diff --git a/qwen3-tts-frontend/src/components/ParamInput.tsx b/qwen3-tts-frontend/src/components/ParamInput.tsx
index 4a37e36..cb2c0da 100644
--- a/qwen3-tts-frontend/src/components/ParamInput.tsx
+++ b/qwen3-tts-frontend/src/components/ParamInput.tsx
@@ -28,8 +28,8 @@ export function ParamInput<T extends FieldValues>({
   max,
 }: ParamInputProps<T>) {
   return (
-    <div className="space-y-2">
-      <div className="flex items-center gap-2">
+    <div className="space-y-1">
+      <div className="flex items-center gap-1.5">
         <Label htmlFor={name}>{label}</Label>
         <TooltipProvider>
           <Tooltip>
diff --git a/qwen3-tts-frontend/src/components/PresetSelector.tsx b/qwen3-tts-frontend/src/components/PresetSelector.tsx
index 1f23775..5ab21d8 100644
--- a/qwen3-tts-frontend/src/components/PresetSelector.tsx
+++ b/qwen3-tts-frontend/src/components/PresetSelector.tsx
@@ -1,5 +1,6 @@
 import { memo, useMemo } from 'react'
 import { Button } from '@/components/ui/button'
+import { Shuffle } from 'lucide-react'
 
 interface Preset {
   label: string
@@ -20,16 +21,33 @@ const PresetSelectorInner = <T extends Preset>({ presets, onSelect }: PresetSele
         variant="outline"
         size="sm"
         onClick={() => onSelect(preset)}
-        className="text-xs md:text-sm px-2.5 md:px-3 h-7 md:h-8"
+        className="text-xs md:text-sm px-2 h-6 md:h-7"
       >
         {preset.label}
       </Button>
     ))
   }, [presets, onSelect])
 
+  const handleRandomSelect = () => {
+    const randomIndex = Math.floor(Math.random() * presets.length)
+    onSelect(presets[randomIndex])
+  }
+
   return (
-    <div className="flex flex-wrap gap-1.5 md:gap-2 mt-1.5 md:mt-2">
-      {presetButtons}
+    <div className="flex items-center gap-2 mt-1">
+      <div className="flex flex-wrap gap-1 flex-1">
+        {presetButtons}
+      </div>
+      <Button
+        type="button"
+        variant="ghost"
+        size="sm"
+        onClick={handleRandomSelect}
+        className="h-6 md:h-7 px-2 flex-shrink-0"
+        title="随机选择"
+      >
+        <Shuffle className="h-3.5 w-3.5" />
+      </Button>
     </div>
   )
 }
diff --git a/qwen3-tts-frontend/src/components/tts/CustomVoiceForm.tsx b/qwen3-tts-frontend/src/components/tts/CustomVoiceForm.tsx
index 1ba81ac..f756fa5 100644
--- a/qwen3-tts-frontend/src/components/tts/CustomVoiceForm.tsx
+++ b/qwen3-tts-frontend/src/components/tts/CustomVoiceForm.tsx
@@ -116,8 +116,8 @@ const CustomVoiceForm = forwardRef<CustomVoiceFormHandle>((_props, ref) => {
   }, [currentJob?.id, currentJob?.audio_url])
 
   return (
-    <form onSubmit={handleSubmit(onSubmit)} className="space-y-4 md:space-y-6">
-      <div className="space-y-1.5 md:space-y-2">
+    <form onSubmit={handleSubmit(onSubmit)} className="space-y-3">
+      <div className="space-y-1">
         <Label htmlFor="language">语言</Label>
         <Select
           value={watch('language')}
@@ -139,7 +139,7 @@ const CustomVoiceForm = forwardRef<CustomVoiceFormHandle>((_props, ref) => {
         )}
       </div>
 
-      <div className="space-y-1.5 md:space-y-2">
+      <div className="space-y-1">
         <Label htmlFor="speaker">发音人</Label>
         <Select
           value={watch('speaker')}
@@ -161,26 +161,24 @@ const CustomVoiceForm = forwardRef<CustomVoiceFormHandle>((_props, ref) => {
         )}
       </div>
 
-      <div className="space-y-1.5 md:space-y-2">
+      <div className="space-y-1">
         <Label htmlFor="text">合成文本</Label>
         <Textarea
           {...register('text')}
           placeholder="输入要合成的文本..."
-          rows={2}
-          className="min-h-[60px] md:min-h-[96px]"
+          className="min-h-[40px] md:min-h-[60px]"
         />
         {errors.text && (
           <p className="text-sm text-destructive">{errors.text.message}</p>
         )}
       </div>
 
-      <div className="space-y-1.5 md:space-y-2">
+      <div className="space-y-1">
         <Label htmlFor="instruct">情绪指导（可选）</Label>
         <Textarea
           {...register('instruct')}
           placeholder="例如：温柔体贴，语速平缓，充满关怀"
-          rows={2}
-          className="min-h-[60px] md:min-h-[80px]"
+          className="min-h-[40px] md:min-h-[60px]"
         />
         <PresetSelector
           presets={PRESET_INSTRUCTS}
@@ -198,12 +196,12 @@ const CustomVoiceForm = forwardRef<CustomVoiceFormHandle>((_props, ref) => {
 
       <Collapsible open={advancedOpen} onOpenChange={setAdvancedOpen}>
         <CollapsibleTrigger asChild>
-          <Button type="button" variant="ghost" className="w-full">
+          <Button type="button" variant="ghost" className="w-full py-1.5">
             高级选项
             <ChevronDown className="ml-2 h-4 w-4" />
           </Button>
         </CollapsibleTrigger>
-        <CollapsibleContent className="space-y-3 md:space-y-4 pt-3 md:pt-4">
+        <CollapsibleContent className="space-y-2 pt-2">
           <ParamInput
             name="max_new_tokens"
             label={ADVANCED_PARAMS_INFO.max_new_tokens.label}
diff --git a/qwen3-tts-frontend/src/components/tts/VoiceCloneForm.tsx b/qwen3-tts-frontend/src/components/tts/VoiceCloneForm.tsx
index e7be332..63bce2d 100644
--- a/qwen3-tts-frontend/src/components/tts/VoiceCloneForm.tsx
+++ b/qwen3-tts-frontend/src/components/tts/VoiceCloneForm.tsx
@@ -101,14 +101,13 @@ function VoiceCloneForm() {
   }, [currentJob?.id, currentJob?.audio_url])
 
   return (
-    <form onSubmit={handleSubmit(onSubmit)} className="space-y-4 md:space-y-6">
-      <div className="space-y-1.5 md:space-y-2">
+    <form onSubmit={handleSubmit(onSubmit)} className="space-y-3">
+      <div className="space-y-1">
         <Label htmlFor="ref_text">参考文稿（可选）</Label>
         <Textarea
           {...register('ref_text')}
           placeholder="参考音频对应的文本..."
-          rows={2}
-          className="min-h-[60px] md:min-h-[80px]"
+          className="min-h-[40px] md:min-h-[60px]"
         />
         <PresetSelector
           presets={PRESET_REF_TEXTS}
@@ -119,7 +118,7 @@ function VoiceCloneForm() {
         )}
       </div>
 
-      <div className="space-y-1.5 md:space-y-2">
+      <div className="space-y-1">
         <Label htmlFor="ref_audio">参考音频</Label>
         <Controller
           name="ref_audio"
@@ -134,7 +133,7 @@ function VoiceCloneForm() {
         />
       </div>
 
-      <div className="space-y-1.5 md:space-y-2">
+      <div className="space-y-1">
         <Label htmlFor="language">语言（可选）</Label>
         <Select
           value={watch('language')}
@@ -153,13 +152,12 @@ function VoiceCloneForm() {
         </Select>
       </div>
 
-      <div className="space-y-1.5 md:space-y-2">
+      <div className="space-y-1">
         <Label htmlFor="text">合成文本</Label>
         <Textarea
           {...register('text')}
           placeholder="输入要合成的文本..."
-          rows={2}
-          className="min-h-[60px] md:min-h-[96px]"
+          className="min-h-[40px] md:min-h-[60px]"
         />
         <PresetSelector
           presets={PRESET_REF_TEXTS}
@@ -170,7 +168,7 @@ function VoiceCloneForm() {
         )}
       </div>
 
-      <div className="flex items-center space-x-4">
+      <div className="flex items-center space-x-3">
         <div className="flex items-center space-x-2">
           <Controller
             name="x_vector_only_mode"
@@ -208,12 +206,12 @@ function VoiceCloneForm() {
 
       <Collapsible open={advancedOpen} onOpenChange={setAdvancedOpen}>
         <CollapsibleTrigger asChild>
-          <Button type="button" variant="ghost" className="w-full">
+          <Button type="button" variant="ghost" className="w-full py-1.5">
             高级选项
             <ChevronDown className="ml-2 h-4 w-4" />
           </Button>
         </CollapsibleTrigger>
-        <CollapsibleContent className="space-y-3 md:space-y-4 pt-3 md:pt-4">
+        <CollapsibleContent className="space-y-2 pt-2">
           <ParamInput
             name="max_new_tokens"
             label={ADVANCED_PARAMS_INFO.max_new_tokens.label}
diff --git a/qwen3-tts-frontend/src/components/tts/VoiceDesignForm.tsx b/qwen3-tts-frontend/src/components/tts/VoiceDesignForm.tsx
index 41681f3..304eada 100644
--- a/qwen3-tts-frontend/src/components/tts/VoiceDesignForm.tsx
+++ b/qwen3-tts-frontend/src/components/tts/VoiceDesignForm.tsx
@@ -107,8 +107,8 @@ const VoiceDesignForm = forwardRef<VoiceDesignFormHandle>((_props, ref) => {
   }, [currentJob?.id, currentJob?.audio_url])
 
   return (
-    <form onSubmit={handleSubmit(onSubmit)} className="space-y-4 md:space-y-6">
-      <div className="space-y-1.5 md:space-y-2">
+    <form onSubmit={handleSubmit(onSubmit)} className="space-y-3">
+      <div className="space-y-1">
         <Label htmlFor="language">语言</Label>
         <Select
           value={watch('language')}
@@ -130,26 +130,24 @@ const VoiceDesignForm = forwardRef<VoiceDesignFormHandle>((_props, ref) => {
         )}
       </div>
 
-      <div className="space-y-1.5 md:space-y-2">
+      <div className="space-y-1">
         <Label htmlFor="text">合成文本</Label>
         <Textarea
           {...register('text')}
           placeholder="输入要合成的文本..."
-          rows={2}
-          className="min-h-[60px] md:min-h-[96px]"
+          className="min-h-[40px] md:min-h-[60px]"
         />
         {errors.text && (
           <p className="text-sm text-destructive">{errors.text.message}</p>
         )}
       </div>
 
-      <div className="space-y-1.5 md:space-y-2">
+      <div className="space-y-1">
         <Label htmlFor="instruct">音色描述</Label>
         <Textarea
           {...register('instruct')}
           placeholder="例如：成熟男性,低沉磁性,充满权威感"
-          rows={2}
-          className="min-h-[60px] md:min-h-[80px]"
+          className="min-h-[40px] md:min-h-[60px]"
         />
         <PresetSelector
           presets={PRESET_VOICE_DESIGNS}
@@ -167,12 +165,12 @@ const VoiceDesignForm = forwardRef<VoiceDesignFormHandle>((_props, ref) => {
 
       <Collapsible open={advancedOpen} onOpenChange={setAdvancedOpen}>
         <CollapsibleTrigger asChild>
-          <Button type="button" variant="ghost" className="w-full">
+          <Button type="button" variant="ghost" className="w-full py-1.5">
             高级选项
             <ChevronDown className="ml-2 h-4 w-4" />
           </Button>
         </CollapsibleTrigger>
-        <CollapsibleContent className="space-y-3 md:space-y-4 pt-3 md:pt-4">
+        <CollapsibleContent className="space-y-2 pt-2">
           <ParamInput
             name="max_new_tokens"
             label={ADVANCED_PARAMS_INFO.max_new_tokens.label}
diff --git a/qwen3-tts-frontend/src/pages/Home.tsx b/qwen3-tts-frontend/src/pages/Home.tsx
index c21c431..f5d7a54 100644
--- a/qwen3-tts-frontend/src/pages/Home.tsx
+++ b/qwen3-tts-frontend/src/pages/Home.tsx
@@ -71,7 +71,7 @@ function Home() {
               </Tabs>
             </CardHeader>
 
-            <CardContent className="pt-4 md:pt-6 px-3 md:px-6">
+            <CardContent className="pt-0 px-3 md:px-6">
               <Tabs value={currentTab}>
                 <TabsContent value="custom-voice" className="mt-0">
                   <Suspense fallback={<FormSkeleton />}>