overflow-hidden">
-    <svg ref="svgRef" class="w-full h-full" />
+    <svg ref="svgRef" class="w-full h-full min-h-[360px]" />
   </div>
 </template>

BillNote_extension/src/logic/task-display.ts

@@ -0,0 +1,21 @@
+import type { TaskRecord } from './types'
+
+const SITE_SUFFIX_RE = /\s*[-_—–|｜]\s*(哔哩哔哩|bilibili|youtube|抖音|douyin|快手|kuaishou)\s*$/i
+
+export function normalizeVideoTitle(title: string | undefined | null): string | undefined {
+  const value = title?.trim()
+  if (!value)
+    return undefined
+  return value
+    .replace(SITE_SUFFIX_RE, '')
+    .trim() || value
+}
+
+export function getTaskDisplayTitle(task: TaskRecord | undefined | null, fallbackTitle?: string): string {
+  if (!task)
+    return normalizeVideoTitle(fallbackTitle) || ''
+  return normalizeVideoTitle((task.result?.audio_meta as { title?: string } | undefined)?.title)
+    || normalizeVideoTitle(task.title)
+    || normalizeVideoTitle(fallbackTitle)
+    || task.videoUrl
+}

BillNote_extension/src/popup/Popup.vue

@@ -5,6 +5,7 @@ import { settings, settingsReady, tasks, tasksReady, upsertTask } from '~/logic/
 import { generateNote, getTaskStatus, resolveImageUrl } from '~/logic/api'
 import { fetchBilibiliSubtitle } from '~/logic/bilibili-subtitle'
 import { NOTE_FORMATS, NOTE_STYLES, type NoteFormat, type TaskRecord } from '~/logic/types'
+import { getTaskDisplayTitle, normalizeVideoTitle } from '~/logic/task-display'
 
 const tabUrl = ref<string>('')
 const tabTitle = ref<string>('')
@@ -43,7 +44,7 @@ async function poll(taskId: string) {
       createdAt: activeTask.value?.createdAt ?? Date.now(),
       updatedAt: Date.now(),
       result: res.result ?? activeTask.value?.result,
-      title: activeTask.value?.title,
+      title: activeTask.value?.title || normalizeVideoTitle(tabTitle.value),
     })
     if (res.status !== 'SUCCESS' && res.status !== 'FAILED')
       pollTimer = setTimeout(() => poll(taskId), 3000)
@@ -95,7 +96,7 @@ async function start() {
       message: '已提交',
       createdAt: Date.now(),
       updatedAt: Date.now(),
-      title: tabTitle.value || undefined,
+      title: normalizeVideoTitle(tabTitle.value),
     })
     poll(task_id)
     // 提交后顺手把侧边栏拉起来，免得用户来回切窗口
@@ -144,10 +145,7 @@ function selectTask(id: string) {
 }
 
 const activeCover = computed(() => activeTask.value?.result?.audio_meta?.cover_url as string | undefined)
-const activeTitle = computed(() =>
+const activeTitle = computed(() => getTaskDisplayTitle(activeTask.value, tabTitle.value))
-  (activeTask.value?.result?.audio_meta?.title as string | undefined)
-  || activeTask.value?.title
-  || tabTitle.value)
 
 function fmtTime(ts?: number) {
   if (!ts)
@@ -182,8 +180,8 @@ onUnmounted(() => {
       <button class="text-xs text-gray-500 hover:text-gray-800" @click="openOptions">设置</button>
     </header>
 
-    <div class="text-xs text-gray-500 truncate" :title="tabUrl">
+    <div class="text-xs text-gray-500 truncate" :title="normalizeVideoTitle(tabTitle) || tabUrl">
-      {{ tabUrl || '当前没有打开的标签页' }}
+      {{ normalizeVideoTitle(tabTitle) || tabUrl || '当前没有打开的标签页' }}
     </div>
 
     <div v-if="!supported" class="text-xs text-amber-700 bg-amber-50 p-2 rounded">
@@ -336,8 +334,8 @@ onUnmounted(() => {
           :class="{ 'bg-blue-50': t.taskId === activeTaskId }"
           @click="selectTask(t.taskId)"
         >
-          <span class="truncate flex-1" :title="t.title || t.videoUrl">
+          <span class="truncate flex-1" :title="getTaskDisplayTitle(t)">
-            {{ (t.result?.audio_meta as { title?: string } | undefined)?.title || t.title || t.videoUrl }}
+            {{ getTaskDisplayTitle(t) }}
           </span>
           <span class="text-gray-500 shrink-0">{{ t.status }}</span>
         </li>

BillNote_extension/src/sidepanel/Sidepanel.vue

@@ -3,14 +3,17 @@ import { computed, onMounted, onUnmounted, ref } from 'vue'
 import { getTaskStatus, resolveImageUrl } from '~/logic/api'
 import { tasks, tasksReady, settingsReady, upsertTask } from '~/logic/storage'
 import type { TaskRecord } from '~/logic/types'
+import { getTaskDisplayTitle } from '~/logic/task-display'
 
 type ViewMode = 'markdown' | 'mindmap' | 'chat'
 
 const activeTaskId = ref<string>('')
 const activeTask = computed<TaskRecord | undefined>(() => tasks.value?.find(t => t.taskId === activeTaskId.value))
 const errorMsg = ref('')
+const successMsg = ref('')
 const viewMode = ref<ViewMode>('markdown')
 const showHistory = ref(false)
+const mindMapRef = ref<{ toPngBlob: () => Promise<Blob> } | null>(null)
 
 const isDone = computed(() => activeTask.value?.status === 'SUCCESS')
 const isFailed = computed(() => activeTask.value?.status === 'FAILED')
@@ -41,7 +44,7 @@ async function poll(taskId: string) {
         message: res.message,
         result: res.result ?? cur.result,
         updatedAt: Date.now(),
-        title: cur.title,
+        title: cur.title || getTaskDisplayTitle(cur),
       })
     }
     if (res.status !== 'SUCCESS' && res.status !== 'FAILED')
@@ -75,11 +78,19 @@ async function copyMarkdown() {
     await navigator.clipboard.writeText(md)
 }
 
+function safeFilename(name: string): string {
+  return (name || 'bilinote')
+    .replace(/[\\/:*?"<>|]/g, '_')
+    .replace(/\s+/g, ' ')
+    .trim()
+    .slice(0, 120) || 'bilinote'
+}
+
 function downloadMarkdown() {
   const md = activeTask.value?.result?.markdown
   if (!md)
     return
-  const title = (activeTask.value?.result?.audio_meta as { title?: string } | undefined)?.title || 'bilinote'
+  const title = safeFilename(getTaskDisplayTitle(activeTask.value))
   const blob = new Blob([md], { type: 'text/markdown;charset=utf-8' })
   const url = URL.createObjectURL(blob)
   const a = document.createElement('a')
@@ -89,11 +100,44 @@ function downloadMarkdown() {
   URL.revokeObjectURL(url)
 }
 
-const activeTitle = computed(() =>
+async function copyMindMapImage() {
-  (activeTask.value?.result?.audio_meta as { title?: string } | undefined)?.title
+  try {
-  || activeTask.value?.title
+    errorMsg.value = ''
-  || activeTask.value?.videoUrl
+    successMsg.value = ''
-  || '')
+    const blob = await mindMapRef.value?.toPngBlob()
+    if (!blob)
+      return
+    await navigator.clipboard.write([
+      new ClipboardItem({ [blob.type]: blob }),
+    ])
+    successMsg.value = '思维导图图片已复制'
+    setTimeout(() => { successMsg.value = '' }, 2000)
+  }
+  catch (e) {
+    errorMsg.value = (e as Error).message || '复制思维导图图片失败'
+  }
+}
+
+async function downloadMindMapImage() {
+  try {
+    errorMsg.value = ''
+    successMsg.value = ''
+    const blob = await mindMapRef.value?.toPngBlob()
+    if (!blob)
+      return
+    const url = URL.createObjectURL(blob)
+    const a = document.createElement('a')
+    a.href = url
+    a.download = `${safeFilename(getTaskDisplayTitle(activeTask.value))}.png`
+    a.click()
+    URL.revokeObjectURL(url)
+  }
+  catch (e) {
+    errorMsg.value = (e as Error).message || '下载思维导图图片失败'
+  }
+}
+
+const activeTitle = computed(() => getTaskDisplayTitle(activeTask.value))
 
 const activeCover = computed(() =>
   (activeTask.value?.result?.audio_meta as { cover_url?: string } | undefined)?.cover_url)
@@ -144,8 +188,8 @@ onUnmounted(() => {
           :class="{ 'bg-white border': t.taskId === activeTaskId }"
           @click="selectTask(t.taskId)"
         >
-          <span class="truncate flex-1" :title="t.title || t.videoUrl">
+          <span class="truncate flex-1" :title="getTaskDisplayTitle(t)">
-            {{ (t.result?.audio_meta as { title?: string } | undefined)?.title || t.title || t.videoUrl }}
+            {{ getTaskDisplayTitle(t) }}
           </span>
           <span class="text-gray-400 shrink-0">{{ STAGE_LABELS[t.status] || t.status }}</span>
         </li>
@@ -155,6 +199,9 @@ onUnmounted(() => {
     <div v-if="errorMsg" class="text-xs text-red-600 px-3 py-1 break-words bg-red-50 shrink-0">
       {{ errorMsg }}
     </div>
+    <div v-if="successMsg" class="text-xs text-green-700 px-3 py-1 break-words bg-green-50 shrink-0">
+      {{ successMsg }}
+    </div>
 
     <section v-if="!activeTask" class="flex-1 flex items-center justify-center text-gray-400 text-xs px-4 text-center">
       还没有任务。在视频页点悬浮按钮、在 popup 提交，或右键菜单选「用 BiliNote 总结」。
@@ -228,6 +275,18 @@ onUnmounted(() => {
           title="下载 .md"
           @click="downloadMarkdown"
         >下载</button>
+        <button
+          v-if="viewMode === 'mindmap'"
+          class="text-gray-500 hover:text-gray-800 px-1.5 py-1 rounded hover:bg-gray-100"
+          title="复制思维导图图片"
+          @click="copyMindMapImage"
+        >复制</button>
+        <button
+          v-if="viewMode === 'mindmap'"
+          class="text-gray-500 hover:text-gray-800 px-1.5 py-1 rounded hover:bg-gray-100"
+          title="下载思维导图 PNG"
+          @click="downloadMindMapImage"
+        >下载</button>
       </div>
 
       <!-- 内容区：占满剩余空间 -->
@@ -240,6 +299,7 @@ onUnmounted(() => {
         />
         <MindMap
           v-else-if="isDone && activeTask.result?.markdown && viewMode === 'mindmap'"
+          ref="mindMapRef"
           :markdown="activeTask.result.markdown"
           class="h-full"
         />

BillNote_frontend/src-tauri/tauri.conf.json

@@ -1,7 +1,7 @@
 {
   "$schema": "../node_modules/@tauri-apps/cli/config.schema.json",
   "productName": "BiliNote",
-  "version": "2.3.4",
+  "version": "2.4.0",
   "identifier": "com.jefferyhuang.bilinote",
   "build": {
     "frontendDist": "../dist",

BillNote_frontend/src/pages/HomePage/components/MarkmapComponent.tsx

@@ -5,6 +5,171 @@ import { Toolbar } from 'markmap-toolbar'
 import 'markmap-toolbar/dist/style.css'
 import JSZip from 'jszip'
 
+const MIN_EXPORT_FONT_PX = 256
+const MIN_EXPORT_WIDTH = 12800
+const WEB_EXPORT_SCALE_FACTOR = 0.34
+const MAX_EXPORT_SCALE = 24
+const MAX_CANVAS_SIDE = 32767
+const MAX_CANVAS_PIXELS = 268000000
+
+function canvasToBlob(canvas: HTMLCanvasElement): Promise<Blob> {
+  return new Promise((resolve, reject) => {
+    canvas.toBlob((blob) => {
+      if (blob) {
+        resolve(blob)
+      } else {
+        reject(new Error('无法创建PNG图片'))
+      }
+    }, 'image/png')
+  })
+}
+
+function createSvgElement<K extends keyof SVGElementTagNameMap>(tag: K): SVGElementTagNameMap[K] {
+  return document.createElementNS('http://www.w3.org/2000/svg', tag)
+}
+
+function sanitizeSvgForCanvas(svg: SVGSVGElement): SVGSVGElement {
+  const cloned = svg.cloneNode(true) as SVGSVGElement
+
+  // markmap 会在 SVG 的顶层 <g> 上写入当前预览视口的 pan/zoom transform。
+  // 导出时我们按内容 bbox 裁剪，如果保留这个视口 transform，会产生双重偏移，
+  // 导致图片内容跑到角落并留下大片空白。这里只移除顶层视口 transform，
+  // 保留内部节点自身的布局 transform。
+  cloned.querySelector(':scope > g')?.removeAttribute('transform')
+
+  cloned.querySelectorAll('image').forEach(el => el.remove())
+  cloned.querySelectorAll('foreignObject').forEach((foreignObject) => {
+    const textContent = foreignObject.textContent?.replace(/\s+/g, ' ').trim()
+    if (!textContent) {
+      foreignObject.remove()
+      return
+    }
+
+    const x = Number(foreignObject.getAttribute('x') || 0)
+    const y = Number(foreignObject.getAttribute('y') || 0)
+    const height = Number(foreignObject.getAttribute('height') || 20)
+    const text = createSvgElement('text')
+    text.setAttribute('x', String(x))
+    text.setAttribute('y', String(y + height / 2))
+    text.setAttribute('dominant-baseline', 'middle')
+    text.setAttribute('font-size', '14')
+    text.setAttribute('font-family', 'Arial, "Microsoft YaHei", sans-serif')
+    text.setAttribute('fill', '#333')
+    text.textContent = textContent
+    foreignObject.replaceWith(text)
+  })
+
+  return cloned
+}
+
+function getExportFontSize(svg: SVGSVGElement): number {
+  const text = svg.querySelector('text, foreignObject')
+  if (!text) return 14
+
+  const fontSize = Number.parseFloat(getComputedStyle(text).fontSize || '')
+  if (Number.isFinite(fontSize) && fontSize > 0) return fontSize
+
+  const attrSize = Number.parseFloat(text.getAttribute('font-size') || '')
+  return Number.isFinite(attrSize) && attrSize > 0 ? attrSize : 14
+}
+
+function getMindmapBounds(svg: SVGSVGElement) {
+  const target = svg.querySelector('g') || svg
+  const bbox = target.getBBox()
+  const padding = 50
+  return {
+    x: Math.floor(bbox.x - padding),
+    y: Math.floor(bbox.y - padding),
+    width: Math.max(Math.ceil(bbox.width + padding * 2), 1),
+    height: Math.max(Math.ceil(bbox.height + padding * 2), 1),
+  }
+}
+
+function stripMindmapImages(markdown: string) {
+  return (markdown || '')
+    // 思维导图只保留文字结构，图片节点会让预览排版和 PNG 导出效果都很差。
+    .replace(/!\[[^\]]*\]\([^)]*\)/g, '')
+    .replace(/<img\b[^>]*>/gi, '')
+}
+
+function transformMindmap(markdown: string) {
+  return transformer.transform(stripMindmapImages(markdown))
+}
+
+function createExportSvg(svgEl: SVGSVGElement) {
+  const bounds = getMindmapBounds(svgEl)
+  const clonedSvg = sanitizeSvgForCanvas(svgEl)
+
+  clonedSvg.setAttribute('xmlns', 'http://www.w3.org/2000/svg')
+  clonedSvg.setAttribute('xmlns:xlink', 'http://www.w3.org/1999/xlink')
+  clonedSvg.setAttribute('width', String(bounds.width))
+  clonedSvg.setAttribute('height', String(bounds.height))
+  clonedSvg.setAttribute('viewBox', `${bounds.x} ${bounds.y} ${bounds.width} ${bounds.height}`)
+  clonedSvg.setAttribute('preserveAspectRatio', 'xMidYMid meet')
+
+  const bgRect = document.createElementNS('http://www.w3.org/2000/svg', 'rect')
+  bgRect.setAttribute('x', String(bounds.x))
+  bgRect.setAttribute('y', String(bounds.y))
+  bgRect.setAttribute('width', String(bounds.width))
+  bgRect.setAttribute('height', String(bounds.height))
+  bgRect.setAttribute('fill', 'white')
+  const firstG = clonedSvg.querySelector('g')
+  clonedSvg.insertBefore(bgRect, firstG || clonedSvg.firstChild)
+
+  return { clonedSvg, ...bounds }
+}
+
+async function exportSvgToPngBlob(svgEl: SVGSVGElement): Promise<Blob> {
+  const { clonedSvg, width, height } = createExportSvg(svgEl)
+  const svgData = new XMLSerializer().serializeToString(clonedSvg)
+  const svgUrl = URL.createObjectURL(new Blob([svgData], { type: 'image/svg+xml;charset=utf-8' }))
+
+  try {
+    const img = new Image()
+    img.decoding = 'async'
+    img.src = svgUrl
+    await img.decode()
+
+    // 按导图内容尺寸和字号动态反推 PNG 倍率，而不是按预览容器或固定倍率导出。
+    const fontScale = MIN_EXPORT_FONT_PX / getExportFontSize(svgEl)
+    const widthScale = MIN_EXPORT_WIDTH / width
+    const rawScale = Math.max(window.devicePixelRatio || 1, fontScale, widthScale)
+    const sideLimitScale = Math.min(MAX_CANVAS_SIDE / width, MAX_CANVAS_SIDE / height)
+    const pixelLimitScale = Math.sqrt(MAX_CANVAS_PIXELS / (width * height))
+    const baseScale = Math.min(rawScale, MAX_EXPORT_SCALE, sideLimitScale, pixelLimitScale)
+    const scale = Math.max(1, baseScale * WEB_EXPORT_SCALE_FACTOR)
+
+    let currentScale = scale
+    let lastError: unknown
+    while (currentScale >= 1) {
+      try {
+        const canvas = document.createElement('canvas')
+        canvas.width = Math.ceil(width * currentScale)
+        canvas.height = Math.ceil(height * currentScale)
+
+        const ctx = canvas.getContext('2d')
+        if (!ctx) {
+          throw new Error('无法获取Canvas上下文')
+        }
+
+        ctx.fillStyle = '#FFFFFF'
+        ctx.fillRect(0, 0, canvas.width, canvas.height)
+        ctx.setTransform(currentScale, 0, 0, currentScale, 0, 0)
+        ctx.drawImage(img, 0, 0, width, height)
+        ctx.setTransform(1, 0, 0, 1, 0, 0)
+
+        return await canvasToBlob(canvas)
+      } catch (error) {
+        lastError = error
+        currentScale = Math.floor(currentScale / 2)
+      }
+    }
+    throw lastError || new Error('导出PNG失败')
+  } finally {
+    URL.revokeObjectURL(svgUrl)
+  }
+}
+
 export interface MarkmapEditorProps {
   /** 要渲染的 Markdown 文本 */
   value: string
@@ -34,6 +199,13 @@ export default function MarkmapEditor({
 
   // 用于跟踪是否处于全屏状态
   const [isFullscreen, setIsFullscreen] = useState(false)
+  const [pngAction, setPngAction] = useState<'idle' | 'exporting' | 'copying'>('idle')
+  const [pngMessage, setPngMessage] = useState('')
+
+  const showPngMessage = (message: string) => {
+    setPngMessage(message)
+    window.setTimeout(() => setPngMessage(''), 2500)
+  }
 
   // 监听全屏状态变化
   useEffect(() => {
@@ -64,7 +236,7 @@ export default function MarkmapEditor({
   // 导出HTML思维导图
   const exportHtml = () => {
     try {
-      const { root } = transformer.transform(value)
+      const { root } = transformMindmap(value)
       const data = JSON.stringify(root)
       
       // 创建HTML内容
@@ -202,7 +374,7 @@ export default function MarkmapEditor({
   // 导出XMind格式思维导图
   const exportXMind = async () => {
     try {
-      const { root } = transformer.transform(value);
+      const { root } = transformMindmap(value);
 
       // 生成唯一ID
       const generateId = () => Math.random().toString(36).substring(2, 15);
@@ -311,100 +483,44 @@ export default function MarkmapEditor({
     try {
       if (!svgRef.current || !mmRef.current) return;
 
-      const svgEl = svgRef.current;
+      setPngAction('exporting');
-      const mm = mmRef.current;
+      setPngMessage('正在生成高清 PNG…');
-
+      const blob = await exportSvgToPngBlob(svgRef.current);
-      // 先调用fit()确保显示完整的思维导图内容
+      const url = URL.createObjectURL(blob);
-      await mm.fit();
+      const a = document.createElement('a');
-      // 等待渲染完成
+      a.href = url;
-      await new Promise(resolve => setTimeout(resolve, 100));
+      a.download = `${title || 'mindmap'}.png`;
-      
+      document.body.appendChild(a);
-      // 获取SVG实际尺寸
+      a.click();
-      const svgWidth = svgEl.width.baseVal.value || svgEl.clientWidth || 800;
+      document.body.removeChild(a);
-      const svgHeight = svgEl.height.baseVal.value || svgEl.clientHeight || 600;
+      URL.revokeObjectURL(url);
-      
+      showPngMessage('PNG 已开始下载');
-      // 设置足够大的缩放比例以确保高清输出
-      const scale = 3;
-      
-      // 克隆SVG以避免修改原始SVG
-      const clonedSvg = svgEl.cloneNode(true) as SVGSVGElement;
-      
-      // 设置SVG的背景为白色
-      const style = document.createElementNS('http://www.w3.org/2000/svg', 'style');
-      style.textContent = 'svg { background-color: white; }';
-      clonedSvg.insertBefore(style, clonedSvg.firstChild);
-      
-      // 确保SVG有正确的命名空间
-      clonedSvg.setAttribute('xmlns', 'http://www.w3.org/2000/svg');
-      clonedSvg.setAttribute('width', svgWidth.toString());
-      clonedSvg.setAttribute('height', svgHeight.toString());
-      
-      // 将SVG转换为Data URI (避免使用Blob URL来解决跨域问题)
-      const svgData = new XMLSerializer().serializeToString(clonedSvg);
-      const svgBase64 = btoa(unescape(encodeURIComponent(svgData)));
-      const dataUri = `data:image/svg+xml;base64,${svgBase64}`;
-      
-      // 创建Canvas
-      const canvas = document.createElement('canvas');
-      canvas.width = svgWidth * scale;
-      canvas.height = svgHeight * scale;
-      
-      // 获取上下文并设置白色背景
-      const ctx = canvas.getContext('2d');
-      if (!ctx) {
-        throw new Error('无法获取Canvas上下文');
-      }
-      
-      // 设置白色背景
-      ctx.fillStyle = '#FFFFFF';
-      ctx.fillRect(0, 0, canvas.width, canvas.height);
-      
-      // 创建Image对象
-      const img = new Image();
-      
-      // 当图片加载完成后，在Canvas上绘制并导出
-      img.onload = () => {
-        try {
-          // 应用缩放
-          ctx.setTransform(scale, 0, 0, scale, 0, 0);
-          
-          // 绘制SVG
-          ctx.drawImage(img, 0, 0);
-          
-          // 重置变换
-          ctx.setTransform(1, 0, 0, 1, 0, 0);
-          
-          // 将Canvas转换为PNG Blob
-          canvas.toBlob((blob) => {
-            if (blob) {
-              // 创建下载链接
-              const url = URL.createObjectURL(blob);
-              const a = document.createElement('a');
-              a.href = url;
-              a.download = `${title || 'mindmap'}.png`;
-              document.body.appendChild(a);
-              a.click();
-              document.body.removeChild(a);
-              URL.revokeObjectURL(url);
-            } else {
-              console.error('无法创建Blob对象');
-            }
-          }, 'image/png');
-        } catch (err) {
-          console.error('Canvas处理失败:', err);
-        }
-      };
-      
-      // 设置图片加载错误处理
-      img.onerror = (error) => {
-        console.error('导出PNG失败（图片加载错误）:', error);
-      };
-      
-      // 开始加载SVG图像 (使用Data URI而不是Blob URL)
-      img.src = dataUri;
-      
     } catch (error) {
       console.error('导出PNG失败:', error);
+      showPngMessage('导出 PNG 失败，请查看控制台');
+    } finally {
+      setPngAction('idle');
+    }
+  };
+
+  // 复制PNG思维导图
+  const copyPng = async () => {
+    try {
+      if (!svgRef.current || !mmRef.current) return;
+
+      setPngAction('copying');
+      setPngMessage('正在复制高清 PNG…');
+      await navigator.clipboard.write([
+        new ClipboardItem({
+          'image/png': exportSvgToPngBlob(svgRef.current),
+        }),
+      ]);
+      showPngMessage('PNG 已复制');
+    } catch (error) {
+      console.error('复制PNG失败:', error);
+      showPngMessage('复制 PNG 失败，请查看控制台');
+    } finally {
+      setPngAction('idle');
     }
   };
 
@@ -428,7 +544,7 @@ export default function MarkmapEditor({
   useEffect(() => {
     const mm = mmRef.current
     if (!mm) return
-    const { root } = transformer.transform(value)
+    const { root } = transformMindmap(value)
     mm.setData(root).then(() => mm.fit())
   }, [value])
 
@@ -459,8 +575,17 @@ export default function MarkmapEditor({
           onClick={exportPng}
           className="rounded p-1 hover:bg-gray-200"
           title="导出PNG图片"
+          disabled={pngAction !== 'idle'}
         >
-          🖼️
+          {pngAction === 'exporting' ? '⏳' : '🖼️'}
+        </button>
+        <button
+          onClick={copyPng}
+          className="rounded p-1 hover:bg-gray-200"
+          title="复制PNG图片"
+          disabled={pngAction !== 'idle'}
+        >
+          {pngAction === 'copying' ? '⏳' : '📋'}
         </button>
         <button
           onClick={exportHtml}
@@ -483,6 +608,11 @@ export default function MarkmapEditor({
           </button>
         )}
       </div>
+      {pngMessage && (
+        <div className="absolute top-11 right-2 z-20 rounded bg-white/95 px-2 py-1 text-xs text-gray-600 shadow">
+          {pngMessage}
+        </div>
+      )}
 
       {/* 如果需要编辑区，就自己加一个 <textarea> 并把 handleChange 绑上 */}
       {/* <textarea value={value} onChange={handleChange} className="mb-2 p-2 border rounded" /> */}

BillNote_frontend/src/pages/SettingPage/about.tsx

@@ -8,7 +8,7 @@ import logo from '@/assets/icon.svg'
 
 // 二维码统一走 GitHub 在线地址：后续在仓库替换图片后，已打包的 App 会自动更新，无需重新发版
 const GITHUB_DOC_RAW = 'https://raw.githubusercontent.com/JefferyHcool/BiliNote/master/doc'
-const wechatQr = `${GITHUB_DOC_RAW}/wechat-group-1.png`
+const wechatQr = `${GITHUB_DOC_RAW}/wechat-gzh.png`
 const remoteInstallQr = `${GITHUB_DOC_RAW}/remote-install-wechat.png`
 
 export default function AboutPage() {
@@ -204,11 +204,11 @@ export default function AboutPage() {
           <div className="mx-auto max-w-3xl">
             <div className="flex flex-col items-center justify-center gap-10 md:flex-row md:items-start">
               <div className="text-center">
-                <h3 className="mb-3 text-xl font-semibold">BiliNote 交流微信群</h3>
+                <h3 className="mb-3 text-xl font-semibold">BiliNote 公众号</h3>
                 <div className="bg-muted mx-auto flex h-52 w-52 items-center justify-center rounded-md">
-                  <img src={wechatQr} alt="BiliNote 交流微信群" className="h-full w-full object-contain" />
+                  <img src={wechatQr} alt="BiliNote 公众号" className="h-full w-full object-contain" />
                 </div>
-                <p className="text-muted-foreground mt-3 text-sm">扫码加入交流群，一起讨论使用问题</p>
+                <p className="text-muted-foreground mt-3 text-sm">扫描公众号，回复「交流群」获取二维码</p>
               </div>
               <div className="text-center">
                 <h3 className="mb-3 text-xl font-semibold">BiliNote AI笔记系统一对一搭建服务</h3>

BillNote_frontend/src/pages/SettingPage/transcriber.tsx

@@ -10,13 +10,16 @@ import {
   SelectValue,
 } from '@/components/ui/select'
 import { Alert, AlertDescription } from '@/components/ui/alert'
-import { AudioLines, AlertTriangle, CheckCircle2, Download, Loader2, Save, XCircle } from 'lucide-react'
+import { Input } from '@/components/ui/input'
+import { AudioLines, AlertTriangle, CheckCircle2, Download, Loader2, Save, XCircle, Plus, Trash2, Boxes } from 'lucide-react'
 import { toast } from 'react-hot-toast'
 import {
   getTranscriberConfig,
   updateTranscriberConfig,
   getModelsStatus,
   downloadModel,
+  addWhisperModel,
+  deleteWhisperModel,
   TranscriberConfig,
   ModelStatus,
 } from '@/services/transcriber'
@@ -33,6 +36,19 @@ export default function Transcriber() {
   const [modelStatuses, setModelStatuses] = useState<ModelStatus[]>([])
   const [mlxModelStatuses, setMlxModelStatuses] = useState<ModelStatus[]>([])
   const [mlxAvailable, setMlxAvailable] = useState(false)
+  // 自定义模型表单
+  const [newModelName, setNewModelName] = useState('')
+  const [newModelTarget, setNewModelTarget] = useState('')
+  const [addingModel, setAddingModel] = useState(false)
+
+  // 重新拉取配置（不重置用户当前的选择），用于增删自定义模型后刷新下拉与列表
+  const reloadConfig = useCallback(async () => {
+    try {
+      setConfig(await getTranscriberConfig())
+    } catch {
+      // 静默
+    }
+  }, [])
 
   const fetchModelsStatus = useCallback(async () => {
     try {
@@ -123,6 +139,41 @@ export default function Transcriber() {
     }
   }
 
+  const handleAddCustomModel = async () => {
+    const name = newModelName.trim()
+    const target = newModelTarget.trim()
+    if (!name || !target) {
+      toast.error('请填写模型名称和 HF repo_id / 本地路径')
+      return
+    }
+    setAddingModel(true)
+    try {
+      await addWhisperModel({ name, target })
+      toast.success(`已添加自定义模型 ${name}`)
+      setNewModelName('')
+      setNewModelTarget('')
+      await reloadConfig()
+      await fetchModelsStatus()
+    } catch {
+      // 后端的具体错误（如重名）已由请求拦截器 toast，这里不重复提示
+    } finally {
+      setAddingModel(false)
+    }
+  }
+
+  const handleDeleteCustomModel = async (name: string) => {
+    try {
+      await deleteWhisperModel(name)
+      toast.success(`已删除自定义模型 ${name}`)
+      // 删的正好是当前选中的，回退到 tiny，避免选中一个不存在的名称
+      if (selectedModelSize === name) setSelectedModelSize('tiny')
+      await reloadConfig()
+      await fetchModelsStatus()
+    } catch {
+      // 拦截器已提示
+    }
+  }
+
   if (loading) {
     return (
       <div className="flex h-64 items-center justify-center">
@@ -272,6 +323,97 @@ export default function Transcriber() {
           </CardContent>
         </Card>
       )}
+
+      {/* 自定义 Whisper 模型（仅 fast-whisper：名称不符合内置 Systran 约定的模型在此登记映射） */}
+      {selectedType === 'fast-whisper' && (
+        <Card>
+          <CardHeader>
+            <CardTitle className="flex items-center gap-2 text-lg">
+              <Boxes className="h-5 w-5" />
+              自定义模型
+              <span className="text-sm font-normal text-neutral-400">
+                登记名称不符合内置约定的模型
+              </span>
+            </CardTitle>
+          </CardHeader>
+          <CardContent className="space-y-4">
+            <Alert className="text-sm">
+              <AlertDescription>
+                填 <strong>HF repo_id</strong>（如{' '}
+                <code className="rounded bg-neutral-100 px-1">Systran/faster-whisper-large-v3</code>
+                ，会自动下载）或<strong>本地模型目录</strong>（如{' '}
+                <code className="rounded bg-neutral-100 px-1">/app/backend/models/my-whisper</code>
+                ，目录内需含 <code className="rounded bg-neutral-100 px-1">model.bin</code>，下载会跳过）。
+                添加后即可在上方「模型大小」下拉中选用。Docker 部署请把模型目录挂载进容器（见 README 的{' '}
+                <code className="rounded bg-neutral-100 px-1">models</code> 卷）。
+              </AlertDescription>
+            </Alert>
+
+            {config.whisper_custom_models &&
+            Object.keys(config.whisper_custom_models).length > 0 ? (
+              <div className="space-y-2">
+                {Object.entries(config.whisper_custom_models).map(([name, target]) => {
+                  const status = modelStatuses.find(m => m.model_size === name)
+                  return (
+                    <div
+                      key={name}
+                      className="flex items-center justify-between gap-3 rounded-md border px-4 py-2.5"
+                    >
+                      <div className="min-w-0">
+                        <div className="flex items-center gap-2 font-medium">
+                          {name}
+                          {status?.downloaded && (
+                            <CheckCircle2 className="h-3.5 w-3.5 text-green-500" />
+                          )}
+                          {status?.downloading && (
+                            <Loader2 className="h-3.5 w-3.5 animate-spin text-neutral-400" />
+                          )}
+                        </div>
+                        <div className="truncate text-xs text-neutral-400" title={target}>
+                          {target}
+                        </div>
+                      </div>
+                      <Button
+                        size="sm"
+                        variant="ghost"
+                        className="text-red-500 hover:text-red-600"
+                        onClick={() => handleDeleteCustomModel(name)}
+                      >
+                        <Trash2 className="h-4 w-4" />
+                      </Button>
+                    </div>
+                  )
+                })}
+              </div>
+            ) : (
+              <p className="text-sm text-neutral-400">还没有自定义模型</p>
+            )}
+
+            <div className="flex flex-col gap-2 sm:flex-row sm:items-center">
+              <Input
+                placeholder="模型名称（自定义，如 my-large-v3）"
+                value={newModelName}
+                onChange={e => setNewModelName(e.target.value)}
+                className="sm:max-w-[220px]"
+              />
+              <Input
+                placeholder="HF repo_id 或本地路径"
+                value={newModelTarget}
+                onChange={e => setNewModelTarget(e.target.value)}
+                className="flex-1"
+              />
+              <Button onClick={handleAddCustomModel} disabled={addingModel}>
+                {addingModel ? (
+                  <Loader2 className="mr-1 h-4 w-4 animate-spin" />
+                ) : (
+                  <Plus className="mr-1 h-4 w-4" />
+                )}
+                添加
+              </Button>
+            </div>
+          </CardContent>
+        </Card>
+      )}
     </div>
   )
 }

BillNote_frontend/src/services/transcriber.ts

@@ -5,6 +5,10 @@ export interface TranscriberConfig {
   whisper_model_size: string
   available_types: { value: string; label: string }[]
   whisper_model_sizes: string[]
+  /** 内置模型映射：size → HF repo_id */
+  whisper_builtin_models?: Record<string, string>
+  /** 用户自定义模型映射：名称 → HF repo_id 或本地路径 */
+  whisper_custom_models?: Record<string, string>
   mlx_whisper_available: boolean
 }
 
@@ -41,3 +45,23 @@ export const downloadModel = async (data: {
 }) => {
   return await request.post('/transcriber_download', data)
 }
+
+export interface WhisperModelsResponse {
+  builtin: Record<string, string>
+  custom: Record<string, string>
+}
+
+/** 列出内置 + 自定义 whisper 模型映射 */
+export const listWhisperModels = async (): Promise<WhisperModelsResponse> => {
+  return await request.get('/whisper_models')
+}
+
+/** 新增自定义模型映射（名称 → HF repo_id 或本地路径） */
+export const addWhisperModel = async (data: { name: string; target: string }) => {
+  return await request.post('/whisper_models', data)
+}
+
+/** 删除自定义模型映射（不会删除已下载的模型文件） */
+export const deleteWhisperModel = async (name: string) => {
+  return await request.delete(`/whisper_models/${encodeURIComponent(name)}`)
+}

CHANGELOG.md

@@ -2,6 +2,27 @@
 
 本项目所有重要变更记录于此。格式参考 [Keep a Changelog](https://keepachangelog.com/zh-CN/1.1.0/)，遵循 [语义化版本](https://semver.org/lang/zh-CN/)。
 
+## [2.4.0] - 2026-06-07
+
+### Added
+
+- **可配置 Whisper 模型**：转写设置支持自定义 Whisper 模型与名称映射，可指定自定义 HuggingFace repo 或本地路径（新增 `backend/app/transcriber/whisper_models.py` + 测试），转写设置页可选择 / 配置模型。
+- **关注公众号获取交流群**：关于页群二维码改为公众号二维码，关注公众号后回复「交流群」即可获取最新群二维码，避免群码过期失效。
+
+### Fixed
+
+- **浏览器扩展**：修复标题显示异常，优化脑图（Markmap）导出。
+
+### Docs
+
+- **GPU/CUDA 部署**：README 补全 GPU/CUDA 部署说明。
+
+## [2.3.4] - 2026-05-27
+
+### Added
+
+- **一对一搭建服务二维码**：新增「BiliNote AI 笔记系统一对一搭建服务」二维码（README + 关于页），扫码加微信、备注「搭建服务」即可咨询。
+
 ## [2.3.3] - 2026-05-22
 
 ### Fixed

README.md

@@ -3,7 +3,7 @@
     <p align="center">
   <img src="./doc/icon.svg" alt="BiliNote Banner" width="50" height="50"  />
 </p>
-<h1 align="center" > BiliNote v2.3.4</h1>
+<h1 align="center" > BiliNote v2.4.0</h1>
 </div>
 
 <p align="center"><i>AI 视频笔记生成工具 让 AI 为你的视频做笔记</i></p>
@@ -303,10 +303,42 @@ sudo apt install ffmpeg
 >
 > Docker 部署已内置 FFmpeg，无需额外安装。
 
-### 🚀 CUDA 加速（可选）
+### 🚀 CUDA / GPU 加速（可选）
-若你希望更快地执行音频转写任务，可使用具备 NVIDIA GPU 的机器，并启用 fast-whisper + CUDA 加速版本：
 
-具体 `fast-whisper` 配置方法，请参考：[fast-whisper 项目地址](http://github.com/SYSTRAN/faster-whisper#requirements)
+本地 **Faster Whisper** 转写可用 NVIDIA GPU 加速（在线引擎 Groq / 必剪 / 快手 与 GPU 无关）。仓库已自带 GPU 镜像与编排，**无需改代码、无需手动配置 device**——后端会自动检测 CUDA，可用就走 GPU，否则回退 CPU。
+
+**1. 宿主机前提**
+
+- NVIDIA 显卡 + 较新驱动（CUDA ≥ 12.4），宿主机 `nvidia-smi` 能正常输出；
+- 安装 [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html)（最易漏的一步，没它 Docker 进不去 GPU）。装完验证：
+  ```bash
+  docker run --rm --gpus all nvidia/cuda:12.4.1-base-ubuntu22.04 nvidia-smi
+  ```
+  能列出显卡即 OK。
+
+**2. 切换到 GPU 编排**（在源码目录里）
+
+CPU 与 GPU 两套 compose 用了相同的容器名，先停掉当前栈再起 GPU 栈：
+
+```bash
+docker-compose down                                     # 停掉当前（CPU）栈
+docker-compose -f docker-compose.gpu.yml up --build -d   # 用 GPU 栈重建
+```
+
+- GPU 栈用 `backend/Dockerfile.gpu`（CUDA 12.4.1 + cuDNN 基础镜像，并额外装 torch 用于 CUDA 检测），compose 已声明 `deploy...devices: nvidia` 自动透传 GPU。
+- **数据不丢**：两套 compose 都把 `./backend` 整目录绑挂进容器，数据库 / 配置 / 已下载模型都保留。
+- 首次构建较大较慢（CUDA 基础镜像数 GB + torch），耐心等。
+
+**3. 启用并确认**
+
+- 「设置 → 音频转写配置」转写引擎选 **Faster Whisper（本地）**，GPU 下可放心选大模型（如 `large-v3`）。
+- 确认真的走了 GPU：`docker logs bilinote-backend | grep -i cuda` 看到 `CUDA 可用，使用 GPU`；或转写时宿主机 `nvidia-smi` 能看到 python 进程占显存。
+
+**国内镜像**：GPU compose 支持 `BASE_REGISTRY` / `APT_MIRROR` / `PIP_INDEX` 这几个 build-arg（注意 `BASE_REGISTRY` 选的源必须支持 `nvidia/cuda` 命名空间，否则拉不到 CUDA 基础镜像）。
+
+**起来了但没走 GPU？** 依次排查：① 宿主机 `nvidia-smi` 是否正常 → ② NVIDIA Container Toolkit 是否装好（上面 `--gpus all` 测试是否通过）→ ③ `docker logs bilinote-backend` 是否有 CUDA / cuDNN 报错（驱动 CUDA 版本需 ≥ 12.4）。
+
+`fast-whisper` 本身的 GPU 依赖说明可参考：[faster-whisper 项目](https://github.com/SYSTRAN/faster-whisper#requirements)
 
 ### 🐳 使用 Docker 一键部署
 
@@ -338,8 +370,8 @@ docker run -d -p 80:80 \
 # 标准部署
 docker-compose up -d
 
-# GPU 加速部署（需要 NVIDIA GPU）
+# GPU 加速部署（需要 NVIDIA GPU + NVIDIA Container Toolkit，详见上方「CUDA / GPU 加速」）
-docker-compose -f docker-compose.gpu.yml up -d
+docker-compose -f docker-compose.gpu.yml up --build -d
 ```
 
 ## 🧠 TODO
@@ -354,18 +386,11 @@ docker-compose -f docker-compose.gpu.yml up -d
 
 ### Contact and Join-联系和加入社区
 
-扫码加入 BiliNote 交流微信群（共 5 个群，任选一个即可；二维码会定期更新，如已失效请到 [Issues](https://github.com/JefferyHcool/BiliNote/issues) 反馈）：
+扫描下方公众号二维码，关注后回复 **「交流群」** 即可获取最新的微信交流群二维码（群码会自动更新，避免过期失效）：
 
 <table align="center">
   <tr>
-    <td align="center"><img src="./doc/wechat-group-1.png" alt="BiliNote 交流群 1" width="200" /><br/>交流群 1</td>
+    <td align="center"><img src="./doc/wechat-gzh.png" alt="BiliNote 公众号" width="200" /><br/>BiliNote 公众号</td>
-    <td align="center"><img src="./doc/wechat-group-2.png" alt="BiliNote 交流群 2" width="200" /><br/>交流群 2</td>
-    <td align="center"><img src="./doc/wechat-group-3.png" alt="BiliNote 交流群 3" width="200" /><br/>交流群 3</td>
-  </tr>
-  <tr>
-    <td align="center"><img src="./doc/wechat-group-4.png" alt="BiliNote 交流群 4" width="200" /><br/>交流群 4</td>
-    <td align="center"><img src="./doc/wechat-group-5.png" alt="BiliNote 交流群 5" width="200" /><br/>交流群 5</td>
-    <td></td>
   </tr>
 </table>
 

backend/app/downloaders/bilibili_downloader.py

@@ -1,6 +1,9 @@
+import base64
 import os
 import json
 import logging
+import random
+import string
 import tempfile
 from abc import ABC
 from typing import Union, Optional, List
@@ -18,6 +21,54 @@ from app.services.cookie_manager import CookieConfigManager
 logger = logging.getLogger(__name__)
 
 
+def _patch_bilibili_extractor():
+    """
+    Monkey-patch yt-dlp's BilibiliBaseIE._download_playinfo to inject the
+    dm_img_* / web_location risk-control parameters that Bilibili's
+    wbi/playurl gateway started requiring (returns HTTP 412 without them).
+
+    The parameter format (string.printable source + [:-2] base64 truncation)
+    mirrors yt-dlp's own implementation for the same fields in the channel
+    search endpoint (yt_dlp/extractor/bilibili.py, BiliBiliSpaceIE).
+    """
+    try:
+        from yt_dlp.extractor.bilibili import BilibiliBaseIE
+
+        # Guard: skip if already patched (e.g. module reloaded in tests)
+        if getattr(BilibiliBaseIE._download_playinfo, '_bili_dm_patched', False):
+            return
+
+        _original_download_playinfo = BilibiliBaseIE._download_playinfo
+
+        def _patched_download_playinfo(self, bvid, cid, headers=None, query=None):
+            # Inject dummy risk-control fingerprints expected by Bilibili's gateway.
+            # The [:-2] truncation and string.printable source intentionally match
+            # yt-dlp's own pattern used for the x/space/wbi/arc/search endpoint.
+            extra = {
+                'web_location': 1550101,
+                'dm_img_list': '[]',
+                'dm_img_str': base64.b64encode(
+                    ''.join(random.choices(string.printable, k=random.randint(16, 64))).encode()
+                )[:-2].decode(),
+                'dm_cover_img_str': base64.b64encode(
+                    ''.join(random.choices(string.printable, k=random.randint(32, 128))).encode()
+                )[:-2].decode(),
+                'dm_img_inter': '{"ds":[],"wh":[6093,6631,31],"of":[430,760,380]}',
+            }
+            # Caller-supplied query params take precedence over the dummy values
+            merged_query = {**extra, **(query or {})}
+            return _original_download_playinfo(self, bvid, cid, headers=headers, query=merged_query)
+
+        _patched_download_playinfo._bili_dm_patched = True
+        BilibiliBaseIE._download_playinfo = _patched_download_playinfo
+        logger.info("Applied Bilibili wbi/playurl dm_img patch to yt-dlp BilibiliBaseIE")
+    except Exception as e:
+        logger.warning("Failed to apply Bilibili dm_img patch: %s", e)
+
+
+_patch_bilibili_extractor()
+
+
 class BilibiliDownloader(Downloader, ABC):
     def __init__(self):
         super().__init__()

backend/app/routers/config.py

@@ -61,16 +61,53 @@ WHISPER_MODEL_SIZES = ["tiny", "base", "small", "medium", "large-v3", "large-v3-
 @router.get("/transcriber_config")
 def get_transcriber_config():
     from app.transcriber.transcriber_provider import MLX_WHISPER_AVAILABLE
+    from app.transcriber.whisper_models import get_registry, BUILTIN_WHISPER_MODELS
 
+    registry = get_registry()
     config = transcriber_config_manager.get_config()
     return R.success(data={
         **config,
         "available_types": AVAILABLE_TRANSCRIBER_TYPES,
-        "whisper_model_sizes": WHISPER_MODEL_SIZES,
+        # 内置可见档位 + 用户自定义模型，供前端下拉
+        "whisper_model_sizes": registry.visible_model_names(),
+        "whisper_builtin_models": BUILTIN_WHISPER_MODELS,
+        "whisper_custom_models": registry.get_custom_models(),
         "mlx_whisper_available": MLX_WHISPER_AVAILABLE,
     })
 
 
+class WhisperCustomModelRequest(BaseModel):
+    name: str
+    target: str  # HF repo_id（如 Systran/faster-whisper-large-v3）或本地模型目录路径
+
+
+@router.get("/whisper_models")
+def list_whisper_models():
+    """列出内置 + 用户自定义的 whisper 模型映射。"""
+    from app.transcriber.whisper_models import get_registry, BUILTIN_WHISPER_MODELS
+    reg = get_registry()
+    return R.success(data={"builtin": BUILTIN_WHISPER_MODELS, "custom": reg.get_custom_models()})
+
+
+@router.post("/whisper_models")
+def add_whisper_model(data: WhisperCustomModelRequest):
+    """新增自定义 whisper 模型映射（名称 → HF repo_id 或本地路径）。"""
+    from app.transcriber.whisper_models import get_registry
+    try:
+        custom = get_registry().add_custom_model(data.name, data.target)
+    except ValueError as e:
+        return R.error(msg=str(e))
+    return R.success(data={"custom": custom}, msg="已添加自定义模型")
+
+
+@router.delete("/whisper_models/{name}")
+def delete_whisper_model(name: str):
+    """删除自定义 whisper 模型映射（不会删除已下载的模型文件）。"""
+    from app.transcriber.whisper_models import get_registry
+    custom = get_registry().remove_custom_model(name)
+    return R.success(data={"custom": custom}, msg="已删除自定义模型")
+
+
 @router.post("/transcriber_config")
 def update_transcriber_config(data: TranscriberConfigRequest):
     config = transcriber_config_manager.update_config(
@@ -119,14 +156,27 @@ _downloading: dict[str, str] = {}  # model_size -> status ("downloading" | "done
 def _check_whisper_model_exists(model_size: str, subdir: str = "whisper") -> bool:
     """检查指定 whisper 模型是否已下载完整到本地。
 
-    faster-whisper 把模型缓存在 HF cache 布局下：
+    先把模型名 resolve 成可加载标识，再按类型判定：
-      <model_dir>/models--Systran--faster-whisper-{size}/snapshots/<hash>/model.bin
+      - 本地路径模型 → 直接看该目录下有没有 model.bin
-    必须能在某个 snapshot 目录里找到 model.bin 才算完成。
+      - HF repo_id → 看 HF cache 布局
-    （历史 modelscope 布局 <model_dir>/whisper-{size}/model.bin 也兼容识别。）
+        <model_dir>/models--{org}--{name}/snapshots/<hash>/model.bin
+        （历史 modelscope 布局 <model_dir>/whisper-{size}/model.bin 也兼容识别）
     """
+    from app.transcriber.whisper_models import (
+        resolve_whisper_model,
+        is_local_target,
+        hf_cache_dirname,
+    )
+    try:
+        target = resolve_whisper_model(model_size)
+    except Exception:
+        return False
+    if is_local_target(target):
+        return (Path(target) / "model.bin").exists()
+
     model_dir = Path(get_model_dir(subdir))
-    # HF cache 布局
+    # HF cache 布局（适配任意 org/repo，不再写死 Systran）
-    hf_repo_dir = model_dir / f"models--Systran--faster-whisper-{model_size}" / "snapshots"
+    hf_repo_dir = model_dir / hf_cache_dirname(target) / "snapshots"
     if hf_repo_dir.exists():
         for snapshot in hf_repo_dir.iterdir():
             if (snapshot / "model.bin").exists():
@@ -157,9 +207,10 @@ def _check_mlx_whisper_model_exists(model_size: str) -> bool:
 
 @router.get("/transcriber_models_status")
 def get_transcriber_models_status():
-    """返回所有 whisper 模型的下载状态。"""
+    """返回所有 whisper 模型的下载状态（含用户自定义模型）。"""
+    from app.transcriber.whisper_models import get_registry
     statuses = []
-    for size in WHISPER_MODEL_SIZES:
+    for size in get_registry().visible_model_names():
         downloaded = _check_whisper_model_exists(size, "whisper")
         download_status = _downloading.get(size)
         statuses.append({
@@ -198,13 +249,15 @@ class ModelDownloadRequest(BaseModel):
 
 
 def _do_download_whisper(model_size: str):
-    """后台下载 faster-whisper 模型。
+    """后台下载 faster-whisper 模型（支持内置 size / 自定义 repo_id / 本地路径）。
 
-    直接走 huggingface_hub.snapshot_download，把模型放到 HF cache 布局里——
+    模型名先 resolve：
-    这样 faster-whisper 加载时（WhisperModel(model_size_or_path=size_name,
+      - 本地路径模型：无需下载，目录里有 model.bin 即 done，否则 failed；
-    download_root=model_dir)）能直接命中缓存，跟加载路径完全对齐。
+      - HF repo_id：snapshot_download 到 HF cache 布局（cache_dir=model_dir），
+        与加载逻辑 WhisperModel(download_root=model_dir) 完全对齐。
     """
     from huggingface_hub import snapshot_download
+    from app.transcriber.whisper_models import resolve_whisper_model, is_local_target
 
     try:
         _downloading[model_size] = "downloading"
@@ -214,12 +267,21 @@ def _do_download_whisper(model_size: str):
         if _check_whisper_model_exists(model_size, "whisper"):
             _downloading[model_size] = "done"
             return
-        repo_id = f"Systran/faster-whisper-{model_size}"
+
-        logger.info(f"开始下载 whisper 模型: {repo_id}")
+        target = resolve_whisper_model(model_size)
+        if is_local_target(target):
+            # 本地模型不下载，只校验 model.bin 是否就位
+            ok = (Path(target) / "model.bin").exists()
+            _downloading[model_size] = "done" if ok else "failed"
+            if not ok:
+                logger.warning(f"本地模型 {model_size} 路径 {target} 下没有 model.bin，无法使用")
+            return
+
+        logger.info(f"开始下载 whisper 模型: {model_size} ← {target}")
         # 跟 faster-whisper utils.py 用同样的 allow_patterns，避免多下无关文件；
         # 不传 local_dir 让它走 HF 默认 cache 布局（与加载逻辑对齐）
         snapshot_download(
-            repo_id,
+            target,
             cache_dir=model_dir,
             allow_patterns=[
                 "config.json",
@@ -268,11 +330,11 @@ def _do_download_mlx_whisper(model_size: str):
 
 @router.post("/transcriber_download")
 def download_transcriber_model(data: ModelDownloadRequest, background_tasks: BackgroundTasks):
-    """触发后台下载指定的 whisper 模型。"""
+    """触发后台下载指定的 whisper 模型（fast-whisper 支持内置档位 + 自定义模型）。"""
-    if data.model_size not in WHISPER_MODEL_SIZES:
-        return R.error(msg=f"不支持的模型大小: {data.model_size}")
-
     if data.transcriber_type == "mlx-whisper":
+        # mlx 只认内置档位（mlx-community 的固定映射）
+        if data.model_size not in WHISPER_MODEL_SIZES:
+            return R.error(msg=f"MLX 不支持的模型大小: {data.model_size}")
         if platform.system() != "Darwin":
             return R.error(msg="MLX Whisper 仅支持 macOS")
         key = f"mlx-{data.model_size}"
@@ -280,6 +342,10 @@ def download_transcriber_model(data: ModelDownloadRequest, background_tasks: Bac
             return R.success(msg="模型正在下载中")
         background_tasks.add_task(_do_download_mlx_whisper, data.model_size)
     else:
+        # fast-whisper：内置档位 / 自定义 repo_id / 本地路径都允许
+        from app.transcriber.whisper_models import get_registry
+        if not get_registry().is_known(data.model_size):
+            return R.error(msg=f"不支持的模型: {data.model_size}（请先在自定义模型中登记）")
         if _downloading.get(data.model_size) == "downloading":
             return R.success(msg="模型正在下载中")
         background_tasks.add_task(_do_download_whisper, data.model_size)

backend/app/transcriber/whisper.py

@@ -3,6 +3,11 @@ from faster_whisper import WhisperModel
 from app.decorators.timeit import timeit
 from app.models.transcriber_model import TranscriptSegment, TranscriptResult
 from app.transcriber.base import Transcriber
+from app.transcriber.whisper_models import (
+    resolve_whisper_model,
+    is_local_target,
+    hf_cache_dirname,
+)
 from app.utils.env_checker import is_cuda_available, is_torch_installed
 from app.utils.logger import get_logger
 from app.utils.path_helper import get_model_dir
@@ -55,8 +60,12 @@ class WhisperTranscriber(Transcriber):
             self.model = self._build_model(model_size, model_dir)
 
     def _build_model(self, model_size: str, model_dir: str) -> WhisperModel:
+        # resolve 把模型名映射成可加载标识：内置 size→Systran repo_id、自定义映射、
+        # 直通的 repo_id 或本地路径。faster-whisper 对本地目录走 os.path.isdir 分支，
+        # 对 repo_id 走 download_model(cache_dir=download_root)，两者都吃 model_size_or_path。
+        target = resolve_whisper_model(model_size)
         return WhisperModel(
-            model_size_or_path=model_size,  # 传 size name，让 faster-whisper 自己映射到 Systran/faster-whisper-*
+            model_size_or_path=target,
             device=self.device,
             compute_type=self.compute_type,
             download_root=model_dir,
@@ -64,14 +73,23 @@ class WhisperTranscriber(Transcriber):
 
     @staticmethod
     def _purge_cache(model_dir: str, model_size: str) -> None:
-        """删掉 HF cache 里这个 size 对应的 snapshot 目录，强制下次重新下载。
+        """加载失败时清掉对应 HF cache 的 snapshot 目录，强制下次重下。
 
-        HF cache 布局：<model_dir>/models--Systran--faster-whisper-{size}/
+        关键：本地路径模型**绝不删**——那是用户自己的文件，删了就是数据丢失；
-        没找到也不报错——可能用户改了 endpoint 或者 cache 布局变了。
+        只清 HF cache 布局 <model_dir>/models--{org}--{name}/（含历史 modelscope 目录）。
         """
+        try:
+            target = resolve_whisper_model(model_size)
+        except Exception:
+            target = model_size
+        if is_local_target(target):
+            logger.warning(
+                f"模型 {model_size} 指向本地路径 {target}，加载失败不清理用户文件，请检查该目录是否完整"
+            )
+            return
         candidates = [
-            Path(model_dir) / f"models--Systran--faster-whisper-{model_size}",
+            Path(model_dir) / hf_cache_dirname(target),       # HF cache: models--org--name
-            Path(model_dir) / f"whisper-{model_size}",  # 历史 modelscope 目录，顺手清掉
+            Path(model_dir) / f"whisper-{model_size}",        # 历史 modelscope 目录，顺手清掉
         ]
         for path in candidates:
             if path.exists():

backend/app/transcriber/whisper_models.py

@@ -0,0 +1,156 @@
+"""fast-whisper 模型名 → 可加载标识（HF repo_id 或本地路径）的映射注册表。
+
+背景：faster-whisper 加载时 `WhisperModel(model_size_or_path=...)` 接受三种入参——
+内置 size 名、HF repo_id（含 "/"）、或本地模型目录（`os.path.isdir` 命中则直接用）。
+此前后端把「size → Systran/faster-whisper-{size}」这层约定**隐式**散落在加载/下载/
+检测三处，用户想用命名不符合该约定的模型（比如社区微调版、或自己下到本地的模型）就接不上。
+
+本模块把映射**显式化 + 可配置**（对齐 mlx_whisper_transcriber.MLX_MODEL_MAP 的模式）：
+  - 内置：size → Systran/faster-whisper-{size}
+  - 自定义：用户在 config/whisper_models.json 登记 {名称: "<repo_id 或本地路径>"}
+    （JSON 持久化；Docker 下随 config 卷持久化）
+
+解析优先级（resolve）：自定义 > 内置 > 直通（含 "/" 当 repo_id；已存在目录当本地路径）。
+加载 / 下载 / 完整性检测三处统一调用 resolve，路径不再各写各的。
+"""
+import json
+import os
+from pathlib import Path
+from typing import Dict, List
+
+from app.utils.logger import get_logger
+
+logger = get_logger(__name__)
+
+# 内置模型：size → faster-whisper 兼容的 HF repo_id（CTranslate2 转换版，Systran 官方维护）。
+BUILTIN_WHISPER_MODELS: Dict[str, str] = {
+    "tiny": "Systran/faster-whisper-tiny",
+    "base": "Systran/faster-whisper-base",
+    "small": "Systran/faster-whisper-small",
+    "medium": "Systran/faster-whisper-medium",
+    "large-v1": "Systran/faster-whisper-large-v1",
+    "large-v2": "Systran/faster-whisper-large-v2",
+    "large-v3": "Systran/faster-whisper-large-v3",
+    "large-v3-turbo": "Systran/faster-whisper-large-v3-turbo",
+}
+
+# 前端下拉默认展示的内置档位（保持与历史 WHISPER_MODEL_SIZES 一致，不把 8 个全列出来）
+DEFAULT_VISIBLE_BUILTINS: List[str] = ["tiny", "base", "small", "medium", "large-v3", "large-v3-turbo"]
+
+
+def is_local_target(target: str) -> bool:
+    """判断解析出的 target 是本地路径而非 HF repo_id。
+
+    HF repo_id 形如 'Org/Name'（恰一个斜杠、无前导斜杠、非已存在目录）。
+    本地路径：绝对路径 / 以 . 或 ~ 开头 / 已存在的目录。
+    """
+    if not target:
+        return False
+    if os.path.isabs(target) or target.startswith(".") or target.startswith("~"):
+        return True
+    return os.path.isdir(target)
+
+
+def hf_cache_dirname(repo_id: str) -> str:
+    """huggingface_hub snapshot 的本地缓存目录名：Org/Name → models--Org--Name。"""
+    return "models--" + repo_id.replace("/", "--")
+
+
+class WhisperModelRegistry:
+    """内置 + 用户自定义的 whisper 模型映射，自定义部分持久化到 JSON。"""
+
+    def __init__(self, filepath: str = "config/whisper_models.json"):
+        self.path = Path(filepath)
+        self.path.parent.mkdir(parents=True, exist_ok=True)
+
+    # ---- 持久化 ----
+    def _read_custom(self) -> Dict[str, str]:
+        if not self.path.exists():
+            return {}
+        try:
+            with self.path.open("r", encoding="utf-8") as f:
+                data = json.load(f) or {}
+        except Exception as e:
+            logger.warning(f"读取自定义 whisper 模型配置失败，按空处理: {e}")
+            return {}
+        out: Dict[str, str] = {}
+        for name, val in data.items():
+            if isinstance(val, str) and val.strip():
+                out[name] = val.strip()
+            elif isinstance(val, dict) and isinstance(val.get("target"), str):
+                out[name] = val["target"].strip()
+        return out
+
+    def _write_custom(self, data: Dict[str, str]) -> None:
+        with self.path.open("w", encoding="utf-8") as f:
+            json.dump(data, f, ensure_ascii=False, indent=2)
+
+    # ---- 查询 ----
+    def get_custom_models(self) -> Dict[str, str]:
+        return self._read_custom()
+
+    def visible_model_names(self) -> List[str]:
+        """给前端下拉 / 下载状态用：默认可见内置档位 + 全部自定义名称。"""
+        names = list(DEFAULT_VISIBLE_BUILTINS)
+        for name in self._read_custom():
+            if name not in names:
+                names.append(name)
+        return names
+
+    def is_known(self, name: str) -> bool:
+        try:
+            self.resolve(name)
+            return True
+        except ValueError:
+            return False
+
+    def resolve(self, name: str) -> str:
+        """模型名 → 可加载标识（HF repo_id 或本地路径）。
+
+        优先级：自定义映射 > 内置映射 > 直通（含 "/" 的 repo_id / 已存在的本地目录）。
+        无法识别时抛 ValueError。
+        """
+        name = (name or "").strip()
+        custom = self._read_custom()
+        if name in custom:
+            return custom[name]
+        if name in BUILTIN_WHISPER_MODELS:
+            return BUILTIN_WHISPER_MODELS[name]
+        # 直通：用户直接把 repo_id（含 "/"）或本地已存在目录当 model_size 传进来
+        if "/" in name or os.path.isdir(name):
+            return name
+        raise ValueError(
+            f"未知 whisper 模型 '{name}'。内置可选: {', '.join(BUILTIN_WHISPER_MODELS)}；"
+            "或在「音频转写配置」添加自定义模型（HF repo_id 或本地路径）。"
+        )
+
+    # ---- 增删 ----
+    def add_custom_model(self, name: str, target: str) -> Dict[str, str]:
+        name = (name or "").strip()
+        target = (target or "").strip()
+        if not name or not target:
+            raise ValueError("模型名称与目标（HF repo_id 或本地路径）都不能为空")
+        if name in BUILTIN_WHISPER_MODELS:
+            raise ValueError(f"'{name}' 与内置模型重名，请换一个名称")
+        data = self._read_custom()
+        data[name] = target
+        self._write_custom(data)
+        return data
+
+    def remove_custom_model(self, name: str) -> Dict[str, str]:
+        data = self._read_custom()
+        data.pop((name or "").strip(), None)
+        self._write_custom(data)
+        return data
+
+
+# 模块级单例
+_registry = WhisperModelRegistry()
+
+
+def get_registry() -> WhisperModelRegistry:
+    return _registry
+
+
+def resolve_whisper_model(name: str) -> str:
+    return _registry.resolve(name)

backend/tests/test_whisper_models.py

@@ -0,0 +1,132 @@
+"""Unit tests for app.transcriber.whisper_models（whisper 模型名→标识 的映射注册表）。
+
+直接按文件路径加载被测模块，并桩掉 app.utils.logger，避免触发 app/__init__.py
+（会 import faster_whisper / ctranslate2 等重依赖），使本测试无需安装转写依赖即可运行。
+"""
+import importlib.util
+import logging
+import os
+import pathlib
+import sys
+import tempfile
+import types
+import unittest
+
+ROOT = pathlib.Path(__file__).resolve().parents[1]
+MODULE_PATH = ROOT / "app" / "transcriber" / "whisper_models.py"
+
+
+def _load_module():
+    if "app" not in sys.modules:
+        app_pkg = types.ModuleType("app")
+        app_pkg.__path__ = []  # 标记为 package
+        sys.modules["app"] = app_pkg
+    if "app.utils" not in sys.modules:
+        utils_pkg = types.ModuleType("app.utils")
+        utils_pkg.__path__ = []
+        sys.modules["app.utils"] = utils_pkg
+    if "app.utils.logger" not in sys.modules:
+        logger_mod = types.ModuleType("app.utils.logger")
+        logger_mod.get_logger = lambda name=None: logging.getLogger(name or "test")
+        sys.modules["app.utils.logger"] = logger_mod
+    spec = importlib.util.spec_from_file_location("whisper_models_under_test", MODULE_PATH)
+    assert spec and spec.loader
+    mod = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(mod)
+    return mod
+
+
+wm = _load_module()
+
+
+class TestResolve(unittest.TestCase):
+    def setUp(self):
+        self.tmp = tempfile.TemporaryDirectory()
+        self.cfg = os.path.join(self.tmp.name, "whisper_models.json")
+        self.reg = wm.WhisperModelRegistry(filepath=self.cfg)
+
+    def tearDown(self):
+        self.tmp.cleanup()
+
+    def test_builtin_resolves_to_systran(self):
+        self.assertEqual(self.reg.resolve("tiny"), "Systran/faster-whisper-tiny")
+        self.assertEqual(self.reg.resolve("large-v3-turbo"), "Systran/faster-whisper-large-v3-turbo")
+
+    def test_passthrough_repo_id(self):
+        # 用户直接把 HF repo_id 当 model_size 传进来（含 "/"）
+        self.assertEqual(self.reg.resolve("SomeOrg/my-whisper-ct2"), "SomeOrg/my-whisper-ct2")
+
+    def test_unknown_raises(self):
+        with self.assertRaises(ValueError):
+            self.reg.resolve("definitely-not-a-model")
+
+    def test_custom_overrides_and_persists(self):
+        self.reg.add_custom_model("myhf", "someorg/whisper-ct2")
+        self.assertEqual(self.reg.resolve("myhf"), "someorg/whisper-ct2")
+        # 新实例读同一文件 → 确认持久化（Docker 下随 config 卷保留）
+        reg2 = wm.WhisperModelRegistry(filepath=self.cfg)
+        self.assertEqual(reg2.resolve("myhf"), "someorg/whisper-ct2")
+
+    def test_custom_can_override_builtin_key_resolution(self):
+        # 自定义优先级高于内置：把 "tiny" 强行指到别的 repo（resolve 层允许；add 层禁止重名）
+        self.reg._write_custom({"tiny": "Other/tiny-ct2"})
+        self.assertEqual(self.reg.resolve("tiny"), "Other/tiny-ct2")
+
+    def test_local_path_resolution_and_detection(self):
+        model_dir = os.path.join(self.tmp.name, "mymodel")
+        os.makedirs(model_dir)
+        self.reg.add_custom_model("local1", model_dir)
+        self.assertEqual(self.reg.resolve("local1"), model_dir)
+        self.assertTrue(wm.is_local_target(self.reg.resolve("local1")))
+
+    def test_bare_existing_dir_passthrough(self):
+        # 没登记、但直接传一个已存在目录 → 直通为本地路径
+        model_dir = os.path.join(self.tmp.name, "bare")
+        os.makedirs(model_dir)
+        self.assertEqual(self.reg.resolve(model_dir), model_dir)
+
+    def test_add_rejects_builtin_collision_and_empty(self):
+        with self.assertRaises(ValueError):
+            self.reg.add_custom_model("tiny", "x/y")  # 与内置重名
+        with self.assertRaises(ValueError):
+            self.reg.add_custom_model("", "x/y")
+        with self.assertRaises(ValueError):
+            self.reg.add_custom_model("ok", "")
+
+    def test_remove(self):
+        self.reg.add_custom_model("tmpm", "a/b")
+        self.assertIn("tmpm", self.reg.get_custom_models())
+        self.reg.remove_custom_model("tmpm")
+        self.assertNotIn("tmpm", self.reg.get_custom_models())
+
+    def test_visible_includes_builtin_and_custom(self):
+        self.reg.add_custom_model("zzz", "a/b")
+        names = self.reg.visible_model_names()
+        self.assertIn("tiny", names)
+        self.assertIn("large-v3", names)
+        self.assertIn("zzz", names)
+
+    def test_is_known(self):
+        self.assertTrue(self.reg.is_known("base"))
+        self.assertTrue(self.reg.is_known("Org/Name"))
+        self.assertFalse(self.reg.is_known("nope-not-real"))
+
+
+class TestHelpers(unittest.TestCase):
+    def test_hf_cache_dirname(self):
+        self.assertEqual(
+            wm.hf_cache_dirname("Systran/faster-whisper-tiny"),
+            "models--Systran--faster-whisper-tiny",
+        )
+        self.assertEqual(wm.hf_cache_dirname("Org/Name"), "models--Org--Name")
+
+    def test_is_local_target(self):
+        self.assertTrue(wm.is_local_target("/abs/path"))
+        self.assertTrue(wm.is_local_target("./rel"))
+        self.assertTrue(wm.is_local_target("~/home/model"))
+        self.assertFalse(wm.is_local_target("Org/Name"))  # repo_id 不是本地路径
+        self.assertFalse(wm.is_local_target(""))
+
+
+if __name__ == "__main__":
+    unittest.main()