Upload 15 files

LTX2.3/API issues-API问题办法.bat

@@ -0,0 +1,43 @@
+@echo off
+chcp 65001 >nul
+title LTX 本地显卡模式修复工具
+
+echo ========================================
+echo    LTX 本地显卡模式修复工具
+echo ========================================
+echo.
+
+:: 检查管理员权限
+net session >nul 2>&1
+if %errorlevel% neq 0 (
+    echo [!] 请右键选择"以管理员身份运行"此脚本
+    pause
+    exit /b 1
+)
+
+echo [1/2] 正在修改 VRAM 阈值...
+set "policy_file=C:\Program Files\LTX Desktop\resources\backend\runtime_config\runtime_policy.py"
+
+if exist "%policy_file%" (
+    powershell -Command "(Get-Content '%policy_file%') -replace 'vram_gb < 31', 'vram_gb < 6' | Set-Content '%policy_file%'"
+    echo     ^_^ VRAM 阈值已修改为 6GB
+) else (
+    echo     [!] 未找到 runtime_policy.py，请确认 LTX Desktop 已安装
+)
+
+echo.
+echo [2/2] 正在清空 API Key...
+set "settings_file=%USERPROFILE%\AppData\Local\LTXDesktop\settings.json"
+
+if exist "%settings_file%" (
+    powershell -Command "$content = Get-Content '%settings_file%' -Raw; $content = $content -replace '\"fal_api_key\": \"[^\"]*\"', '\"fal_api_key\": \"\"'; Set-Content -Path '%settings_file%' -Value $content -NoNewline"
+    echo     ^_^ API Key 已清空
+) else (
+    echo     [!] 未找到 settings.json，首次运行后会自动创建
+)
+
+echo.
+echo ========================================
+echo    修复完成！请重启 LTX Desktop
+echo ========================================
+pause

LTX2.3/API issues-API问题办法.txt

@@ -0,0 +1,50 @@
+1. 复制LTX桌面版的快捷方式到LTX_Shortcut
+
+2. 运行run.bat
+----
+1. Copy the LTX desktop shortcut to LTX_Shortcut
+
+2. Run run.bat
+----
+
+
+
+【问题描述 / Problem】
+系统强制使用 FAL API 生成图片，即使本地有 GPU 可用。
+System forces FAL API generation even when local GPU is available.
+
+【原因 / Cause】
+LTX 强制要求 GPU 有 31GB VRAM 才会使用本地显卡，低于此值会强制走 API 模式。
+LTX requires 31GB VRAM to use local GPU. Below this, it forces API mode.
+
+================================================================================
+【修复方法 / Fix Method】
+================================================================================
+
+运行: API issues.bat.bat (以管理员身份)
+Run: API issues.bat.bat (as Administrator)
+
+================================================================================
+================================================================================
+
+【或者手动 / Or Manual】
+
+1. 修改 VRAM 阈值 / Modify VRAM Threshold
+   文件路径 / File: C:\Program Files\LTX Desktop\resources\backend\runtime_config\runtime_policy.py
+   第16行 / Line 16:
+   原 / Original: return vram_gb < 31
+   改为 / Change:  return vram_gb < 6
+
+2. 清空 API Key / Clear API Key
+   文件路径 / File: C:\Users\<用户名>\AppData\Local\LTXDesktop\settings.json
+   原 / Original: "fal_api_key": "xxxxx"
+   改为 / Change:  "fal_api_key": ""
+
+【说明 / Note】
+- VRAM 阈值改为 6GB，意味着 6GB 及以上显存都会使用本地显卡
+- VRAM threshold set to 6GB means 6GB+ VRAM will use local GPU
+- 清空 fal_api_key 避免系统误判为已配置 API
+- Clear fal_api_key to avoid system thinking API is configured
+- 修改后重启程序即可生效
+- Restart LTX Desktop after changes
+================================================================================

LTX2.3/UI/index.css

@@ -0,0 +1,298 @@
+:root {
+            --accent:      #2563EB;     /* Refined blue – not too bright, not purple */
+            --accent-hover:#3B82F6;
+            --accent-dim:  rgba(37,99,235,0.14);
+            --accent-ring: rgba(37,99,235,0.35);
+            --bg:          #111113;
+            --panel:       #18181B;
+            --panel-2:     #1F1F23;
+            --item:        rgba(255,255,255,0.035);
+            --border:      rgba(255,255,255,0.08);
+            --border-2:    rgba(255,255,255,0.05);
+            --text-dim:    #71717A;
+            --text-sub:    #A1A1AA;
+            --text:        #FAFAFA;
+        }
+
+        * { box-sizing: border-box; -webkit-font-smoothing: antialiased; min-width: 0; }
+        body {
+            background: var(--bg); margin: 0; color: var(--text);
+            font-family: -apple-system, "SF Pro Display", "Segoe UI", sans-serif;
+            display: flex; height: 100vh; overflow: hidden;
+            font-size: 13px; line-height: 1.5;
+        }
+
+        .sidebar {
+            width: 460px; min-width: 460px;
+            background: var(--panel);
+            border-right: 1px solid var(--border);
+            display: flex; flex-direction: column; z-index: 20;
+            overflow-y: auto; overflow-x: hidden;
+        }
+
+        /* Scrollbar */
+        ::-webkit-scrollbar { width: 5px; height: 5px; }
+        ::-webkit-scrollbar-track { background: transparent; }
+        ::-webkit-scrollbar-thumb { background: rgba(255,255,255,0.08); border-radius: 10px; }
+        ::-webkit-scrollbar-thumb:hover { background: rgba(255,255,255,0.18); }
+
+        .sidebar-header { padding: 24px 24px 4px; }
+        .sidebar-section { padding: 8px 24px 18px; border-bottom: 1px solid var(--border); }
+
+        .setting-group {
+            background: rgba(255,255,255,0.025);
+            border: 1px solid var(--border-2);
+            border-radius: 10px;
+            padding: 14px;
+            margin-bottom: 12px;
+        }
+        .group-title {
+            font-size: 10px; color: var(--text-dim); font-weight: 700;
+            text-transform: uppercase; letter-spacing: 0.7px;
+            margin-bottom: 12px;
+            padding-bottom: 5px;
+            border-bottom: 1px solid var(--border-2);
+        }
+
+        /* Mode Tabs */
+        .tabs {
+            display: flex; gap: 4px; margin-bottom: 14px;
+            background: rgba(255,255,255,0.04);
+            padding: 4px; border-radius: 10px;
+            border: 1px solid var(--border-2);
+        }
+        .tab {
+            flex: 1; padding: 9px 0; text-align: center; border-radius: 7px;
+            cursor: pointer; font-size: 12px; color: var(--text-dim);
+            transition: all 0.2s; font-weight: 600;
+            display: flex; align-items: center; justify-content: center;
+        }
+        .tab.active { background: var(--accent); color: #fff; box-shadow: 0 1px 6px rgba(10,132,255,0.45); }
+        .tab:hover:not(.active) { background: rgba(255,255,255,0.06); color: var(--text); }
+
+        .label-group { display: flex; justify-content: space-between; align-items: center; margin-bottom: 6px; }
+        label { display: block; font-size: 11px; color: var(--text-dim); font-weight: 600; text-transform: uppercase; letter-spacing: 0.5px; margin-bottom: 6px; }
+        .val-badge { font-size: 11px; color: var(--accent); font-family: "SF Mono", ui-monospace, monospace; font-weight: 600; }
+
+        input[type="text"], input[type="number"], select, textarea {
+            width: 100%; background: var(--panel-2);
+            border: 1px solid var(--border);
+            border-radius: 7px; color: var(--text);
+            padding: 8px 11px; font-size: 12.5px; outline: none; margin-bottom: 9px;
+            /* Only transition border/shadow – NOT background-image to prevent arrow flicker */
+            transition: border-color 0.15s, box-shadow 0.15s;
+        }
+        input:focus, select:focus, textarea:focus {
+            border-color: var(--accent);
+            box-shadow: 0 0 0 2px var(--accent-ring);
+        }
+        select {
+            -webkit-appearance: none; -moz-appearance: none; appearance: none;
+            /* Stable grey arrow – no background shorthand so it won't animate */
+            background-color: var(--panel-2);
+            background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='12' viewBox='0 0 24 24' fill='none' stroke='%2371717A' stroke-width='2.5' stroke-linecap='round' stroke-linejoin='round'%3E%3Cpolyline points='6 9 12 15 18 9'/%3E%3C/svg%3E");
+            background-repeat: no-repeat;
+            background-position: right 10px center;
+            background-size: 12px;
+            padding-right: 28px;
+            cursor: pointer;
+            /* Explicitly do NOT transition background properties */
+            transition: border-color 0.15s, box-shadow 0.15s;
+        }
+        select:focus { background-color: var(--panel-2); }
+        select option { background: #27272A; color: var(--text); }
+        textarea { resize: vertical; min-height: 78px; font-family: inherit; }
+
+        .slider-container { display: flex; align-items: center; gap: 12px; margin-bottom: 14px; }
+        input[type="range"] { flex: 1; accent-color: var(--accent); height: 4px; cursor: pointer; border-radius: 2px; }
+
+        .upload-zone {
+            border: 1px dashed var(--border); border-radius: 10px;
+            padding: 18px 10px; text-align: center; cursor: pointer;
+            background: rgba(255,255,255,0.01); margin-bottom: 10px; position: relative;
+            transition: all 0.2s;
+        }
+        .upload-zone:hover, .upload-zone.dragover { background: var(--accent-dim); border-color: var(--accent); }
+        .upload-icon { font-size: 18px; margin-bottom: 6px; opacity: 0.45; }
+        .upload-text { font-size: 11px; color: var(--text); }
+        .upload-hint { font-size: 10px; color: var(--text-dim); margin-top: 3px; }
+        .preview-thumb { width: 100%; height: auto; max-height: 100px; object-fit: contain; border-radius: 8px; display: none; margin-top: 10px; }
+        .clear-img-overlay {
+            position: absolute; top: 8px; right: 8px; background: rgba(255,59,48,0.85); color: white;
+            width: 20px; height: 20px; border-radius: 10px; display: none; align-items: center; justify-content: center;
+            font-size: 11px; cursor: pointer; z-index: 5;
+        }
+
+        .btn-outline {
+            background: var(--panel-2);
+            border: 1px solid var(--border);
+            color: var(--text-sub); padding: 5px 12px; border-radius: 7px;
+            font-size: 11.5px; font-weight: 600; cursor: pointer;
+            transition: background 0.15s, border-color 0.15s, color 0.15s;
+            display: inline-flex; align-items: center; justify-content: center; gap: 5px;
+            white-space: nowrap;
+        }
+        .btn-outline:hover:not(:disabled) { background: rgba(255,255,255,0.08); color: var(--text); border-color: rgba(255,255,255,0.18); }
+        .btn-outline:active { opacity: 0.7; }
+        .btn-outline:disabled { opacity: 0.3; cursor: not-allowed; }
+
+        .btn-icon {
+            padding: 5px; background: transparent; border: none; color: var(--text-dim);
+            border-radius: 6px; cursor: pointer; display: flex; align-items: center; justify-content: center;
+            transition: color 0.15s, background 0.15s;
+        }
+        .btn-icon:hover { color: var(--text-sub); background: rgba(255,255,255,0.07); }
+
+        .btn-primary {
+            width: 100%; padding: 13px;
+            background: var(--accent); border: none;
+            border-radius: 9px; color: #fff; font-weight: 700; font-size: 13.5px;
+            letter-spacing: 0.2px; cursor: pointer; margin-top: 14px;
+            transition: background 0.15s;
+        }
+        .btn-primary:hover:not(:disabled) { background: var(--accent-hover); }
+        .btn-primary:active { opacity: 0.82; }
+        .btn-primary:disabled { background: rgba(255,255,255,0.08); color: var(--text-dim); cursor: not-allowed; }
+
+        .btn-danger {
+            width: 100%; padding: 12px; background: #DC2626; border: none;
+            border-radius: 9px; color: #fff; font-weight: 700; font-size: 13.5px;
+            cursor: pointer; margin-top: 8px; display: none; transition: background 0.15s;
+        }
+        .btn-danger:hover { background: #EF4444; }
+
+        /* Workspace */
+        .workspace { flex: 1; display: flex; flex-direction: column; background: #0A0A0A; position: relative; overflow: hidden; }
+        .viewer { flex: 2; display: flex; align-items: center; justify-content: center; padding: 16px; background: #0A0A0A; position: relative; min-height: 40vh; }
+        .monitor {
+            width: 100%; height: 100%; max-width: 1650px; border-radius: 10px; border: 1px solid var(--border);
+            overflow: hidden; position: relative; background: #070707;
+            display: flex; align-items: center; justify-content: center;
+            background-image: radial-gradient(rgba(255,255,255,0.02) 1px, transparent 1px);
+            background-size: 18px 18px;
+        }
+        .monitor img, .monitor video {
+            width: auto; height: auto; max-width: 100%; max-height: 100%;
+            object-fit: contain; display: none; z-index: 2; border-radius: 3px;
+        }
+
+        .progress-container { position: absolute; bottom: 0; left: 0; width: 100%; height: 2px; background: var(--border-2); z-index: 10; }
+        #progress-fill { width: 0%; height: 100%; background: var(--accent); transition: width 0.5s; }
+        #loading-txt { font-size: 12px; color: var(--text-sub); font-weight: 600; z-index: 5; position: absolute; display: none; }
+
+
+
+        .spinner {
+            width: 12px; height: 12px;
+            border: 2px solid rgba(255,255,255,0.2);
+            border-top-color: currentColor;
+            border-radius: 50%;
+            animation: spin 1s linear infinite;
+        }
+        @keyframes spin { to { transform: rotate(360deg); } }
+        
+        .loading-card {
+            display: flex; align-items: center; justify-content: center;
+            flex-direction: column; gap: 6px; color: var(--text-dim); font-size: 10px;
+            background: rgba(37,99,235,0.07) !important;
+            border-color: rgba(37,99,235,0.3) !important;
+        }
+        .loading-card .spinner { width: 28px; height: 28px; border-width: 3px; color: var(--accent); }
+        .loading-card:hover { background: rgba(37,99,235,0.14) !important; border-color: var(--accent) !important; }
+
+        .library { flex: 1.5; border-top: 1px solid var(--border); padding: 14px 20px; display: flex; flex-direction: column; background: #0F0F11; overflow-y: hidden; }
+        #log-container { flex: 1; overflow-y: auto; padding-right: 4px; }
+        #log { font-family: ui-monospace, "SF Mono", monospace; font-size: 10.5px; color: var(--text-dim); line-height: 1.7; }
+
+        /* History wrapper: scrollable area for thumbnails only */
+        #history-wrapper {
+            flex: 1;
+            overflow-y: auto;
+            min-height: 110px;   /* always show at least one row */
+            padding-right: 4px;
+        }
+        #history-container {
+            display: grid;
+            grid-template-columns: repeat(auto-fill, minmax(150px, 1fr));
+            justify-content: start;
+            gap: 10px; align-content: flex-start;
+            padding-bottom: 4px;
+        }
+        /* Pagination row: hidden, using infinite scroll instead */
+        #pagination-bar {
+            display: none;
+        }
+
+        .history-card {
+            width: 100%; max-width: 200px; aspect-ratio: 16 / 9;
+            background: #1A1A1E; border-radius: 7px;
+            overflow: hidden; border: 1px solid var(--border);
+            cursor: pointer; position: relative; transition: border-color 0.15s, transform 0.15s;
+        }
+        .history-card:hover { border-color: var(--accent); transform: translateY(-1px); }
+        .history-card img, .history-card video { width: 100%; height: 100%; object-fit: cover; }
+        .history-type-badge {
+            position: absolute; top: 5px; left: 5px; font-size: 8px; padding: 1px 5px; border-radius: 3px;
+            background: rgba(0,0,0,0.8); color: var(--text-sub); border: 1px solid rgba(255,255,255,0.06);
+            z-index: 2; font-weight: 700; letter-spacing: 0.4px;
+        }
+        .history-delete-btn {
+            position: absolute; top: 5px; right: 5px; width: 20px; height: 20px;
+            border-radius: 50%; border: none; background: rgba(255,50,50,0.8); color: #fff;
+            font-size: 10px; cursor: pointer; z-index: 3; display: flex; align-items: center; justify-content: center;
+            opacity: 0; transition: opacity 0.2s;
+        }
+        .history-card:hover .history-delete-btn { opacity: 1; }
+        .history-delete-btn:hover { background: rgba(255,0,0,0.9); }
+
+        .vram-bar { width: 160px; height: 5px; background: rgba(255,255,255,0.08); border-radius: 999px; overflow: hidden; display: inline-block; vertical-align: middle; }
+        .vram-used { height: 100%; background: var(--accent); width: 0%; transition: width 0.5s; }
+
+        .sub-mode-toggle { display: flex; background: var(--panel-2); border-radius: 7px; padding: 3px; border: 1px solid var(--border); }
+        .sub-mode-btn { flex: 1; padding: 6px 0; border-radius: 5px; border: none; background: transparent; font-size: 11.5px; color: var(--text-dim); font-weight: 600; cursor: pointer; transition: background 0.15s, color 0.15s; }
+        .sub-mode-btn.active { background: var(--accent); color: #fff; }
+        .sub-mode-btn:hover:not(.active) { background: rgba(255,255,255,0.05); color: var(--text-sub); }
+
+        .vid-section { display: none; margin-top: 12px; }
+        .vid-section.active-section { display: block; animation: fadeIn 0.25s ease; }
+        @keyframes fadeIn { from { opacity: 0; transform: translateY(4px); } to { opacity: 1; transform: translateY(0); } }
+
+        /* Status indicator */
+        @keyframes breathe-orange {
+            0%,100% { box-shadow: 0 0 4px #FF9F0A; opacity: 0.7; }
+            50%      { box-shadow: 0 0 10px #FF9F0A; opacity: 1; }
+        }
+        .indicator-busy    { background: #FF9F0A !important; animation: breathe-orange 1.6s infinite ease-in-out !important; box-shadow: none !important; transition: all 0.3s; }
+        .indicator-ready   { background: #30D158 !important; box-shadow: 0 0 8px rgba(48,209,88,0.6) !important; animation: none !important; transition: all 0.3s; }
+        .indicator-offline { background: #636366 !important; box-shadow: none !important; animation: none !important; transition: all 0.3s; }
+
+        .res-preview-tag { font-size: 11px; color: var(--accent); margin-bottom: 10px; font-family: ui-monospace, monospace; }
+        .top-status { display: flex; justify-content: space-between; font-size: 12px; color: var(--text-dim); margin-bottom: 8px; align-items: center; }
+        .checkbox-container { display: flex; align-items: center; gap: 8px; cursor: pointer; background: rgba(255,255,255,0.02); padding: 10px; border-radius: 8px; border: 1px solid var(--border-2); }
+        .checkbox-container input { width: 15px; height: 15px; accent-color: var(--accent); cursor: pointer; margin: 0; }
+        .checkbox-container label { margin-bottom: 0; cursor: pointer; text-transform: none; color: var(--text); }
+        .flex-row { display: flex; gap: 10px; }
+        .flex-1 { flex: 1; min-width: 0; }
+
+        @media (max-width: 1024px) {
+            body { flex-direction: column; overflow-y: auto; }
+            .sidebar { width: 100%; min-width: 100%; border-right: none; border-bottom: 1px solid var(--border); height: auto; overflow: visible; }
+            .workspace { height: auto; min-height: 100vh; overflow: visible; }
+        }
+:root {
+    --plyr-color-main: #3F51B5;
+    --plyr-video-control-background-hover: rgba(255,255,255,0.1);
+    --plyr-control-radius: 6px;
+    --plyr-player-width: 100%;
+}
+.plyr {
+    border-radius: 8px;
+    overflow: hidden;
+    width: 100%;
+    height: 100%;
+}
+.plyr--video .plyr__controls {
+    background: linear-gradient(rgba(0,0,0,0), rgba(0,0,0,0.8));
+    padding: 20px 15px 15px 15px;
+}
+

LTX2.3/UI/index.html

@@ -0,0 +1,302 @@
+<!DOCTYPE html>
+<html lang="zh-CN">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>LTX-2 | Multi-GPU Cinematic Studio</title>
+        <link rel="stylesheet" href="index.css">
+        <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/plyr/3.7.8/plyr.css" />
+</head>
+<body>
+
+<aside class="sidebar">
+    <div class="sidebar-header">
+        <div style="display: flex; align-items: center; justify-content: space-between; margin-bottom: 12px;">
+            <div style="display: flex; align-items: center; gap: 10px;">
+                <div id="sys-indicator" class="indicator-ready" style="width: 12px; height: 12px; border-radius: 50%;"></div>
+                <span style="font-weight: 800; font-size: 18px;">LTX-2 STUDIO</span>
+            </div>
+            <div style="display: flex; gap: 8px; align-items: center;">
+                <button id="langBtn" onclick="toggleLang()" class="btn-icon" title="切换语言">
+                    <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><circle cx="12" cy="12" r="10"></circle><line x1="2" y1="12" x2="22" y2="12"></line><path d="M12 2a15.3 15.3 0 0 1 4 10 15.3 15.3 0 0 1-4 10 15.3 15.3 0 0 1-4-10 15.3 15.3 0 0 1 4-10z"></path></svg>
+                </button>
+                <button id="clearGpuBtn" onclick="clearGpu()" class="btn-outline" data-lang-zh="释放显存" data-lang-en="Clear VRAM">释放显存</button>
+            </div>
+        </div>
+        
+        <div class="top-status" style="margin-bottom: 5px;">
+            <div style="display: flex; align-items: center; gap: 8px;">
+                <span id="sys-status" style="font-weight:bold; color: var(--text-dim); font-size: 12px;" data-lang-zh="正在扫描 GPU..." data-lang-en="Scanning GPU...">正在扫描 GPU...</span>
+            </div>
+            
+            <button onclick="const el = document.getElementById('sys-settings'); el.style.display = el.style.display === 'none' ? 'block' : 'none';" class="btn-icon" title="系统高级设置">
+                <svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><circle cx="12" cy="12" r="3"></circle><path d="M19.4 15a1.65 1.65 0 0 0 .33 1.82l.06.06a2 2 0 0 1 0 2.83 2 2 0 0 1-2.83 0l-.06-.06a1.65 1.65 0 0 0-1.82-.33 1.65 1.65 0 0 0-1 1.51V21a2 2 0 0 1-2 2 2 2 0 0 1-2-2v-.09A1.65 1.65 0 0 0 9 19.4a1.65 1.65 0 0 0-1.82.33l-.06.06a2 2 0 0 1-2.83 0 2 2 0 0 1 0-2.83l.06-.06a1.65 1.65 0 0 0 .33-1.82 1.65 1.65 0 0 0-1.51-1H3a2 2 0 0 1-2-2 2 2 0 0 1 2-2h.09A1.65 1.65 0 0 0 4.6 9a1.65 1.65 0 0 0-.33-1.82l-.06-.06a2 2 0 0 1 0-2.83 2 2 0 0 1 2.83 0l.06.06a1.65 1.65 0 0 0 1.82.33H9a1.65 1.65 0 0 0 1-1.51V3a2 2 0 0 1 2-2 2 2 0 0 1 2 2v.09a1.65 1.65 0 0 0 1 1.51 1.65 1.65 0 0 0 1.82-.33l.06-.06a2 2 0 0 1 2.83 0 2 2 0 0 1 0 2.83l-.06.06a1.65 1.65 0 0 0-.33 1.82V9a1.65 1.65 0 0 0 1.51 1H21a2 2 0 0 1 2 2 2 2 0 0 1-2 2h-.09a1.65 1.65 0 0 0-1.51 1z"></path></svg>
+            </button>
+
+        </div>
+        
+        <div style="font-size: 11px; color: var(--text-dim); margin-bottom: 20px; display: flex; align-items: center; width: 100%;">
+            <div class="vram-bar" style="width: 120px; min-width: 120px; margin-top: 0; margin-right: 12px;"><div class="vram-used" id="vram-fill"></div></div> 
+            <span id="vram-text" style="font-variant-numeric: tabular-nums; flex-shrink: 0; text-align: right;">0/32 GB</span>
+            <span id="gpu-name" style="display: none;"></span>
+        </div>
+        
+        <div id="sys-settings" style="display: none; padding: 14px; background: rgba(0,0,0,0.4) !important; border-radius: 12px; border: 1px solid rgba(255,255,255,0.1); margin-bottom: 15px; box-shadow: 0 4px 15px rgba(0,0,0,0.5); backdrop-filter: blur(10px);">
+            <div style="font-size: 13px; font-weight: bold; margin-bottom: 12px; color: #fff;" data-lang-zh="高级设置" data-lang-en="Advanced Settings">高级设置</div>
+            
+            <label style="font-size: 11px; margin-bottom: 6px;" data-lang-zh="工作设备" data-lang-en="GPU Device">工作设备</label>
+            <select id="gpu-selector" onchange="switchGpu(this.value)" style="margin-bottom: 12px; font-size: 11px; padding: 6px;">
+                <option value="" data-lang-zh="正在检测 GPU..." data-lang-en="Detecting GPU...">正在检测 GPU...</option>
+            </select>
+            
+            <label style="font-size: 11px; margin-bottom: 6px;" data-lang-zh="输出路径" data-lang-en="Output Path">输出路径</label>
+            <div style="display:flex; gap:6px;">
+                <input type="text" id="global-out-dir" onchange="setOutputDir()" placeholder="默认: LTXDesktop/outputs" style="margin-bottom: 0; padding: 6px 8px; font-size: 11px;">
+                <button class="btn-outline" style="padding: 6px;" onclick="browseOutputDir()" title="浏览本地目录">
+                    <svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 19a2 2 0 0 1-2 2H4a2 2 0 0 1-2-2V5a2 2 0 0 1 2-2h5l2 3h9a2 2 0 0 1 2 2z"></path></svg>
+                </button>
+            </div>
+            <div style="font-size: 10px; color: var(--text-dim); margin-top: 8px; line-height: 1.4;" data-lang-zh="系统默认会在 C 盘保留输出文件。您可以浏览更换目录。" data-lang-en="Default output in C drive. Browse to change path.">系统默认会在 C 盘保留输出文件。您可以浏览更换目录。</div>
+        </div>
+    </div>
+
+    <div class="sidebar-section">
+        <div class="tabs">
+            <div id="tab-video" class="tab active" onclick="switchMode('video')">
+                <svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" style="margin-right: 6px;"><rect x="2" y="2" width="20" height="20" rx="2.18" ry="2.18"></rect><line x1="7" y1="2" x2="7" y2="22"></line><line x1="17" y1="2" x2="17" y2="22"></line><line x1="2" y1="12" x2="22" y2="12"></line><line x1="2" y1="7" x2="7" y2="7"></line><line x1="2" y1="17" x2="7" y2="17"></line><line x1="17" y1="17" x2="22" y2="17"></line><line x1="17" y1="7" x2="22" y2="7"></line></svg>
+                <span data-lang-zh="视频生成" data-lang-en="Video">视频生成</span>
+            </div>
+            <div id="tab-upscale" class="tab" onclick="switchMode('upscale')">
+                <svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" style="margin-right: 6px;"><polygon points="13 2 3 14 12 14 11 22 21 10 12 10 13 2"></polygon></svg>
+                <span data-lang-zh="视频增强" data-lang-en="Upscale">视频增强</span>
+            </div>
+            <div id="tab-image" class="tab" onclick="switchMode('image')">
+                <svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" style="margin-right: 6px;"><rect x="3" y="3" width="18" height="18" rx="2" ry="2"></rect><circle cx="8.5" cy="8.5" r="1.5"></circle><polyline points="21 15 16 10 5 21"></polyline></svg>
+                <span data-lang-zh="图像生成" data-lang-en="Image">图像生成</span>
+            </div>
+        </div>
+        
+        <label data-lang-zh="视觉描述词" data-lang-en="Prompt">视觉描述词</label>
+        <textarea id="prompt" placeholder="在此输入视觉描述词..." style="height: 90px; margin-bottom: 0;"></textarea>
+    </div>
+
+    <!-- 视频模式选项 -->
+    <div class="sidebar-section" id="video-opts" style="display:none">
+        <div class="setting-group">
+            <div class="group-title" data-lang-zh="基础设置" data-lang-en="Basic Settings">基础设置</div>
+            <div class="flex-row">
+                <div class="flex-1">
+                    <label data-lang-zh="清晰度" data-lang-en="Quality">清晰度</label>
+                    <select id="vid-quality" onchange="updateResPreview()">
+                        <option value="1080">1080P Full HD</option>
+                        <option value="720" selected>720P Standard</option>
+                        <option value="576">576P Preview</option>
+                    </select>
+                </div>
+                <div class="flex-1">
+                    <label data-lang-zh="画幅比例" data-lang-en="Aspect Ratio">画幅比例</label>
+                    <select id="vid-ratio" onchange="updateResPreview()">
+                        <option value="16:9">16:9</option>
+                        <option value="9:16">9:16</option>
+                    </select>
+                </div>
+            </div>
+            <div id="res-preview" class="res-preview-tag" style="margin-top: -5px; margin-bottom: 12px;">最终发送: 1280x720</div>
+
+            <div class="flex-row">
+                <div class="flex-1">
+                    <label data-lang-zh="帧率" data-lang-en="FPS">帧率</label>
+                    <select id="vid-fps">
+                        <option value="24" selected>24 FPS</option>
+                        <option value="25">25 FPS</option>
+                        <option value="30">30 FPS</option>
+                        <option value="48">48 FPS</option>
+                        <option value="60">60 FPS</option>
+                    </select>
+                </div>
+                <div class="flex-1">
+                    <label data-lang-zh="时长 (秒)" data-lang-en="Duration (sec)">时长 (秒)</label>
+                    <input type="number" id="vid-duration" value="5" min="1" max="30" step="1">
+                </div>
+            </div>
+        </div>
+
+        <div class="setting-group">
+            <div class="group-title" data-lang-zh="镜头与音频" data-lang-en="Camera & Audio">镜头与音频</div>
+            <label data-lang-zh="镜头运动" data-lang-en="Camera Motion">镜头运动</label>
+            <select id="vid-motion">
+                <option value="static" selected>Static (静止机位)</option>
+                <option value="dolly_in">Dolly In (推近)</option>
+                <option value="dolly_out">Dolly Out (拉远)</option>
+                <option value="dolly_left">Dolly Left (向左)</option>
+                <option value="dolly_right">Dolly Right (向右)</option>
+                <option value="jib_up">Jib Up (升臂)</option>
+                <option value="jib_down">Jib Down (降臂)</option>
+                <option value="focus_shift">Focus Shift (焦点)</option>
+            </select>
+            <div class="checkbox-container">
+                <input type="checkbox" id="vid-audio" checked>
+                <label for="vid-audio" data-lang-zh="生成 AI 环境音" data-lang-en="Generate AI Audio">生成 AI 环境音</label>
+            </div>
+        </div>
+
+        <!-- 生成媒介组 -->
+        <div class="setting-group" id="video-source-group">
+            <div class="group-title" data-lang-zh="生成媒介" data-lang-en="Source">生成媒介</div>
+            
+            <div class="flex-row" style="margin-bottom: 10px;">
+                <div class="flex-1">
+                    <label data-lang-zh="起始帧" data-lang-en="Start Frame">起始帧</label>
+                    <div class="upload-zone" id="start-frame-drop-zone" onclick="document.getElementById('start-frame-input').click()">
+                        <div class="clear-img-overlay" id="clear-start-frame-overlay" onclick="event.stopPropagation(); clearFrame('start')">×</div>
+                        <div id="start-frame-placeholder">
+                            <div class="upload-icon">🖼️</div>
+                            <div class="upload-text" data-lang-zh="上传首帧" data-lang-en="Upload">上传首帧</div>
+                        </div>
+                        <img id="start-frame-preview" class="preview-thumb">
+                        <input type="file" id="start-frame-input" accept="image/*" style="display:none" onchange="handleFrameUpload(this.files[0], 'start')">
+                    </div>
+                    <input type="hidden" id="start-frame-path">
+                </div>
+                <div class="flex-1">
+                    <label data-lang-zh="结束帧" data-lang-en="End Frame">结束帧</label>
+                    <div class="upload-zone" id="end-frame-drop-zone" onclick="document.getElementById('end-frame-input').click()">
+                        <div class="clear-img-overlay" id="clear-end-frame-overlay" onclick="event.stopPropagation(); clearFrame('end')">×</div>
+                        <div id="end-frame-placeholder">
+                            <div class="upload-icon">🏁</div>
+                            <div class="upload-text" data-lang-zh="上传尾帧" data-lang-en="Upload">上传尾帧</div>
+                        </div>
+                        <img id="end-frame-preview" class="preview-thumb">
+                        <input type="file" id="end-frame-input" accept="image/*" style="display:none" onchange="handleFrameUpload(this.files[0], 'end')">
+                    </div>
+                    <input type="hidden" id="end-frame-path">
+                </div>
+            </div>
+
+            <div class="flex-row">
+                <div class="flex-1">
+                    <label data-lang-zh="参考音频" data-lang-en="Audio">参考音频</label>
+                    <div class="upload-zone" id="audio-drop-zone" onclick="document.getElementById('vid-audio-input').click()">
+                        <div class="clear-img-overlay" id="clear-audio-overlay" onclick="event.stopPropagation(); clearUploadedAudio()">×</div>
+                        <div id="audio-upload-placeholder">
+                            <div class="upload-icon">🎵</div>
+                            <div class="upload-text" data-lang-zh="点击上传音频" data-lang-en="Upload">点击上传音频</div>
+                        </div>
+                        <div id="audio-upload-status" style="display:none;">
+                            <div class="upload-icon" style="color:var(--accent); opacity:1;">✔️</div>
+                            <div id="audio-filename-status" class="upload-text"></div>
+                        </div>
+                        <input type="file" id="vid-audio-input" accept="audio/*" style="display:none" onchange="handleAudioUpload(this.files[0])">
+                    </div>
+                    <input type="hidden" id="uploaded-audio-path">
+                </div>
+            </div>
+            <div style="font-size: 10px; color: var(--text-dim); text-align: center; margin-top: 5px;" data-lang-zh="💡 仅首帧=图生视频，首尾帧=插帧" data-lang-en="💡 Start only=Img2Vid, Both=Interpolation">
+                💡 仅首帧=图生视频，首尾帧=插帧
+            </div>
+        </div>
+    </div>
+
+    <!-- 图像模式选项 -->
+    <div id="image-opts" class="sidebar-section" style="display:none">
+        <label data-lang-zh="分辨率预设" data-lang-en="Resolution">分辨率预设</label>
+        <select id="img-res-preset" onchange="applyImgPreset(this.value)">
+            <option value="1024x1024">1:1 (1024x1024)</option>
+            <option value="1280x720">16:9 (1280x720)</option>
+            <option value="720x1280">9:16 (720x1280)</option>
+            <option value="custom">Custom</option>
+        </select>
+
+        <div id="img-custom-res" class="flex-row" style="margin-top: 10px;">
+            <div class="flex-1"><label data-lang-zh="宽度" data-lang-en="Width">宽度</label><input type="number" id="img-w" value="1024" onchange="updateImgResPreview()"></div>
+            <div class="flex-1"><label data-lang-zh="高度" data-lang-en="Height">高度</label><input type="number" id="img-h" value="1024" onchange="updateImgResPreview()"></div>
+        </div>
+        <div id="img-res-preview" class="res-preview-tag">最终发送: 1024x1024</div>
+        
+        <div class="label-group" style="margin-top: 15px;">
+            <label data-lang-zh="采样步数" data-lang-en="Steps">采样步数</label>
+            <span class="val-badge" id="stepsVal">28</span>
+        </div>
+        <div class="slider-container">
+            <input type="range" id="img-steps" min="1" max="50" value="28" oninput="document.getElementById('stepsVal').innerText=this.value">
+        </div>
+    </div>
+
+    <!-- 超分模式选项 -->
+    <div id="upscale-opts" class="sidebar-section" style="display:none">
+        <div class="setting-group">
+            <label data-lang-zh="待超分视频" data-lang-en="Video">待超分视频</label>
+            <div class="upload-zone" id="upscale-drop-zone" onclick="document.getElementById('upscale-video-input').click()" style="margin-bottom: 0;">
+                <div class="clear-img-overlay" id="clear-upscale-overlay" onclick="event.stopPropagation(); clearUpscaleVideo()">×</div>
+                <div id="upscale-placeholder">
+                    <div class="upload-icon">📹</div>
+                    <div class="upload-text" data-lang-zh="拖入视频" data-lang-en="Drop Video">拖入视频</div>
+                </div>
+                <div id="upscale-status" style="display:none;">
+                    <div class="upload-icon" style="color:var(--accent); opacity:1;">✔️</div>
+                    <div id="upscale-filename" class="upload-text"></div>
+                </div>
+                <input type="file" id="upscale-video-input" accept="video/*" style="display:none" onchange="handleUpscaleVideoUpload(this.files[0])">
+            </div>
+            <input type="hidden" id="upscale-video-path">
+        </div>
+
+        <div class="setting-group">
+            <label data-lang-zh="目标分辨率" data-lang-en="Target">目标分辨率</label>
+            <select id="upscale-res" style="margin-bottom: 0;">
+                <option value="1080p">1080P (2x)</option>
+                <option value="720p">720P</option>
+            </select>
+        </div>
+    </div>
+
+    <div style="padding: 0 30px 30px 30px;">
+        <button class="btn-primary" id="mainBtn" onclick="run()" data-lang-zh="开始渲染" data-lang-en="Generate">开始渲染</button>
+    </div>
+</aside>
+
+<main class="workspace">
+    <section class="viewer" id="viewer-section">
+        <div class="monitor" id="viewer">
+            <div id="loading-txt" data-lang-zh="等待分配渲染任务..." data-lang-en="Waiting for task...">等待分配渲染任务...</div>
+            <img id="res-img" src="">
+            <div id="video-wrapper" style="width:100%; height:100%; display:none; max-height:100%; align-items:center; justify-content:center;">
+                <video id="res-video" autoplay loop playsinline></video>
+            </div>
+            <div class="progress-container"><div id="progress-fill"></div></div>
+        </div>
+    </section>
+
+    <!-- Drag Handle -->
+    <div id="resize-handle" style="
+        height: 5px; background: transparent; cursor: row-resize;
+        flex-shrink: 0; position: relative; z-index: 50;
+        display: flex; align-items: center; justify-content: center;
+    " title="拖动调整面板高度">
+        <div style="width: 40px; height: 3px; background: var(--border); border-radius: 999px; pointer-events: none;"></div>
+    </div>
+
+    <section class="library" id="library-section">
+        <div style="display: flex; justify-content: space-between; margin-bottom: 15px; align-items: center; border-bottom: 1px solid var(--border); padding-bottom: 10px;">
+            <div style="display: flex; gap: 20px;">
+                <span id="tab-history" style="font-size: 11px; font-weight: 800; color: var(--accent); cursor: pointer; border-bottom: 2px solid var(--accent); padding-bottom: 11px; margin-bottom: -11px;" onclick="switchLibTab('history')" data-lang-zh="历史资产" data-lang-en="ASSETS">历史资产</span>
+                <span id="tab-log" style="font-size: 11px; font-weight: 800; color: var(--text-dim); cursor: pointer; border-bottom: 2px solid transparent; padding-bottom: 11px; margin-bottom: -11px;" onclick="switchLibTab('log')" data-lang-zh="系统日志" data-lang-en="LOGS">系统日志</span>
+            </div>
+                <button onclick="fetchHistory(currentHistoryPage)" style="background: var(--item); border: 1px solid var(--border); border-radius: 6px; color: var(--text-dim); font-size: 11px; padding: 4px 10px; cursor: pointer;" data-lang-zh="刷新" data-lang-en="Refresh">刷新</button>
+        </div>
+        
+        <div id="log-container" style="display: none; flex: 1; flex-direction: column;">
+            <div id="log">> LTX-2 Studio Ready. Expecting commands...</div>
+        </div>
+        
+        <div id="history-wrapper">
+            <div id="history-container"></div>
+        </div>
+        <div id="pagination-bar" style="display:none;"></div>
+    </section>
+</main>
+<script src="https://cdnjs.cloudflare.com/ajax/libs/plyr/3.7.8/plyr.min.js"></script>
+<script src="index.js"></script>
+
+</body>
+</html>

LTX2.3/UI/index.js

@@ -0,0 +1,1055 @@
+// ─── Resizable panel drag logic ───────────────────────────────────────────────
+(function() {
+    const handle = document.getElementById('resize-handle');
+    const viewer = document.getElementById('viewer-section');
+    const library = document.getElementById('library-section');
+    const workspace = document.querySelector('.workspace');
+    let dragging = false, startY = 0, startVH = 0;
+
+    handle.addEventListener('mousedown', (e) => {
+        dragging = true;
+        startY = e.clientY;
+        startVH = viewer.getBoundingClientRect().height;
+        document.body.style.cursor = 'row-resize';
+        document.body.style.userSelect = 'none';
+        handle.querySelector('div').style.background = 'var(--accent)';
+        e.preventDefault();
+    });
+    document.addEventListener('mousemove', (e) => {
+        if (!dragging) return;
+        const wsH = workspace.getBoundingClientRect().height;
+        const delta = e.clientY - startY;
+        let newVH = startVH + delta;
+        // Clamp: viewer min 150px, library min 100px
+        newVH = Math.max(150, Math.min(wsH - 100 - 5, newVH));
+        viewer.style.flex = 'none';
+        viewer.style.height = newVH + 'px';
+        library.style.flex = '1';
+    });
+    document.addEventListener('mouseup', () => {
+        if (dragging) {
+            dragging = false;
+            document.body.style.cursor = '';
+            document.body.style.userSelect = '';
+            handle.querySelector('div').style.background = 'var(--border)';
+        }
+    });
+    // Hover highlight
+    handle.addEventListener('mouseenter', () => { handle.querySelector('div').style.background = 'var(--text-dim)'; });
+    handle.addEventListener('mouseleave', () => { if (!dragging) handle.querySelector('div').style.background = 'var(--border)'; });
+})();
+// ──────────────────────────────────────────────────────────────────────────────
+
+
+
+
+
+
+// 动态获取当前访问的域名或 IP，自动对齐 3000 端口
+    const BASE = `http://${window.location.hostname}:3000`;
+    
+    let currentMode = 'image';
+    let pollInterval = null;
+    let isEnglish = false;
+
+    function toggleLang() {
+        isEnglish = !isEnglish;
+        const lang = isEnglish ? 'en' : 'zh';
+        
+        document.querySelectorAll('[data-lang-zh]').forEach(el => {
+            if (el.tagName === 'INPUT' || el.tagName === 'TEXTAREA') {
+                el.placeholder = el.getAttribute('data-lang-' + lang) || el.placeholder;
+            } else {
+                el.textContent = el.getAttribute('data-lang-' + lang) || el.textContent;
+            }
+        });
+        
+        document.getElementById('langBtn').title = isEnglish ? '切换语言' : 'Toggle Language';
+    }
+
+    // 建议增加一个简单的调试日志，方便在控制台确认地址是否正确
+    console.log("Connecting to Backend API at:", BASE);
+
+    // 分辨率自动计算逻辑
+    function updateResPreview() {
+        const q = document.getElementById('vid-quality').value; // "1080", "720", "544"
+        const r = document.getElementById('vid-ratio').value;
+        
+        let resLabel = q === "1080" ? "1080p" : q === "720" ? "720p" : "576p";
+        
+        let resDisplay;
+        if (r === "16:9") {
+            resDisplay = q === "1080" ? "1920x1080" : q === "720" ? "1280x720" : "1024x576";
+        } else {
+            resDisplay = q === "1080" ? "1080x1920" : q === "720" ? "720x1280" : "576x1024";
+        }
+        
+        document.getElementById('res-preview').innerText = `最终发送规格: ${resLabel} (${resDisplay})`;
+        return resLabel;
+    }
+
+    // 图片分辨率预览
+    function updateImgResPreview() {
+        const w = document.getElementById('img-w').value;
+        const h = document.getElementById('img-h').value;
+        document.getElementById('img-res-preview').innerText = `最终发送规格: ${w}x${h}`;
+    }
+
+    // 切换图片预设分辨率
+    function applyImgPreset(val) {
+        if (val === "custom") {
+            document.getElementById('img-custom-res').style.display = 'flex';
+        } else {
+            const [w, h] = val.split('x');
+            document.getElementById('img-w').value = w;
+            document.getElementById('img-h').value = h;
+            updateImgResPreview();
+            // 隐藏自定义区域或保持显示供微调
+            // document.getElementById('img-custom-res').style.display = 'none';
+        }
+    }
+
+
+
+    // 处理帧图片上传
+    async function handleFrameUpload(file, frameType) {
+        if (!file) return;
+
+        const preview = document.getElementById(`${frameType}-frame-preview`);
+        const placeholder = document.getElementById(`${frameType}-frame-placeholder`);
+        const clearOverlay = document.getElementById(`clear-${frameType}-frame-overlay`);
+
+        const previewReader = new FileReader();
+        previewReader.onload = (e) => {
+            preview.src = e.target.result;
+            preview.style.display = 'block';
+            placeholder.style.display = 'none';
+            clearOverlay.style.display = 'flex';
+        };
+        previewReader.readAsDataURL(file);
+
+        const reader = new FileReader();
+        reader.onload = async (e) => {
+            const b64Data = e.target.result;
+            addLog(`正在上传 ${frameType === 'start' ? '起始帧' : '结束帧'}: ${file.name}...`);
+            try {
+                const res = await fetch(`${BASE}/api/system/upload-image`, {
+                    method: 'POST',
+                    headers: { 'Content-Type': 'application/json' },
+                    body: JSON.stringify({ image: b64Data, filename: file.name })
+                });
+                const data = await res.json();
+                if (res.ok && data.path) {
+                    document.getElementById(`${frameType}-frame-path`).value = data.path;
+                    addLog(`✅ ${frameType === 'start' ? '起始帧' : '结束帧'}上传成功`);
+                } else {
+                    throw new Error(data.error || data.detail || "上传失败");
+                }
+            } catch (e) {
+                addLog(`❌ 帧图片上传失败: ${e.message}`);
+            }
+        };
+        reader.readAsDataURL(file);
+    }
+
+    function clearFrame(frameType) {
+        document.getElementById(`${frameType}-frame-input`).value = "";
+        document.getElementById(`${frameType}-frame-path`).value = "";
+        document.getElementById(`${frameType}-frame-preview`).style.display = 'none';
+        document.getElementById(`${frameType}-frame-preview`).src = "";
+        document.getElementById(`${frameType}-frame-placeholder`).style.display = 'block';
+        document.getElementById(`clear-${frameType}-frame-overlay`).style.display = 'none';
+        addLog(`🧹 已清除${frameType === 'start' ? '起始帧' : '结束帧'}`);
+    }
+
+    // 处理图片上传
+    async function handleImageUpload(file) {
+        if (!file) return;
+        
+        // 预览图片
+        const preview = document.getElementById('upload-preview');
+        const placeholder = document.getElementById('upload-placeholder');
+        const clearOverlay = document.getElementById('clear-img-overlay');
+        
+        const previewReader = new FileReader();
+        preview.onload = () => {
+            preview.style.display = 'block';
+            placeholder.style.display = 'none';
+            clearOverlay.style.display = 'flex';
+        };
+        previewReader.onload = (e) => preview.src = e.target.result;
+        previewReader.readAsDataURL(file);
+
+        // 使用 FileReader 转换为 Base64，绕过后端缺失 python-multipart 的问题
+        const reader = new FileReader();
+        reader.onload = async (e) => {
+            const b64Data = e.target.result;
+            addLog(`正在上传参考图: ${file.name}...`);
+            try {
+                const res = await fetch(`${BASE}/api/system/upload-image`, {
+                    method: 'POST',
+                    headers: { 'Content-Type': 'application/json' },
+                    body: JSON.stringify({
+                        image: b64Data,
+                        filename: file.name
+                    })
+                });
+                const data = await res.json();
+                if (res.ok && data.path) {
+                    document.getElementById('uploaded-img-path').value = data.path;
+                    addLog(`✅ 参考图上传成功: ${file.name}`);
+                } else {
+                    const errMsg = data.error || data.detail || "上传失败";
+                    throw new Error(typeof errMsg === 'string' ? errMsg : JSON.stringify(errMsg));
+                }
+            } catch (e) {
+                addLog(`❌ 图片上传失败: ${e.message}`);
+            }
+        };
+        reader.onerror = () => addLog("❌ 读取本地文件失败");
+        reader.readAsDataURL(file);
+    }
+
+    function clearUploadedImage() {
+        document.getElementById('vid-image-input').value = "";
+        document.getElementById('uploaded-img-path').value = "";
+        document.getElementById('upload-preview').style.display = 'none';
+        document.getElementById('upload-preview').src = "";
+        document.getElementById('upload-placeholder').style.display = 'block';
+        document.getElementById('clear-img-overlay').style.display = 'none';
+        addLog("🧹 已清除参考图");
+    }
+
+    // 处理音频上传
+    async function handleAudioUpload(file) {
+        if (!file) return;
+
+        const placeholder = document.getElementById('audio-upload-placeholder');
+        const statusDiv = document.getElementById('audio-upload-status');
+        const filenameStatus = document.getElementById('audio-filename-status');
+        const clearOverlay = document.getElementById('clear-audio-overlay');
+
+        placeholder.style.display = 'none';
+        filenameStatus.innerText = file.name;
+        statusDiv.style.display = 'block';
+        clearOverlay.style.display = 'flex';
+
+        const reader = new FileReader();
+        reader.onload = async (e) => {
+            const b64Data = e.target.result;
+            addLog(`正在上传音频: ${file.name}...`);
+            try {
+                // 复用图片上传接口，后端已支持任意文件类型
+                const res = await fetch(`${BASE}/api/system/upload-image`, {
+                    method: 'POST',
+                    headers: { 'Content-Type': 'application/json' },
+                    body: JSON.stringify({
+                        image: b64Data,
+                        filename: file.name
+                    })
+                });
+                const data = await res.json();
+                if (res.ok && data.path) {
+                    document.getElementById('uploaded-audio-path').value = data.path;
+                    addLog(`✅ 音频上传成功: ${file.name}`);
+                } else {
+                    const errMsg = data.error || data.detail || "上传失败";
+                    throw new Error(typeof errMsg === 'string' ? errMsg : JSON.stringify(errMsg));
+                }
+            } catch (e) {
+                addLog(`❌ 音频上传失败: ${e.message}`);
+            }
+        };
+        reader.onerror = () => addLog("❌ 读取本地音频文件失败");
+        reader.readAsDataURL(file);
+    }
+
+    function clearUploadedAudio() {
+        document.getElementById('vid-audio-input').value = "";
+        document.getElementById('uploaded-audio-path').value = "";
+        document.getElementById('audio-upload-placeholder').style.display = 'block';
+        document.getElementById('audio-upload-status').style.display = 'none';
+        document.getElementById('clear-audio-overlay').style.display = 'none';
+        addLog("🧹 已清除音频文件");
+    }
+
+    // 处理超分视频上传
+    async function handleUpscaleVideoUpload(file) {
+        if (!file) return;
+        const placeholder = document.getElementById('upscale-placeholder');
+        const statusDiv = document.getElementById('upscale-status');
+        const filenameStatus = document.getElementById('upscale-filename');
+        const clearOverlay = document.getElementById('clear-upscale-overlay');
+
+        filenameStatus.innerText = file.name;
+        placeholder.style.display = 'none';
+        statusDiv.style.display = 'block';
+        clearOverlay.style.display = 'flex';
+
+        const reader = new FileReader();
+        reader.onload = async (e) => {
+            const b64Data = e.target.result;
+            addLog(`正在上传待超分视频: ${file.name}...`);
+            try {
+                const res = await fetch(`${BASE}/api/system/upload-image`, {
+                    method: 'POST',
+                    headers: { 'Content-Type': 'application/json' },
+                    body: JSON.stringify({ image: b64Data, filename: file.name })
+                });
+                const data = await res.json();
+                if (res.ok && data.path) {
+                    document.getElementById('upscale-video-path').value = data.path;
+                    addLog(`✅ 视频上传成功`);
+                } else {
+                    throw new Error(data.error || "上传失败");
+                }
+            } catch (e) {
+                addLog(`❌ 视频上传失败: ${e.message}`);
+            }
+        };
+        reader.readAsDataURL(file);
+    }
+
+    function clearUpscaleVideo() {
+        document.getElementById('upscale-video-input').value = "";
+        document.getElementById('upscale-video-path').value = "";
+        document.getElementById('upscale-placeholder').style.display = 'block';
+        document.getElementById('upscale-status').style.display = 'none';
+        document.getElementById('clear-upscale-overlay').style.display = 'none';
+        addLog("🧹 已清除待超分视频");
+    }
+
+    // 初始化拖拽上传逻辑
+    function initDragAndDrop() {
+        const audioDropZone = document.getElementById('audio-drop-zone');
+        const startFrameDropZone = document.getElementById('start-frame-drop-zone');
+        const endFrameDropZone = document.getElementById('end-frame-drop-zone');
+        const upscaleDropZone = document.getElementById('upscale-drop-zone');
+        
+        const zones = [audioDropZone, startFrameDropZone, endFrameDropZone, upscaleDropZone];
+
+        ['dragenter', 'dragover', 'dragleave', 'drop'].forEach(eventName => {
+            zones.forEach(zone => {
+                if (!zone) return;
+                zone.addEventListener(eventName, (e) => {
+                    e.preventDefault();
+                    e.stopPropagation();
+                }, false);
+            });
+        });
+
+        ['dragenter', 'dragover'].forEach(eventName => {
+            zones.forEach(zone => {
+                if (!zone) return;
+                zone.addEventListener(eventName, () => zone.classList.add('dragover'), false);
+            });
+        });
+
+        ['dragleave', 'drop'].forEach(eventName => {
+            zones.forEach(zone => {
+                if (!zone) return;
+                zone.addEventListener(eventName, () => zone.classList.remove('dragover'), false);
+            });
+        });
+
+        audioDropZone.addEventListener('drop', (e) => {
+            const file = e.dataTransfer.files[0];
+            if (file && file.type.startsWith('audio/')) handleAudioUpload(file);
+        }, false);
+
+        startFrameDropZone.addEventListener('drop', (e) => {
+            const file = e.dataTransfer.files[0];
+            if (file && file.type.startsWith('image/')) handleFrameUpload(file, 'start');
+        }, false);
+
+        endFrameDropZone.addEventListener('drop', (e) => {
+            const file = e.dataTransfer.files[0];
+            if (file && file.type.startsWith('image/')) handleFrameUpload(file, 'end');
+        }, false);
+
+        upscaleDropZone.addEventListener('drop', (e) => {
+            const file = e.dataTransfer.files[0];
+            if (file && file.type.startsWith('video/')) handleUpscaleVideoUpload(file);
+        }, false);
+    }
+
+    let _isGeneratingFlag = false;
+
+    // 系统状态轮询
+    async function checkStatus() {
+        try {
+            const h = await fetch(`${BASE}/health`).then(r => r.json()).catch(() => ({status: "error"}));
+            const g = await fetch(`${BASE}/api/gpu-info`).then(r => r.json()).catch(() => ({gpu_info: {}}));
+            const p = await fetch(`${BASE}/api/generation/progress`).then(r => r.json()).catch(() => ({progress: 0}));
+            const sysGpus = await fetch(`${BASE}/api/system/list-gpus`).then(r => r.json()).catch(() => ({gpus: []}));
+            
+            const activeGpu = (sysGpus.gpus || []).find(x => x.active) || (sysGpus.gpus || [])[0] || {};
+            const gpuName = activeGpu.name || g.gpu_info?.name || "GPU";
+            
+            const s = document.getElementById('sys-status');
+            const indicator = document.getElementById('sys-indicator');
+            
+            const isReady = h.status === "ok" || h.status === "ready" || h.models_loaded;
+            const backendActive = (p && p.progress > 0);
+            
+            if (_isGeneratingFlag || backendActive) {
+                s.innerText = `${gpuName}: 运算中...`;
+                if(indicator) indicator.className = 'indicator-busy';
+            } else {
+                s.innerText = isReady ? `${gpuName}: 在线 / 就绪` : `${gpuName}: 启动中...`;
+                if(indicator) indicator.className = isReady ? 'indicator-ready' : 'indicator-offline';
+            }
+            s.style.color = "var(--text-dim)";
+
+            const vUsedMB = g.gpu_info?.vramUsed || 0;
+            const vTotalMB = activeGpu.vram_mb || g.gpu_info?.vram || 32768; 
+            const vUsedGB = vUsedMB / 1024;
+            const vTotalGB = vTotalMB / 1024;
+            
+            document.getElementById('vram-fill').style.width = (vUsedMB / vTotalMB * 100) + "%";
+            document.getElementById('vram-text').innerText = `${vUsedGB.toFixed(1)} / ${vTotalGB.toFixed(0)} GB`;
+        } catch(e) { document.getElementById('sys-status').innerText = "未检测到后端 (Port 3000)"; }
+    }
+    setInterval(checkStatus, 1000); // 提升到 1 秒一次实时监控
+    checkStatus();
+    initDragAndDrop();
+    listGpus(); // 初始化 GPU 列表
+    getOutputDir(); // 获取当前的保存路径
+
+    async function setOutputDir() {
+        const dir = document.getElementById('global-out-dir').value.trim();
+        try {
+            const res = await fetch(`${BASE}/api/system/set-dir`, {
+                method: 'POST',
+                headers: { 'Content-Type': 'application/json' },
+                body: JSON.stringify({ directory: dir })
+            });
+            if (res.ok) {
+                addLog(`✅ 存储路径更新成功! 当前路径: ${dir || '默认路径'}`);
+                if (typeof fetchHistory === 'function') fetchHistory(currentHistoryPage);
+            }
+        } catch (e) {
+            addLog(`❌ 设置路径时连接异常: ${e.message}`);
+        }
+    }
+
+    async function browseOutputDir() {
+        try {
+            const res = await fetch(`${BASE}/api/system/browse-dir`);
+            const data = await res.json();
+            if (data.status === "success" && data.directory) {
+                document.getElementById('global-out-dir').value = data.directory;
+                // auto apply immediately
+                setOutputDir();
+                addLog(`📂 检测到新路径，已自动套用！`);
+            } else if (data.error) {
+                addLog(`❌ 内部系统权限拦截了弹窗: ${data.error}`);
+            }
+        } catch (e) {
+            addLog(`❌ 无法调出文件夹浏览弹窗, 请直接复制粘贴绝对路径。`);
+        }
+    }
+
+    async function getOutputDir() {
+        try {
+            const res = await fetch(`${BASE}/api/system/get-dir`);
+            const data = await res.json();
+            if (data.directory && data.directory.indexOf('LTXDesktop') === -1 && document.getElementById('global-out-dir')) {
+                document.getElementById('global-out-dir').value = data.directory;
+            }
+        } catch (e) {}
+    }
+
+    function switchMode(m) {
+        currentMode = m;
+        document.getElementById('tab-image').classList.toggle('active', m === 'image');
+        document.getElementById('tab-video').classList.toggle('active', m === 'video');
+        document.getElementById('tab-upscale').classList.toggle('active', m === 'upscale');
+        
+        document.getElementById('image-opts').style.display = m === 'image' ? 'block' : 'none';
+        document.getElementById('video-opts').style.display = m === 'video' ? 'block' : 'none';
+        document.getElementById('upscale-opts').style.display = m === 'upscale' ? 'block' : 'none';
+        
+        // 如果切到图像模式，隐藏提示词框外的其他东西
+        document.getElementById('prompt').placeholder = m === 'upscale' ? "输入画面增强引导词 (可选)..." : "在此输入视觉描述词 (Prompt)...";
+    }
+
+    function showGeneratingView() {
+        if (!_isGeneratingFlag) return;
+        const resImg = document.getElementById('res-img');
+        const videoWrapper = document.getElementById('video-wrapper');
+        if (resImg) resImg.style.display = "none";
+        if (videoWrapper) videoWrapper.style.display = "none";
+        if (player) {
+            try { player.stop(); } catch(_) {}
+        } else {
+            const vid = document.getElementById('res-video');
+            if (vid) { vid.pause(); vid.removeAttribute('src'); vid.load(); }
+        }
+        const loadingTxt = document.getElementById('loading-txt');
+        if (loadingTxt) loadingTxt.style.display = "flex";
+    }
+
+    async function run() {
+        // 防止重复点击（_isGeneratingFlag 比 btn.disabled 更可靠）
+        if (_isGeneratingFlag) {
+            addLog("⚠️ 当前正在生成中，请等待完成");
+            return;
+        }
+
+        const btn = document.getElementById('mainBtn');
+        const prompt = document.getElementById('prompt').value.trim();
+
+        if (currentMode !== 'upscale' && !prompt) {
+            addLog("⚠️ 请输入提示词后再开始渲染");
+            return;
+        }
+
+        // 先设置标志 + 禁用按钮，然后用顶层 try/finally 保证一定能解锁
+        _isGeneratingFlag = true;
+        btn.disabled = true;
+
+        try {
+            // 安全地操作 UI 元素（改用 if 判空，防止 Plyr 接管后 getElementById 返回 null）
+            const loader = document.getElementById('loading-txt');
+            const resImg = document.getElementById('res-img');
+            const resVideo = document.getElementById('res-video');
+
+            if (loader) {
+                loader.style.display = "flex";
+                loader.style.flexDirection = "column";
+                loader.style.alignItems = "center";
+                loader.style.gap = "12px";
+                loader.innerHTML = `
+                    <div class="spinner" style="width:48px;height:48px;border-width:4px;color:var(--accent);"></div>
+                    <div id="loader-step-text" style="font-size:13px;font-weight:700;color:var(--text-sub);">GPU 正在分配资源...</div>
+                `;
+            }
+            if (resImg) resImg.style.display = "none";
+            // 必须隐藏整个 video-wrapper（Plyr 外层容器），否则第二次生成时视频会与 spinner 叠加
+            const videoWrapper = document.getElementById('video-wrapper');
+            if (videoWrapper) videoWrapper.style.display = "none";
+            if (player) { try { player.stop(); } catch(_) {} }
+            else if (resVideo) { resVideo.pause?.(); resVideo.removeAttribute?.('src'); }
+
+        checkStatus();
+
+            // 重置后端状态锁（非关键，失败不影响主流程）
+            try { await fetch(`${BASE}/api/system/reset-state`, { method: 'POST' }); } catch(_) {}
+
+            startProgressPolling();
+
+            // ---- 新增：在历史记录区插入「正在渲染」缩略图卡片 ----
+            const historyContainer = document.getElementById('history-container');
+            if (historyContainer) {
+                const old = document.getElementById('current-loading-card');
+                if (old) old.remove();
+                const loadingCard = document.createElement('div');
+                loadingCard.className = 'history-card loading-card';
+                loadingCard.id = 'current-loading-card';
+                loadingCard.onclick = showGeneratingView;
+                loadingCard.innerHTML = `
+                    <div class="spinner"></div>
+                    <div id="loading-card-step" style="font-size:10px;color:var(--text-dim);margin-top:4px;">等待中...</div>
+                `;
+                historyContainer.prepend(loadingCard);
+            }
+
+            // ---- 构建请求 ----
+            let endpoint, payload;
+            if (currentMode === 'image') {
+                const w = parseInt(document.getElementById('img-w').value);
+                const h = parseInt(document.getElementById('img-h').value);
+                endpoint = '/api/generate-image';
+                payload = {
+                    prompt, width: w, height: h,
+                    numSteps: parseInt(document.getElementById('img-steps').value),
+                    numImages: 1
+                };
+                addLog(`正在发起图像渲染: ${w}x${h}, Steps: ${payload.numSteps}`);
+
+            } else if (currentMode === 'video') {
+                const res = updateResPreview();
+                const dur = parseFloat(document.getElementById('vid-duration').value);
+                const fps = document.getElementById('vid-fps').value;
+                if (dur > 20) addLog(`⚠️ 时长设定为 ${dur}s 极长，可能导致显存溢出或耗时较久。`);
+
+                const audio = document.getElementById('vid-audio').checked ? "true" : "false";
+                const audioPath = document.getElementById('uploaded-audio-path').value;
+                const startFramePathValue = document.getElementById('start-frame-path').value;
+                const endFramePathValue = document.getElementById('end-frame-path').value;
+
+                let finalImagePath = null, finalStartFramePath = null, finalEndFramePath = null;
+                if (startFramePathValue && endFramePathValue) {
+                    finalStartFramePath = startFramePathValue;
+                    finalEndFramePath = endFramePathValue;
+                } else if (startFramePathValue) {
+                    finalImagePath = startFramePathValue;
+                }
+
+                endpoint = '/api/generate';
+                payload = {
+                    prompt, resolution: res, model: "ltx-2",
+                    cameraMotion: document.getElementById('vid-motion').value,
+                    negativePrompt: "low quality, blurry, noisy, static noise, distorted",
+                    duration: String(dur), fps, audio,
+                    imagePath: finalImagePath,
+                    audioPath: audioPath || null,
+                    startFramePath: finalStartFramePath,
+                    endFramePath: finalEndFramePath,
+                    aspectRatio: document.getElementById('vid-ratio').value
+                };
+                addLog(`正在发起视频渲染: ${res}, 时长: ${dur}s, FPS: ${fps}, 音频: ${audio}, 参考图: ${finalImagePath ? '已加载' : '无'}, 参考音频: ${audioPath ? '已加载' : '无'}, 插帧: ${finalStartFramePath && finalEndFramePath ? '已加载' : '无'}, 镜头: ${payload.cameraMotion}`);
+
+            } else if (currentMode === 'upscale') {
+                const videoPath = document.getElementById('upscale-video-path').value;
+                const targetRes = document.getElementById('upscale-res').value;
+                if (!videoPath) throw new Error("请先上传待超分的视频");
+                endpoint = '/api/system/upscale-video';
+                payload = { video_path: videoPath, resolution: targetRes, prompt: "high quality, detailed, 4k", strength: 0.7 };
+                addLog(`正在发起视频超分: 目标 ${targetRes}`);
+            }
+
+            // ---- 发送请求 ----
+            const res = await fetch(BASE + endpoint, {
+                method: 'POST',
+                headers: { 'Content-Type': 'application/json' },
+                body: JSON.stringify(payload)
+            });
+            const data = await res.json();
+            if (!res.ok) {
+                const errMsg = data.error || data.detail || "API 拒绝了请求";
+                throw new Error(typeof errMsg === 'string' ? errMsg : JSON.stringify(errMsg));
+            }
+
+            // ---- 显示结果 ----
+            const rawPath = data.image_paths ? data.image_paths[0] : data.video_path;
+            if (rawPath) {
+                try { displayOutput(rawPath); } catch (dispErr) { addLog(`⚠️ 播放器显示异常: ${dispErr.message}`); }
+            }
+
+            // 强制刷新历史记录（不依赖 isLoadingHistory 标志，确保新生成的视频立即显示）
+            setTimeout(() => {
+                isLoadingHistory = false; // 强制重置状态
+                if (typeof fetchHistory === 'function') fetchHistory(1);
+            }, 500);
+
+        } catch (e) {
+            addLog(`❌ 渲染中断: ${e.message}`);
+            const loader = document.getElementById('loading-txt');
+            if (loader) loader.innerText = "渲染失败，请检查显存或参数";
+
+        } finally {
+            // ✅ 无论发生什么，这里一定执行，确保按钮永远可以再次点击
+            _isGeneratingFlag = false;
+            btn.disabled = false;
+            stopProgressPolling();
+            checkStatus();
+            // 生成完毕后自动释放显存，降低 VRAM 压力（不 await 避免阻塞 UI 解锁）
+            setTimeout(() => clearGpu(), 500);
+        }
+    }
+
+    async function clearGpu() {
+        const btn = document.getElementById('clearGpuBtn');
+        btn.disabled = true;
+        btn.innerText = "清理中...";
+        try {
+            const res = await fetch(`${BASE}/api/system/clear-gpu`, {
+                method: 'POST',
+                headers: { 'Content-Type': 'application/json' }
+            });
+            const data = await res.json();
+            if (res.ok) {
+                addLog(`🧹 显存清理成功: ${data.message}`);
+                // 立即触发状态刷新
+                checkStatus();
+                setTimeout(checkStatus, 1000); 
+            } else {
+                const errMsg = data.error || data.detail || "后端未实现此接口 (404)";
+                throw new Error(errMsg);
+            }
+        } catch(e) {
+            addLog(`❌ 清理显存失败: ${e.message}`);
+        } finally {
+            btn.disabled = false;
+            btn.innerText = "释放显存";
+        }
+    }
+
+    async function listGpus() {
+        try {
+            const res = await fetch(`${BASE}/api/system/list-gpus`);
+            const data = await res.json();
+            if (res.ok && data.gpus) {
+                const selector = document.getElementById('gpu-selector');
+                selector.innerHTML = data.gpus.map(g => 
+                    `<option value="${g.id}" ${g.active ? 'selected' : ''}>GPU ${g.id}: ${g.name} (${g.vram})</option>`
+                ).join('');
+                
+                // 更新当前显示的 GPU 名称
+                const activeGpu = data.gpus.find(g => g.active);
+                if (activeGpu) document.getElementById('gpu-name').innerText = activeGpu.name;
+            }
+        } catch (e) {
+            console.error("Failed to list GPUs", e);
+        }
+    }
+
+    async function switchGpu(id) {
+        if (!id) return;
+        addLog(`🔄 正在切换到 GPU ${id}...`);
+        try {
+            const res = await fetch(`${BASE}/api/system/switch-gpu`, {
+                method: 'POST',
+                headers: { 'Content-Type': 'application/json' },
+                body: JSON.stringify({ gpu_id: parseInt(id) })
+            });
+            const data = await res.json();
+            if (res.ok) {
+                addLog(`✅ 已成功切换到 GPU ${id}，模型将重新加载。`);
+                listGpus(); // 重新获取列表以同步状态
+                setTimeout(checkStatus, 1000);
+            } else {
+                throw new Error(data.error || "切换失败");
+            }
+        } catch (e) {
+            addLog(`❌ GPU 切换失败: ${e.message}`);
+        }
+    }
+
+    function startProgressPolling() {
+        if (pollInterval) clearInterval(pollInterval);
+        pollInterval = setInterval(async () => {
+            try {
+                const res = await fetch(`${BASE}/api/generation/progress`);
+                const d = await res.json();
+                if (d.progress > 0) {
+                    const phaseMap = {
+                        'loading_model': '加载权重',
+                        'encoding_text': 'T5 编码',
+                        'validating_request': '校验请求',
+                        'uploading_audio': '上传音频',
+                        'uploading_image': '上传图像',
+                        'inference': 'AI 推理',
+                        'downloading_output': '下载结果',
+                        'complete': '完成'
+                    };
+                    const phaseStr = phaseMap[d.phase] || d.phase || '推理';
+
+                    // 步骤格式：优先显示 current_step/total_steps，降级用百分比
+                    let stepLabel;
+                    if (d.current_step !== undefined && d.current_step !== null && d.total_steps) {
+                        stepLabel = `${d.current_step}/${d.total_steps} 步`;
+                    } else {
+                        stepLabel = `${d.progress}%`;
+                    }
+
+                    document.getElementById('progress-fill').style.width = d.progress + "%";
+                    // 更��主预览区的进度文字（内嵌子元素）
+                    const loaderStep = document.getElementById('loader-step-text');
+                    if (loaderStep) loaderStep.innerText = `GPU 运算中: ${stepLabel}  [ ${phaseStr} ]`;
+                    else {
+                        const loadingTxt = document.getElementById('loading-txt');
+                        if (loadingTxt) loadingTxt.innerText = `GPU 运算中: ${stepLabel} [${phaseStr}]`;
+                    }
+
+                    // 同步更新历史缩略图卡片上的进度文字
+                    const cardStep = document.getElementById('loading-card-step');
+                    if (cardStep) cardStep.innerText = stepLabel;
+                }
+            } catch(e) {}
+        }, 1000);
+    }
+
+    function stopProgressPolling() {
+        clearInterval(pollInterval);
+        pollInterval = null;
+        document.getElementById('progress-fill').style.width = "0%";
+        // 移除渲染中的卡片（生成已结束）
+        const lc = document.getElementById('current-loading-card');
+        if (lc) lc.remove();
+    }
+
+    function displayOutput(fileOrPath) {
+        const img = document.getElementById('res-img');
+        const vid = document.getElementById('res-video');
+        const loader = document.getElementById('loading-txt');
+        
+        // 关键BUG修复：切换前强制清除并停止现有视频和声音，避免后台继续播放
+        if(player) {
+            player.stop();
+        } else {
+            vid.pause();
+            vid.removeAttribute('src');
+            vid.load();
+        }
+        
+        let url = "";
+        let fileName = fileOrPath;
+        if (fileOrPath.indexOf('\\') !== -1 || fileOrPath.indexOf('/') !== -1) {
+            url = `${BASE}/api/system/file?path=${encodeURIComponent(fileOrPath)}&t=${Date.now()}`;
+            fileName = fileOrPath.split(/[\\/]/).pop();
+        } else {
+            const outInput = document.getElementById('global-out-dir');
+            const globalDir = outInput ? outInput.value.replace(/\\/g, '/').replace(/\/$/, '') : "";
+            if (globalDir && globalDir !== "") {
+                url = `${BASE}/api/system/file?path=${encodeURIComponent(globalDir + '/' + fileOrPath)}&t=${Date.now()}`;
+            } else {
+                url = `${BASE}/outputs/${fileOrPath}?t=${Date.now()}`;
+            }
+        }
+
+        loader.style.display = "none";
+        if (currentMode === 'image') {
+            img.src = url;
+            img.style.display = "block";
+            addLog(`✅ 图像渲染成功: ${fileName}`);
+        } else {
+            document.getElementById('video-wrapper').style.display = "flex";
+            
+            if(player) {
+                player.source = {
+                    type: 'video',
+                    sources: [{ src: url, type: 'video/mp4' }]
+                };
+                player.play();
+            } else {
+                vid.src = url;
+            }
+            addLog(`✅ 视频渲染成功: ${fileName}`);
+        }
+    }
+
+
+
+    function addLog(msg) {
+        const log = document.getElementById('log');
+        const time = new Date().toLocaleTimeString();
+        log.innerHTML += `<div style="margin-bottom:5px"> <span style="color:var(--text-dim)">[${time}]</span> ${msg}</div>`;
+        log.scrollTop = log.scrollHeight;
+    }
+
+
+// Force switch to video mode on load
+window.addEventListener('DOMContentLoaded', () => switchMode('video'));
+
+
+    
+    
+    
+    
+    
+    
+    
+    
+    
+    
+    let isLoadingHistory = false;
+
+    function switchLibTab(tab) {
+        document.getElementById('log-container').style.display = tab === 'log' ? 'flex' : 'none';
+        const hw = document.getElementById('history-wrapper');
+        if (hw) hw.style.display = tab === 'history' ? 'block' : 'none';
+        
+        document.getElementById('tab-log').style.color = tab === 'log' ? 'var(--accent)' : 'var(--text-dim)';
+        document.getElementById('tab-log').style.borderColor = tab === 'log' ? 'var(--accent)' : 'transparent';
+        
+        document.getElementById('tab-history').style.color = tab === 'history' ? 'var(--accent)' : 'var(--text-dim)';
+        document.getElementById('tab-history').style.borderColor = tab === 'history' ? 'var(--accent)' : 'transparent';
+        
+        if (tab === 'history') {
+            fetchHistory();
+        }
+    }
+
+    async function fetchHistory(isFirstLoad = false, silent = false) {
+        if (isLoadingHistory) return;
+        isLoadingHistory = true;
+        
+        try {
+            // 加载所有历史，不分页
+            const res = await fetch(`${BASE}/api/system/history?page=1&limit=10000`);
+            if (!res.ok) {
+                isLoadingHistory = false;
+                return;
+            }
+            const data = await res.json();
+            
+            if (data.history && data.history.length > 0) {
+                const container = document.getElementById('history-container');
+                
+                // 清空容器
+                let loadingCardHtml = "";
+                const lc = document.getElementById('current-loading-card');
+                if (lc && _isGeneratingFlag) {
+                    loadingCardHtml = lc.outerHTML;
+                }
+                container.innerHTML = loadingCardHtml;
+
+                const outInput = document.getElementById('global-out-dir');
+                const globalDir = outInput ? outInput.value.replace(/\\/g, '/').replace(/\/$/, '') : "";
+                
+                // 过滤无效数据
+                const validHistory = data.history.filter(item => item && item.filename);
+                
+                const cardsHtml = validHistory.map((item, index) => {
+                    const url = (globalDir && globalDir !== "") 
+                        ? `${BASE}/api/system/file?path=${encodeURIComponent(globalDir + '/' + item.filename)}` 
+                        : `${BASE}/outputs/${item.filename}`;
+                        
+                    const safeFilename = item.filename.replace(/'/g, "\\'").replace(/"/g, '\\"');
+                    const media = item.type === 'video' 
+                        ? `<video data-src="${url}#t=0.001" class="lazy-load" muted loop preload="none" onmouseover="if(this.readyState >= 2) this.play()" onmouseout="this.pause()" style="pointer-events: none; object-fit: cover; width: 100%; height: 100%;"></video>` 
+                        : `<img data-src="${url}" class="lazy-load" style="object-fit: cover; width: 100%; height: 100%;">`;
+                    return `<div class="history-card" onclick="displayHistoryOutput('${safeFilename}', '${item.type}')">
+                                <div class="history-type-badge">${item.type === 'video' ? '🎬 VID' : '🎨 IMG'}</div>
+                                <button class="history-delete-btn" onclick="event.stopPropagation(); deleteHistoryItem('${safeFilename}', '${item.type}', this)">✕</button>
+                                ${media}
+                            </div>`;
+                }).join('');
+
+                container.insertAdjacentHTML('beforeend', cardsHtml);
+
+                // 重新绑定loading card点击事件
+                const newLc = document.getElementById('current-loading-card');
+                if (newLc) newLc.onclick = showGeneratingView;
+                
+                // 加载可见的图片
+                loadVisibleImages();
+            }
+        } catch(e) {
+            console.error("Failed to load history", e);
+        } finally {
+            isLoadingHistory = false;
+        }
+    }
+    
+    async function deleteHistoryItem(filename, type, btn) {
+        if (!confirm(`确定要删除 "${filename}" 吗？`)) return;
+        
+        try {
+            const res = await fetch(`${BASE}/api/system/delete-file`, {
+                method: 'POST',
+                headers: {'Content-Type': 'application/json'},
+                body: JSON.stringify({filename: filename, type: type})
+            });
+            
+            if (res.ok) {
+                // 删除成功后移除元素
+                const card = btn.closest('.history-card');
+                if (card) {
+                    card.remove();
+                }
+            } else {
+                alert('删除失败');
+            }
+        } catch(e) {
+            console.error('Delete failed', e);
+            alert('删除失败');
+        }
+    }
+
+    function loadVisibleImages() {
+        const hw = document.getElementById('history-wrapper');
+        if (!hw) return;
+        
+        const lazyMedias = document.querySelectorAll('#history-container .lazy-load');
+        
+        // 每次只加载3个媒体元素（图片或视频）
+        let loadedCount = 0;
+        lazyMedias.forEach(media => {
+            if (loadedCount >= 3) return;
+            
+            const src = media.dataset.src;
+            if (!src) return;
+            
+            // 检查是否在可见区域附近
+            const rect = media.getBoundingClientRect();
+            const containerRect = hw.getBoundingClientRect();
+            
+            if (rect.top < containerRect.bottom + 300 && rect.bottom > containerRect.top - 100) {
+                media.src = src;
+                media.classList.remove('lazy-load');
+                
+                // 视频需要额外设置 preload
+                if (media.tagName === 'VIDEO') {
+                    media.preload = 'metadata';
+                }
+                
+                loadedCount++;
+            }
+        });
+        
+        // 继续检查直到没有更多媒体需要加载
+        if (loadedCount > 0) {
+            setTimeout(loadVisibleImages, 100);
+        }
+    }
+
+    // 监听history-wrapper���滚动事件来懒加载
+    function initHistoryScrollListener() {
+        const hw = document.getElementById('history-wrapper');
+        if (!hw) return;
+        
+        let scrollTimeout;
+        hw.addEventListener('scroll', () => {
+            if (scrollTimeout) clearTimeout(scrollTimeout);
+            scrollTimeout = setTimeout(() => {
+                loadVisibleImages();
+            }, 100);
+        });
+    }
+
+    // 页面加载时初始化滚动监听
+    window.addEventListener('DOMContentLoaded', () => {
+        setTimeout(initHistoryScrollListener, 500);
+    });
+
+    function displayHistoryOutput(file, type) {
+        document.getElementById('res-img').style.display = 'none';
+        document.getElementById('video-wrapper').style.display = 'none';
+
+        const mode = type === 'video' ? 'video' : 'image';
+        switchMode(mode);
+        displayOutput(file);
+    }
+    
+    window.addEventListener('DOMContentLoaded', () => {
+        // Initialize Plyr Custom Video Component
+        if(window.Plyr) {
+            player = new Plyr('#res-video', {
+                controls: [
+                    'play-large', 'play', 'progress', 'current-time', 
+                    'mute', 'volume', 'fullscreen'
+                ],
+                settings: [],
+                loop: { active: true },
+                autoplay: true
+            });
+        }
+        
+        // Fetch current directory context to show in UI
+        fetch(`${BASE}/api/system/get-dir`)
+            .then(res => res.json())
+            .then(data => {
+                if(data && data.directory) {
+                    const outInput = document.getElementById('global-out-dir');
+                    if (outInput) outInput.value = data.directory;
+                }
+            }).catch(e => console.error(e));
+
+        setTimeout(() => fetchHistory(1), 500);
+        
+        let historyRefreshInterval = null;
+        function startHistoryAutoRefresh() {
+            if (historyRefreshInterval) return;
+            historyRefreshInterval = setInterval(() => {
+                if (document.getElementById('history-container').style.display === 'flex' && !_isGeneratingFlag) {
+                    fetchHistory(1, true);
+                }
+            }, 5000);
+        }
+        startHistoryAutoRefresh();
+        switchLibTab('history');
+    });

LTX2.3/main.py

@@ -0,0 +1,248 @@
+import os
+import sys
+import subprocess
+import threading
+import time
+import socket
+import logging
+from fastapi import FastAPI
+from fastapi.responses import FileResponse
+from fastapi.staticfiles import StaticFiles
+import uvicorn
+
+# ============================================================
+# 配置区 (动态路径适配与补丁挂载)
+# ============================================================
+def resolve_ltx_path():
+    import glob, tempfile, subprocess
+    sc_dir = os.path.join(os.getcwd(), "LTX_Shortcut")
+    os.makedirs(sc_dir, exist_ok=True)
+    lnk_files = glob.glob(os.path.join(sc_dir, "*.lnk"))
+    if not lnk_files:
+        print("\033[91m[ERROR] 未在 LTX_Shortcut 文件夹中找到快捷方式！\n请打开程序目录下的 LTX_Shortcut 文件夹，并将官方 LTX Desktop 的快捷方式复制进去后重试。\033[0m")
+        sys.exit(1)
+        
+    lnk_path = lnk_files[0]
+    # 使用 VBScript 解析快捷方式，兼容所有 Windows 系统
+    vbs_code = f'''Set sh = CreateObject("WScript.Shell")\nSet obj = sh.CreateShortcut("{os.path.abspath(lnk_path)}")\nWScript.Echo obj.TargetPath'''
+    fd, vbs_path = tempfile.mkstemp(suffix='.vbs')
+    with os.fdopen(fd, 'w') as f:
+        f.write(vbs_code)
+    try:
+        out = subprocess.check_output(['cscript', '//nologo', vbs_path], stderr=subprocess.STDOUT)
+        target_exe = out.decode('ansi').strip()
+    finally:
+        os.remove(vbs_path)
+        
+    if not target_exe or not os.path.exists(target_exe):
+        # 如果快捷方式解析失败，或者解析出来的是朋友电脑的路径（当前电脑不存在），自动全盘搜索默认路径
+        default_paths = [
+            os.path.join(os.environ.get("LOCALAPPDATA", ""), r"Programs\LTX Desktop\LTX Desktop.exe"),
+            r"C:\Program Files\LTX Desktop\LTX Desktop.exe",
+            r"D:\Program Files\LTX Desktop\LTX Desktop.exe",
+            r"E:\Program Files\LTX Desktop\LTX Desktop.exe"
+        ]
+        found = False
+        for p in default_paths:
+            if os.path.exists(p):
+                target_exe = p
+                print(f"\033[96m[INFO] 自动检测到 LTX 原版安装路径: {p}\033[0m")
+                found = True
+                break
+        
+        if not found:
+            print(f"\033[91m[ERROR] 未能找到原版 LTX Desktop 的安装路径！\033[0m")
+            print("请清理 LTX_Shortcut 文件夹，并将您当前电脑上真正的原版快捷方式重贴复制进去。")
+            sys.exit(1)
+        
+    return os.path.dirname(target_exe)
+
+USER_PROFILE = os.path.expanduser("~")
+PYTHON_EXE = os.path.join(USER_PROFILE, r"AppData\Local\LTXDesktop\python\python.exe")
+DATA_DIR = os.path.join(USER_PROFILE, r"AppData\Local\LTXDesktop")
+
+# 1. 动态获取主安装路径
+LTX_INSTALL_DIR = resolve_ltx_path()
+BACKEND_DIR = os.path.join(LTX_INSTALL_DIR, r"resources\backend")
+UI_FILE_NAME = "UI/index.html"
+
+# 环境致命检测：如果官方 Python 还没解压释放，立刻强制中断整个程序
+if not os.path.exists(PYTHON_EXE):
+    print(f"\n\033[1;41m [致命错误] 您的电脑上尚未配置好 LTX 的官方渲染核心框架！ \033[0m")
+    print(f"\033[93m此应用仅是 UI 图形控制台，必需依赖原版软件环境才能生成。在 ({PYTHON_EXE}) 未找到运行引擎。\n")
+    print(">> 解决方案：\n1. 请先在您的电脑上正常安装【LTX Desktop 官方原版软件】。")
+    print("2. 必需：双击打开运行一次原版软件！（运行后原版软件会在后台自动释放环境）")
+    print("3. 把原版软件的快捷方式复制到本文档的 LTX_Shortcut 文件夹里面。")
+    print("4. 全部完成后，再重新启动本 run.bat 脚本即可！\033[0m\n")
+    os._exit(1)
+
+# 2. 从目录读取改动过的 Python 文件 (热修复拦截器)
+PATCHES_DIR = os.path.join(os.getcwd(), "patches")
+os.makedirs(PATCHES_DIR, exist_ok=True)
+
+# 3. 默认输出定向至程序根目录
+LOCAL_OUTPUTS = os.path.join(os.getcwd(), "outputs")
+os.makedirs(LOCAL_OUTPUTS, exist_ok=True)
+
+# 强制注入自定义输出录至 LTX 缓存数据中
+os.makedirs(DATA_DIR, exist_ok=True)
+with open(os.path.join(DATA_DIR, "custom_dir.txt"), 'w', encoding='utf-8') as f:
+    f.write(LOCAL_OUTPUTS)
+
+os.environ["LTX_APP_DATA_DIR"] = DATA_DIR
+
+# 将 patches 目录优先级提升，做到 Python 无损替换
+os.environ["PYTHONPATH"] = f"{PATCHES_DIR};{BACKEND_DIR}"
+
+def get_lan_ip():
+    try:
+        host_name = socket.gethostname()
+        _, _, ip_list = socket.gethostbyname_ex(host_name)
+        
+        candidates = []
+        for ip in ip_list:
+            if ip.startswith("192.168."):
+                return ip
+            elif ip.startswith("10.") or (ip.startswith("172.") and 16 <= int(ip.split('.')[1]) <= 31):
+                candidates.append(ip)
+                
+        if candidates:
+            return candidates[0]
+            
+        # Fallback to the default socket routing approach if no obvious LAN IP found
+        s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
+        s.connect(("8.8.8.8", 80))
+        ip = s.getsockname()[0]
+        s.close()
+        return ip
+    except:
+        return "127.0.0.1"
+
+LAN_IP = get_lan_ip()
+
+# ============================================================
+# 服务启动逻辑
+# ============================================================
+def check_port_in_use(port):
+    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+        return s.connect_ex(('127.0.0.1', port)) == 0
+
+def launch_backend():
+    """启动核心引擎 - 监听 0.0.0.0 确保局域网可调"""
+    if check_port_in_use(3000):
+        print(f"\n\033[1;41m [致命错误] 3000 端口已被占用，无法启动核心引擎！ \033[0m")
+        print("\033[93m>> 绝大多数情况下，这是因为【官方原版 LTX Desktop】正在您的电脑后台运行。\033[0m")
+        print(">> 冲突会导致显存爆炸。请检查右下角系统托盘图标，右键完全退出官方软件。")
+        print(">> 退出后重新双击 run.bat 启动本程序！\n")
+        os._exit(1)
+
+    print(f"\033[96m[CORE] 核心引擎正在启动...\033[0m")
+    # 只开启重要级别的 Python 应用层日志，去除无用的 HTTP 刷屏
+    import logging as _logging
+    _logging.basicConfig(
+        level=_logging.INFO,
+        format="[%(asctime)s] %(levelname)s %(name)s: %(message)s",
+        datefmt="%H:%M:%S",
+        force=True
+    )
+    
+    # 构建绝对无损的环境拦截器：防止其他电脑被 cwd 劫持加载原版文件
+    launcher_code = f"""
+import sys
+import os
+
+patch_dir = r"{PATCHES_DIR}"
+backend_dir = r"{BACKEND_DIR}"
+
+# 防御性清除：强行剥离所有的默认 backend_dir 引用
+sys.path = [p for p in sys.path if p and os.path.normpath(p) != os.path.normpath(backend_dir)]
+sys.path = [p for p in sys.path if p and p != "." and p != ""]
+
+# 绝对插队注入：优先搜索 PATCHES_DIR
+sys.path.insert(0, patch_dir)
+sys.path.insert(1, backend_dir)
+
+import uvicorn
+from ltx2_server import app
+
+if __name__ == '__main__':
+    uvicorn.run(app, host="0.0.0.0", port=3000, log_level="info", access_log=False)
+"""
+    launcher_path = os.path.join(PATCHES_DIR, "launcher.py")
+    with open(launcher_path, "w", encoding="utf-8") as f:
+        f.write(launcher_code)
+
+    cmd = [PYTHON_EXE, launcher_path]
+    env = os.environ.copy()
+    result = subprocess.run(cmd, cwd=BACKEND_DIR, env=env)
+    if result.returncode != 0:
+        print(f"\n\033[1;41m [致命错误] 核心引擎异常崩溃退出！ (Exit Code: {result.returncode})\033[0m")
+        print(">> 请检查上述终端报错信息。确认显卡驱动是否正常。")
+        os._exit(1)
+
+ui_app = FastAPI()
+# 已移除存在安全隐患的静态资源挂载目录
+
+@ui_app.get("/")
+async def serve_index():
+    return FileResponse(os.path.join(os.getcwd(), UI_FILE_NAME))
+
+@ui_app.get("/index.css")
+async def serve_css():
+    return FileResponse(os.path.join(os.getcwd(), "UI/index.css"))
+
+@ui_app.get("/index.js")
+async def serve_js():
+    return FileResponse(os.path.join(os.getcwd(), "UI/index.js"))
+
+def launch_ui_server():
+    print(f"\033[92m[UI] 工作站已就绪！\033[0m")
+    print(f"\033[92m[LOCAL] 本机访问: http://127.0.0.1:4000\033[0m")
+    print(f"\033[93m[WIFI]  局域网访问: http://{LAN_IP}:4000\033[0m")
+    
+    # 彻底压制 WinError 10054 (客户端强制断开) 的底层警告报错
+    if sys.platform == 'win32':
+        # Uvicorn 内部会拉起循环，所以只能通过底层 Logging Filter 拦截控制台噪音
+        class WinError10054Filter(logging.Filter):
+            def filter(self, record):
+                if record.name == 'asyncio' and hasattr(record, 'exc_info') and record.exc_info:
+                    exc_type, exc_value, _ = record.exc_info
+                    if isinstance(exc_value, ConnectionResetError) and getattr(exc_value, 'winerror', None) == 10054:
+                        return False
+                if "10054" in record.getMessage() and "ConnectionResetError" in record.getMessage():
+                    return False
+                return True
+        logging.getLogger("asyncio").addFilter(WinError10054Filter())
+        
+    uvicorn.run(ui_app, host="0.0.0.0", port=4000, log_level="warning", access_log=False)
+
+if __name__ == "__main__":
+    os.system('cls' if os.name == 'nt' else 'clear')
+    print("\033[1;97;44m LTX-2 CINEMATIC WORKSTATION | NETWORK ENABLED \033[0m\n")
+    
+    threading.Thread(target=launch_backend, daemon=True).start()
+    
+    # 强制校验 3000 端口是否存活
+    print("\033[93m[SYS] 正在等待内部核心 3000 端口启动...\033[0m")
+    backend_ready = False
+    for _ in range(30):
+        try:
+            with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+                if s.connect_ex(('127.0.0.1', 3000)) == 0:
+                    backend_ready = True
+                    break
+        except Exception:
+            pass
+        time.sleep(1)
+        
+    if backend_ready:
+        print("\033[92m[SYS] 3000 端口已通过连通性握手验证！后端装载成功。\033[0m")
+    else:
+        print("\033[1;41m [崩坏警告] 等待 30 秒后，3000 端口依然无法连通！ \033[0m")
+        print(">> Uvicorn 可能在后台陷入了死锁，或者被防火墙拦截，前端大概率将无法连接到后端！")
+        print(">> 请检查上方是否有 Python 报错。\n")
+        
+    try:
+        launch_ui_server()
+    except KeyboardInterrupt:
+        sys.exit(0)

LTX2.3/patches/api_types.py

@@ -0,0 +1,306 @@
+"""Pydantic request/response models and TypedDicts for ltx2_server."""
+
+from __future__ import annotations
+
+from typing import Literal, NamedTuple, TypeAlias, TypedDict
+from typing import Annotated
+
+from pydantic import BaseModel, Field, StringConstraints
+
+NonEmptyPrompt = Annotated[str, StringConstraints(strip_whitespace=True, min_length=1)]
+ModelFileType = Literal[
+    "checkpoint",
+    "upsampler",
+    "distilled_lora",
+    "ic_lora",
+    "depth_processor",
+    "person_detector",
+    "pose_processor",
+    "text_encoder",
+    "zit",
+]
+
+
+class ImageConditioningInput(NamedTuple):
+    """Image conditioning triplet used by all video pipelines."""
+
+    path: str
+    frame_idx: int
+    strength: float
+
+
+# ============================================================
+# TypedDicts for module-level state globals
+# ============================================================
+
+
+class GenerationState(TypedDict):
+    id: str | None
+    cancelled: bool
+    result: str | list[str] | None
+    error: str | None
+    status: str  # "idle" | "running" | "complete" | "cancelled" | "error"
+    phase: str
+    progress: int
+    current_step: int
+    total_steps: int
+
+
+JsonObject: TypeAlias = dict[str, object]
+VideoCameraMotion = Literal[
+    "none",
+    "dolly_in",
+    "dolly_out",
+    "dolly_left",
+    "dolly_right",
+    "jib_up",
+    "jib_down",
+    "static",
+    "focus_shift",
+]
+
+
+# ============================================================
+# Response Models
+# ============================================================
+
+
+class ModelStatusItem(BaseModel):
+    id: str
+    name: str
+    loaded: bool
+    downloaded: bool
+
+
+class GpuTelemetry(BaseModel):
+    name: str
+    vram: int
+    vramUsed: int
+
+
+class HealthResponse(BaseModel):
+    status: str
+    models_loaded: bool
+    active_model: str | None
+    gpu_info: GpuTelemetry
+    sage_attention: bool
+    models_status: list[ModelStatusItem]
+
+
+class GpuInfoResponse(BaseModel):
+    cuda_available: bool
+    mps_available: bool = False
+    gpu_available: bool = False
+    gpu_name: str | None
+    vram_gb: int | None
+    gpu_info: GpuTelemetry
+
+
+class RuntimePolicyResponse(BaseModel):
+    force_api_generations: bool
+
+
+class GenerationProgressResponse(BaseModel):
+    status: str
+    phase: str
+    progress: int
+    currentStep: int | None
+    totalSteps: int | None
+
+
+class ModelInfo(BaseModel):
+    id: str
+    name: str
+    description: str
+
+
+class ModelFileStatus(BaseModel):
+    id: ModelFileType
+    name: str
+    description: str
+    downloaded: bool
+    size: int
+    expected_size: int
+    required: bool = True
+    is_folder: bool = False
+    optional_reason: str | None = None
+
+
+class TextEncoderStatus(BaseModel):
+    downloaded: bool
+    size_bytes: int
+    size_gb: float
+    expected_size_gb: float
+
+
+class ModelsStatusResponse(BaseModel):
+    models: list[ModelFileStatus]
+    all_downloaded: bool
+    total_size: int
+    downloaded_size: int
+    total_size_gb: float
+    downloaded_size_gb: float
+    models_path: str
+    has_api_key: bool
+    text_encoder_status: TextEncoderStatus
+    use_local_text_encoder: bool
+
+
+class DownloadProgressResponse(BaseModel):
+    status: str
+    current_downloading_file: ModelFileType | None
+    current_file_progress: int
+    total_progress: int
+    total_downloaded_bytes: int
+    expected_total_bytes: int
+    completed_files: set[ModelFileType]
+    all_files: set[ModelFileType]
+    error: str | None
+    speed_mbps: int
+
+
+class SuggestGapPromptResponse(BaseModel):
+    status: str = "success"
+    suggested_prompt: str
+
+
+class GenerateVideoResponse(BaseModel):
+    status: str
+    video_path: str | None = None
+
+
+class GenerateImageResponse(BaseModel):
+    status: str
+    image_paths: list[str] | None = None
+
+
+class CancelResponse(BaseModel):
+    status: str
+    id: str | None = None
+
+
+class RetakeResponse(BaseModel):
+    status: str
+    video_path: str | None = None
+    result: JsonObject | None = None
+
+
+class IcLoraExtractResponse(BaseModel):
+    conditioning: str
+    original: str
+    conditioning_type: Literal["canny", "depth"]
+    frame_time: float
+
+
+class IcLoraGenerateResponse(BaseModel):
+    status: str
+    video_path: str | None = None
+
+
+class ModelDownloadStartResponse(BaseModel):
+    status: str
+    message: str | None = None
+    sessionId: str | None = None
+
+
+class TextEncoderDownloadResponse(BaseModel):
+    status: str
+    message: str | None = None
+    sessionId: str | None = None
+
+
+class StatusResponse(BaseModel):
+    status: str
+
+
+class ErrorResponse(BaseModel):
+    error: str
+    message: str | None = None
+
+
+# ============================================================
+# Request Models
+# ============================================================
+
+
+class GenerateVideoRequest(BaseModel):
+    prompt: NonEmptyPrompt
+    resolution: str = "512p"
+    model: str = "fast"
+    cameraMotion: VideoCameraMotion = "none"
+    negativePrompt: str = ""
+    duration: str = "2"
+    fps: str = "24"
+    audio: str = "false"
+    imagePath: str | None = None
+    audioPath: str | None = None
+    startFramePath: str | None = None
+    endFramePath: str | None = None
+    aspectRatio: Literal["16:9", "9:16"] = "16:9"
+
+
+class GenerateImageRequest(BaseModel):
+    prompt: NonEmptyPrompt
+    width: int = 1024
+    height: int = 1024
+    numSteps: int = 4
+    numImages: int = 1
+
+
+def _default_model_types() -> set[ModelFileType]:
+    return set()
+
+
+class ModelDownloadRequest(BaseModel):
+    modelTypes: set[ModelFileType] = Field(default_factory=_default_model_types)
+
+
+class RequiredModelsResponse(BaseModel):
+    modelTypes: list[ModelFileType]
+
+
+class SuggestGapPromptRequest(BaseModel):
+    beforePrompt: str = ""
+    afterPrompt: str = ""
+    beforeFrame: str | None = None
+    afterFrame: str | None = None
+    gapDuration: float = 5
+    mode: str = "t2v"
+    inputImage: str | None = None
+
+
+class RetakeRequest(BaseModel):
+    video_path: str
+    start_time: float = 0
+    duration: float = 0
+    prompt: str = ""
+    mode: str = "replace_video_only"
+    width: int | None = None
+    height: int | None = None
+
+
+class IcLoraExtractRequest(BaseModel):
+    video_path: str
+    conditioning_type: Literal["canny", "depth"] = "canny"
+    frame_time: float = 0
+
+
+class IcLoraImageInput(BaseModel):
+    path: str
+    frame: int = 0
+    strength: float = 1.0
+
+
+def _default_ic_lora_images() -> list[IcLoraImageInput]:
+    return []
+
+
+class IcLoraGenerateRequest(BaseModel):
+    video_path: str
+    conditioning_type: Literal["canny", "depth"]
+    prompt: NonEmptyPrompt
+    conditioning_strength: float = 1.0
+    num_inference_steps: int = 30
+    cfg_guidance_scale: float = 1.0
+    negative_prompt: str = ""
+    images: list[IcLoraImageInput] = Field(default_factory=_default_ic_lora_images)

LTX2.3/patches/app_factory.py

@@ -0,0 +1,1396 @@
+"""FastAPI app factory decoupled from runtime bootstrap side effects."""
+
+from __future__ import annotations
+
+import base64
+import hmac
+import os
+
+# 防 OOM 与显存碎片化补丁：在 torch 初始化之前注入环境变量
+os.environ["PYTORCH_ALLOC_CONF"] = "expandable_segments:True"
+import torch  # 提升到顶层导入
+from collections.abc import Awaitable, Callable
+from typing import TYPE_CHECKING
+from pathlib import Path  # 必须导入，用于处理 Windows 路径
+
+from fastapi import FastAPI, Request, UploadFile, File
+from fastapi.exceptions import RequestValidationError
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.responses import JSONResponse
+from pydantic import ConfigDict
+from fastapi.staticfiles import StaticFiles  # 必须导入，用于挂载静态目录
+from starlette.responses import Response as StarletteResponse
+import shutil
+import tempfile
+import time
+from api_types import ImageConditioningInput, GenerateVideoRequest
+
+from _routes._errors import HTTPError
+from _routes.generation import router as generation_router
+from _routes.health import router as health_router
+from _routes.ic_lora import router as ic_lora_router
+from _routes.image_gen import router as image_gen_router
+from _routes.models import router as models_router
+from _routes.suggest_gap_prompt import router as suggest_gap_prompt_router
+from _routes.retake import router as retake_router
+from _routes.runtime_policy import router as runtime_policy_router
+from _routes.settings import router as settings_router
+from logging_policy import log_http_error, log_unhandled_exception
+from state import init_state_service
+
+if TYPE_CHECKING:
+    from app_handler import AppHandler
+
+# 跨域配置：允许所有来源，解决本地网页调用限制
+DEFAULT_ALLOWED_ORIGINS: list[str] = ["*"]
+
+
+def _extend_generate_video_request_model() -> None:
+    """Keep custom video fields working across upstream request-model changes."""
+    annotations = dict(getattr(GenerateVideoRequest, "__annotations__", {}))
+    changed = False
+
+    for field_name in ("startFramePath", "endFramePath"):
+        if field_name not in annotations:
+            annotations[field_name] = str | None
+            setattr(GenerateVideoRequest, field_name, None)
+            changed = True
+
+    if changed:
+        GenerateVideoRequest.__annotations__ = annotations
+
+    existing_config = dict(getattr(GenerateVideoRequest, "model_config", {}) or {})
+    if existing_config.get("extra") != "allow":
+        existing_config["extra"] = "allow"
+        GenerateVideoRequest.model_config = ConfigDict(**existing_config)
+        changed = True
+
+    if changed:
+        GenerateVideoRequest.model_rebuild(force=True)
+
+
+def create_app(
+    *,
+    handler: "AppHandler",
+    allowed_origins: list[str] | None = None,
+    title: str = "LTX-2 Video Generation Server",
+    auth_token: str = "",
+    admin_token: str = "",
+) -> FastAPI:
+    """Create a configured FastAPI app bound to the provided handler."""
+    init_state_service(handler)
+    _extend_generate_video_request_model()
+
+    app = FastAPI(title=title)
+    app.state.admin_token = admin_token  # type: ignore[attr-defined]
+
+    # 彻底压制 WinError 10054 (客户端强制断开) 的底层警告报错
+    import sys, asyncio
+
+    if sys.platform == "win32":
+        try:
+            loop = asyncio.get_event_loop()
+
+            def silence_winerror_10054(loop, context):
+                exc = context.get("exception")
+                if (
+                    isinstance(exc, ConnectionResetError)
+                    and getattr(exc, "winerror", None) == 10054
+                ):
+                    return
+                loop.default_exception_handler(context)
+
+            loop.set_exception_handler(silence_winerror_10054)
+        except Exception:
+            pass
+
+    # --- 核心修复：对准 LTX 真正的输出目录 (AppData) ---
+    def get_dynamic_output_path():
+        base_dir = (
+            Path(os.environ.get("LOCALAPPDATA", os.path.expanduser("~/AppData/Local")))
+            / "LTXDesktop"
+        ).resolve()
+        config_file = base_dir / "custom_dir.txt"
+        if config_file.exists():
+            try:
+                custom_dir = config_file.read_text(encoding="utf-8").strip()
+                if custom_dir:
+                    p = Path(custom_dir)
+                    p.mkdir(parents=True, exist_ok=True)
+                    return p
+            except Exception:
+                pass
+        default_dir = base_dir / "outputs"
+        default_dir.mkdir(parents=True, exist_ok=True)
+        return default_dir
+
+    actual_output_path = get_dynamic_output_path()
+    handler.config.outputs_dir = actual_output_path
+
+    upload_tmp_path = actual_output_path / "uploads"
+
+    # 如果文件夹不存在则创建，防止挂载失败
+    if not actual_output_path.exists():
+        actual_output_path.mkdir(parents=True, exist_ok=True)
+    if not upload_tmp_path.exists():
+        upload_tmp_path.mkdir(parents=True, exist_ok=True)
+
+    # 挂载静态服务：将该目录映射到 http://127.0.0.1:3000/outputs
+    app.mount(
+        "/outputs", StaticFiles(directory=str(actual_output_path)), name="outputs"
+    )
+    # -----------------------------------------------
+
+    # 配置 CORS
+    app.add_middleware(
+        CORSMiddleware,
+        allow_origins=allowed_origins or DEFAULT_ALLOWED_ORIGINS,
+        allow_methods=["*"],
+        allow_headers=["*"],
+    )
+
+    # === [全局隔离补丁] ===
+    # 强制将每一个新的 HTTP 线程/协程请求的默认显卡都强绑定到用户选定的设备上
+    @app.middleware("http")
+    async def _sync_gpu_middleware(
+        request: Request,
+        call_next: Callable[[Request], Awaitable[StarletteResponse]],
+    ) -> StarletteResponse:
+        import torch
+
+        if (
+            torch.cuda.is_available()
+            and getattr(handler.config.device, "type", "") == "cuda"
+        ):
+            idx = handler.config.device.index
+            if idx is not None:
+                # 能够强行夺取那些底层写死了 cuda:0 而忽略 config.device 的第三方库
+                torch.cuda.set_device(idx)
+        return await call_next(request)
+
+    # 认证中间件
+    @app.middleware("http")
+    async def _auth_middleware(
+        request: Request,
+        call_next: Callable[[Request], Awaitable[StarletteResponse]],
+    ) -> StarletteResponse:
+        # 关键修复：如果是获取生成的图片，直接放行，不检查 Token
+        if (
+            request.url.path.startswith("/outputs")
+            or request.url.path == "/api/system/upload-image"
+        ):
+            return await call_next(request)
+
+        if not auth_token:
+            return await call_next(request)
+        if request.method == "OPTIONS":
+            return await call_next(request)
+
+        def _token_matches(candidate: str) -> bool:
+            return hmac.compare_digest(candidate, auth_token)
+
+        # WebSocket 认证
+        if request.headers.get("upgrade", "").lower() == "websocket":
+            if _token_matches(request.query_params.get("token", "")):
+                return await call_next(request)
+            return JSONResponse(status_code=401, content={"error": "Unauthorized"})
+
+        # HTTP 认证 (Bearer/Basic)
+        auth_header = request.headers.get("authorization", "")
+        if auth_header.startswith("Bearer ") and _token_matches(auth_header[7:]):
+            return await call_next(request)
+        if auth_header.startswith("Basic "):
+            try:
+                decoded = base64.b64decode(auth_header[6:]).decode()
+                _, _, password = decoded.partition(":")
+                if _token_matches(password):
+                    return await call_next(request)
+            except Exception:
+                pass
+        return JSONResponse(status_code=401, content={"error": "Unauthorized"})
+
+    # 异常处理逻辑
+    _FALLBACK = "An unexpected error occurred"
+
+    async def _route_http_error_handler(
+        request: Request, exc: Exception
+    ) -> JSONResponse:
+        if isinstance(exc, HTTPError):
+            log_http_error(request, exc)
+            return JSONResponse(
+                status_code=exc.status_code, content={"error": exc.detail or _FALLBACK}
+            )
+        return JSONResponse(status_code=500, content={"error": str(exc) or _FALLBACK})
+
+    async def _validation_error_handler(
+        request: Request, exc: Exception
+    ) -> JSONResponse:
+        if isinstance(exc, RequestValidationError):
+            return JSONResponse(
+                status_code=422, content={"error": str(exc) or _FALLBACK}
+            )
+        return JSONResponse(status_code=422, content={"error": str(exc) or _FALLBACK})
+
+    async def _route_generic_error_handler(
+        request: Request, exc: Exception
+    ) -> JSONResponse:
+        log_unhandled_exception(request, exc)
+        return JSONResponse(status_code=500, content={"error": str(exc) or _FALLBACK})
+
+    app.add_exception_handler(RequestValidationError, _validation_error_handler)
+    app.add_exception_handler(HTTPError, _route_http_error_handler)
+    app.add_exception_handler(Exception, _route_generic_error_handler)
+
+    # --- 系统功能接口 ---
+    @app.post("/api/system/clear-gpu")
+    async def route_clear_gpu():
+        try:
+            import torch
+            import gc
+            import asyncio
+
+            # 1. 尝试终止任务并重置运行状态
+            if getattr(handler.generation, "is_generation_running", lambda: False)():
+                try:
+                    handler.generation.cancel_generation()
+                except Exception:
+                    pass
+                await asyncio.sleep(0.5)
+
+            # 暴力重置死锁状态
+            if hasattr(handler.generation, "_generation_id"):
+                handler.generation._generation_id = None
+            if hasattr(handler.generation, "_is_generating"):
+                handler.generation._is_generating = False
+
+            # 2. 强制卸载模型: 临时屏蔽底层锁定器
+            try:
+                mock_swapped = False
+                orig_running = None
+                if hasattr(handler.pipelines, "_generation_service"):
+                    orig_running = (
+                        handler.pipelines._generation_service.is_generation_running
+                    )
+                    handler.pipelines._generation_service.is_generation_running = (
+                        lambda: False
+                    )
+                    mock_swapped = True
+                try:
+                    handler.pipelines.unload_gpu_pipeline()
+                finally:
+                    if mock_swapped:
+                        handler.pipelines._generation_service.is_generation_running = (
+                            orig_running
+                        )
+            except Exception as e:
+                print(f"Force unload warning: {e}")
+
+            # 3. 深度清理
+            gc.collect()
+            if torch.cuda.is_available():
+                torch.cuda.empty_cache()
+                torch.cuda.ipc_collect()
+            return {
+                "status": "success",
+                "message": "GPU memory cleared and models unloaded",
+            }
+        except Exception as e:
+            return JSONResponse(status_code=500, content={"error": str(e)})
+
+    @app.post("/api/system/reset-state")
+    async def route_reset_state():
+        """轻量级状态重置：只清除 generation 状态锁，不卸载 GPU 管线。
+        在每次新渲染开始前由前端调用，确保后端状态干净可用。"""
+        try:
+            gen = handler.generation
+            # 强制清除所有可能导致 is_generation_running() 返回 True 的标志
+            for attr in (
+                "_is_generating",
+                "_generation_id",
+                "_cancelled",
+                "_is_cancelled",
+            ):
+                if hasattr(gen, attr):
+                    if attr in ("_is_generating", "_cancelled", "_is_cancelled"):
+                        setattr(gen, attr, False)
+                    else:
+                        setattr(gen, attr, None)
+            # 某些实现用 threading.Event
+            for attr in ("_cancel_event",):
+                if hasattr(gen, attr):
+                    try:
+                        getattr(gen, attr).clear()
+                    except Exception:
+                        pass
+            print("[reset-state] Generation state has been reset cleanly.")
+            return {"status": "success", "message": "Generation state reset"}
+        except Exception as e:
+            import traceback
+
+            traceback.print_exc()
+            return JSONResponse(status_code=500, content={"error": str(e)})
+
+    @app.post("/api/system/set-dir")
+    async def route_set_dir(request: Request):
+        try:
+            data = await request.json()
+            new_dir = data.get("directory", "").strip()
+            base_dir = (
+                Path(
+                    os.environ.get(
+                        "LOCALAPPDATA", os.path.expanduser("~/AppData/Local")
+                    )
+                )
+                / "LTXDesktop"
+            ).resolve()
+            config_file = base_dir / "custom_dir.txt"
+            if new_dir:
+                p = Path(new_dir)
+                p.mkdir(parents=True, exist_ok=True)
+                config_file.write_text(new_dir, encoding="utf-8")
+            else:
+                if config_file.exists():
+                    config_file.unlink()
+            # 立即更新全局 config 控制
+            handler.config.outputs_dir = get_dynamic_output_path()
+            return {"status": "success", "directory": str(get_dynamic_output_path())}
+        except Exception as e:
+            return JSONResponse(status_code=500, content={"error": str(e)})
+
+    @app.get("/api/system/get-dir")
+    async def route_get_dir():
+        return {"status": "success", "directory": str(get_dynamic_output_path())}
+
+    @app.get("/api/system/browse-dir")
+    async def route_browse_dir():
+        try:
+            import subprocess
+
+            # 强制将对话框置顶层：通过 STA 线程 + Topmost 属性，避免被窗口锥入后台
+            ps_script = (
+                "[System.Reflection.Assembly]::LoadWithPartialName('System.Windows.Forms') | Out-Null;"
+                "[System.Reflection.Assembly]::LoadWithPartialName('System.Drawing') | Out-Null;"
+                "$f = New-Object System.Windows.Forms.FolderBrowserDialog;"
+                "$f.Description = '\u9009\u62e9 LTX \u89c6\u9891\u548c\u56fe\u50cf\u751f\u6210\u7684\u5168\u5c40\u8f93\u51fa\u76ee\u5f55';"
+                "$f.ShowNewFolderButton = $true;"
+                # 创建一个雐形助手窗口作为 parent 确保对话框在最顶层
+                "$owner = New-Object System.Windows.Forms.Form;"
+                "$owner.TopMost = $true;"
+                "$owner.StartPosition = 'CenterScreen';"
+                "$owner.Size = New-Object System.Drawing.Size(1, 1);"
+                "$owner.Show();"
+                "$owner.BringToFront();"
+                "$owner.Focus();"
+                "if ($f.ShowDialog($owner) -eq 'OK') { echo $f.SelectedPath };"
+                "$owner.Dispose();"
+            )
+
+            def run_ps():
+                process = subprocess.Popen(
+                    ["powershell", "-STA", "-NoProfile", "-Command", ps_script],
+                    stdout=subprocess.PIPE,
+                    stderr=subprocess.PIPE,
+                    text=True,
+                    # 移除 CREATE_NO_WINDOW 以允许 UI 线程正常弹出
+                )
+                stdout, _ = process.communicate()
+                return stdout.strip()
+
+            from starlette.concurrency import run_in_threadpool
+
+            selected_dir = await run_in_threadpool(run_ps)
+            return {"status": "success", "directory": selected_dir}
+        except Exception as e:
+            return JSONResponse(status_code=500, content={"error": str(e)})
+
+    @app.get("/api/system/file")
+    async def route_serve_file(path: str):
+        from fastapi.responses import FileResponse
+
+        if os.path.exists(path):
+            return FileResponse(path)
+        return JSONResponse(status_code=404, content={"error": "File not found"})
+
+    @app.get("/api/system/list-gpus")
+    async def route_list_gpus():
+        try:
+            import torch
+
+            gpus = []
+            if torch.cuda.is_available():
+                current_idx = 0
+                dev = getattr(handler.config, "device", None)
+                if dev is not None and getattr(dev, "index", None) is not None:
+                    current_idx = dev.index
+                for i in range(torch.cuda.device_count()):
+                    try:
+                        name = torch.cuda.get_device_name(i)
+                    except Exception:
+                        name = f"GPU {i}"
+                    try:
+                        vram_bytes = torch.cuda.get_device_properties(i).total_memory
+                        vram_gb = vram_bytes / (1024**3)
+                        vram_mb = vram_bytes / (1024**2)
+                    except Exception:
+                        vram_gb = 0.0
+                        vram_mb = 0
+                    gpus.append(
+                        {
+                            "id": i,
+                            "name": name,
+                            "vram": f"{vram_gb:.1f} GB",
+                            "vram_mb": int(vram_mb),
+                            "active": (i == current_idx),
+                        }
+                    )
+            return {"status": "success", "gpus": gpus}
+        except Exception as e:
+            return JSONResponse(status_code=500, content={"error": str(e)})
+
+    @app.post("/api/system/switch-gpu")
+    async def route_switch_gpu(request: Request):
+        try:
+            import torch
+            import gc
+            import asyncio
+
+            data = await request.json()
+            gpu_id = data.get("gpu_id")
+
+            if (
+                gpu_id is None
+                or not torch.cuda.is_available()
+                or gpu_id >= torch.cuda.device_count()
+            ):
+                return JSONResponse(
+                    status_code=400, content={"error": "Invalid GPU ID"}
+                )
+
+            # 先尝试终止任何可能的卡死任务
+            if getattr(handler.generation, "is_generation_running", lambda: False)():
+                try:
+                    handler.generation.cancel_generation()
+                except Exception:
+                    pass
+                await asyncio.sleep(0.5)
+            if hasattr(handler.generation, "_generation_id"):
+                handler.generation._generation_id = None
+            if hasattr(handler.generation, "_is_generating"):
+                handler.generation._is_generating = False
+
+            # 1. 卸载当前 GPU 上的模型: 临时屏蔽底层锁定器
+            try:
+                mock_swapped = False
+                orig_running = None
+                if hasattr(handler.pipelines, "_generation_service"):
+                    orig_running = (
+                        handler.pipelines._generation_service.is_generation_running
+                    )
+                    handler.pipelines._generation_service.is_generation_running = (
+                        lambda: False
+                    )
+                    mock_swapped = True
+                try:
+                    handler.pipelines.unload_gpu_pipeline()
+                finally:
+                    if mock_swapped:
+                        handler.pipelines._generation_service.is_generation_running = (
+                            orig_running
+                        )
+            except Exception:
+                pass
+            gc.collect()
+            torch.cuda.empty_cache()
+
+            # 2. 切换全局设备配置
+            new_device = torch.device(f"cuda:{gpu_id}")
+            handler.config.device = new_device
+
+            # 3. 核心修复：设置当前进程的默认 CUDA 设备
+            # 这会影响到 torch.cuda.current_device() 和后续的模型加载
+            torch.cuda.set_device(gpu_id)
+
+            # 针对底层库可能直接读取 CUDA_VISIBLE_DEVICES 的情况
+            # 注意：torch 初始化后修改此变量不一定生效，但对某些库可能有引导作用
+            os.environ["CUDA_VISIBLE_DEVICES"] = str(gpu_id)
+
+            # 4. 【核心修复】同步更新 TextEncoder 的设备指针
+            # 根本原因: LTXTextEncoder.self.device 在初始化时硬绑定了旧 GPU，
+            # 切换设备后 text context 仍在旧 GPU 上，与已迁移到新 GPU 的
+            # Transformer 产生 "cuda:0 and cuda:1" 设备不一致冲突。
+            try:
+                te_state = None
+                # 尝试多种路径访问 text_encoder 状态
+                if hasattr(handler, "state") and hasattr(handler.state, "text_encoder"):
+                    te_state = handler.state.text_encoder
+                elif hasattr(handler, "_state") and hasattr(
+                    handler._state, "text_encoder"
+                ):
+                    te_state = handler._state.text_encoder
+
+                if te_state is not None:
+                    # 4a. 更新 LTXTextEncoder 服务自身的 device 属性
+                    if hasattr(te_state, "service") and hasattr(
+                        te_state.service, "device"
+                    ):
+                        te_state.service.device = new_device
+                        print(f"[TextEncoder] device updated to {new_device}")
+
+                    # 4b. 将缓存的 encoder 权重迁移到 CPU，下次推理时再按新设备重加载
+                    if (
+                        hasattr(te_state, "cached_encoder")
+                        and te_state.cached_encoder is not None
+                    ):
+                        try:
+                            te_state.cached_encoder.to(torch.device("cpu"))
+                        except Exception:
+                            pass
+                        te_state.cached_encoder = None
+                        print(
+                            "[TextEncoder] cached encoder cleared (will reload on new GPU)"
+                        )
+
+                    # 4c. 清除 API embeddings 缓存（tensor 绑定旧 GPU）
+                    if hasattr(te_state, "api_embeddings"):
+                        te_state.api_embeddings = None
+
+                    # 4d. 清除 prompt cache（其中 tensor 也绑定旧 GPU）
+                    if hasattr(te_state, "prompt_cache") and te_state.prompt_cache:
+                        te_state.prompt_cache.clear()
+                        print("[TextEncoder] prompt cache cleared")
+            except Exception as _te_err:
+                print(f"[TextEncoder] device sync warning (non-fatal): {_te_err}")
+
+            print(
+                f"Switched active GPU to: {torch.cuda.get_device_name(gpu_id)} (ID: {gpu_id})"
+            )
+            return {"status": "success", "message": f"Switched to GPU {gpu_id}"}
+        except Exception as e:
+            return JSONResponse(status_code=500, content={"error": str(e)})
+
+    # --- 核心增强：首尾帧插值与视频超分支持 ---
+    from handlers.video_generation_handler import VideoGenerationHandler
+    from services.retake_pipeline.ltx_retake_pipeline import LTXRetakePipeline
+    from server_utils.media_validation import normalize_optional_path
+    from PIL import Image
+
+    # 1. 增强插值功能 (Monkey Patch VideoGenerationHandler)
+    _orig_generate = VideoGenerationHandler.generate
+    _orig_generate_video = VideoGenerationHandler.generate_video
+
+    def patched_generate(self, req: GenerateVideoRequest):
+        # === [DEBUG] 打印当前生成状态 ===
+        gen = self._generation
+        is_running = (
+            gen.is_generation_running()
+            if hasattr(gen, "is_generation_running")
+            else "?方法不存在"
+        )
+        gen_id = getattr(gen, "_generation_id", "?属性不存在")
+        is_gen = getattr(gen, "_is_generating", "?属性不存在")
+        cancelled = getattr(
+            gen, "_cancelled", getattr(gen, "_is_cancelled", "?属性不存在")
+        )
+        print(f"\n[PATCH][patched_generate] ==> 收到新请求")
+        print(f"  is_generation_running() = {is_running}")
+        print(f"  _generation_id          = {gen_id}")
+        print(f"  _is_generating          = {is_gen}")
+        print(f"  _cancelled              = {cancelled}")
+        start_frame_path = normalize_optional_path(getattr(req, "startFramePath", None))
+        end_frame_path = normalize_optional_path(getattr(req, "endFramePath", None))
+        aspect_ratio = getattr(req, "aspectRatio", None)
+        print(f"  startFramePath          = {start_frame_path}")
+        print(f"  endFramePath            = {end_frame_path}")
+        print(f"  aspectRatio             = {aspect_ratio}")
+
+        # 检查是否有音频
+        audio_path = normalize_optional_path(getattr(req, "audioPath", None))
+        print(f"[PATCH] audio_path = {audio_path}")
+
+        # 检查是否有图片（图生视频）
+        image_path = normalize_optional_path(getattr(req, "imagePath", None))
+        print(f"[PATCH] image_path = {image_path}")
+
+        # 始终使用自定义逻辑（支持首尾帧和竖屏）
+        print(f"[PATCH] 使用自定义逻辑处理")
+
+        # 计算分辨率
+        import uuid
+
+        resolution = req.resolution
+        duration = int(float(req.duration))
+        fps = int(float(req.fps))
+
+        # 720p 分辨率：横屏 1280x720，竖屏 720x1280
+        RESOLUTION_MAP = {
+            "540p": (960, 540),
+            "720p": (1280, 720),
+            "1080p": (1920, 1080),
+        }
+
+        def get_16_9_size(res):
+            return RESOLUTION_MAP.get(res, (1280, 720))
+
+        def get_9_16_size(res):
+            w, h = get_16_9_size(res)
+            return h, w  # 交换宽高
+
+        if req.aspectRatio == "9:16":
+            width, height = get_9_16_size(resolution)
+        else:
+            width, height = get_16_9_size(resolution)
+
+        # 计算帧数
+        num_frames = ((duration * fps) // 8) * 8 + 1
+        num_frames = max(num_frames, 9)
+
+        print(f"[PATCH] 计算得到的分辨率: {width}x{height}, 帧数: {num_frames}")
+
+        # 设置首尾帧路径
+        self._start_frame_path = start_frame_path
+        self._end_frame_path = end_frame_path
+
+        # 无论有没有音频，都使用自定义逻辑支持首尾帧
+        try:
+            result = patched_generate_video(
+                self,
+                prompt=req.prompt,
+                image=None,
+                image_path=image_path,
+                height=height,
+                width=width,
+                num_frames=num_frames,
+                fps=fps,
+                seed=self._resolve_seed(),
+                camera_motion=req.cameraMotion,
+                negative_prompt=req.negativePrompt,
+                audio_path=audio_path,
+            )
+            print(f"[PATCH][patched_generate] <== 完成, 返回状态: complete")
+            return type("Response", (), {"status": "complete", "video_path": result})()
+        except Exception as e:
+            import traceback
+
+            print(f"[PATCH][patched_generate] 错误: {e}")
+            traceback.print_exc()
+            raise
+
+    def patched_generate_video(
+        self,
+        prompt,
+        image,
+        image_path=None,
+        height=None,
+        width=None,
+        num_frames=None,
+        fps=None,
+        seed=None,
+        camera_motion=None,
+        negative_prompt=None,
+        audio_path=None,
+    ):
+        # === [DEBUG] 打印当前生成状态 ===
+        gen = self._generation
+        is_running = (
+            gen.is_generation_running()
+            if hasattr(gen, "is_generation_running")
+            else "?方法不存在"
+        )
+        gen_id = getattr(gen, "_generation_id", "?属性不存在")
+        is_gen = getattr(gen, "_is_generating", "?属性不存在")
+        print(f"[PATCH][patched_generate_video] ==> 开始推理")
+        print(f"  is_generation_running() = {is_running}")
+        print(f"  _generation_id          = {gen_id}")
+        print(f"  _is_generating          = {is_gen}")
+        print(
+            f"  resolution              = {width}x{height}, frames={num_frames}, fps={fps}"
+        )
+        print(f"  image param             = {type(image)}, {image is not None}")
+        print(f"  image_path              = {image_path}")
+        # ==================================
+        from ltx_pipelines.utils.args import (
+            ImageConditioningInput as LtxImageConditioningInput,
+        )
+
+        images_inputs = []
+        temp_paths = []
+        start_path = getattr(self, "_start_frame_path", None)
+        end_path = getattr(self, "_end_frame_path", None)
+        print(f"[PATCH] start_path={start_path}, end_path={end_path}")
+
+        # 如果没有首尾帧但有 image_path，使用 image_path 作为起始帧
+        if not start_path and not end_path and image_path:
+            print(f"[PATCH] 使用 image_path 作为起始帧: {image_path}")
+            start_path = image_path
+
+        # 检查是否有来自 imagePath 的数据（当只用首帧时）
+        has_image_param = image is not None
+        if has_image_param:
+            print(f"[PATCH] image param is available, will be used as start frame")
+
+        latent_num_frames = (num_frames - 1) // 8 + 1
+        last_latent_idx = latent_num_frames - 1
+        print(
+            f"[PATCH] latent_num_frames={latent_num_frames}, last_latent_idx={last_latent_idx}"
+        )
+
+        target_start_path = start_path if start_path else None
+        if not target_start_path and image is not None:
+            tmp = tempfile.NamedTemporaryFile(suffix=".png", delete=False).name
+            image.save(tmp)
+            temp_paths.append(tmp)
+            target_start_path = tmp
+            print(f"[PATCH] Using image param as start frame: {target_start_path}")
+
+        if target_start_path:
+            start_img = self._prepare_image(target_start_path, width, height)
+            tmp = tempfile.NamedTemporaryFile(suffix=".png", delete=False).name
+            start_img.save(tmp)
+            temp_paths.append(tmp)
+            # 转换 Windows 路径为正斜杠
+            tmp_normalized = tmp.replace("\\", "/")
+            images_inputs.append(
+                LtxImageConditioningInput(
+                    path=tmp_normalized, frame_idx=0, strength=1.0
+                )
+            )
+            print(f"[PATCH] Added start frame: {tmp_normalized}, frame_idx=0")
+
+        if end_path:
+            end_img = self._prepare_image(end_path, width, height)
+            tmp = tempfile.NamedTemporaryFile(suffix=".png", delete=False).name
+            end_img.save(tmp)
+            temp_paths.append(tmp)
+            # 转换 Windows 路径为正斜杠
+            tmp_normalized = tmp.replace("\\", "/")
+            images_inputs.append(
+                LtxImageConditioningInput(
+                    path=tmp_normalized, frame_idx=last_latent_idx, strength=1.0
+                )
+            )
+            print(
+                f"[PATCH] Added end frame: {tmp_normalized}, frame_idx={last_latent_idx}"
+            )
+
+        print(f"[PATCH] images_inputs count: {len(images_inputs)}")
+        if images_inputs:
+            for idx, img in enumerate(images_inputs):
+                print(
+                    f"[PATCH] images_inputs[{idx}]: path={getattr(img, 'path', 'N/A')}, frame_idx={getattr(img, 'frame_idx', 'N/A')}, strength={getattr(img, 'strength', 'N/A')}"
+                )
+
+        print(f"[PATCH] audio_path = {audio_path}")
+
+        if self._generation.is_generation_cancelled():
+            raise RuntimeError("Generation was cancelled")
+
+        # 导入 uuid
+        import uuid
+
+        generation_id = uuid.uuid4().hex[:8]
+
+        # 根据是否有音频选择不同的 pipeline
+        if audio_path:
+            print(f"[PATCH] 加载 A2V pipeline（支持音频）")
+            pipeline_state = self._pipelines.load_a2v_pipeline()
+            num_inference_steps = 11  # A2V 需要这个参数
+        else:
+            print(f"[PATCH] 加载 Fast pipeline")
+            pipeline_state = self._pipelines.load_gpu_pipeline(
+                "fast", should_warm=False
+            )
+            num_inference_steps = None
+
+        # 启动 generation 状态（在 pipeline 加载之后）
+        self._generation.start_generation(generation_id)
+
+        # 处理 negative_prompt
+        neg_prompt = (
+            negative_prompt if negative_prompt else self.config.default_negative_prompt
+        )
+        enhanced_prompt = prompt + self.config.camera_motion_prompts.get(
+            camera_motion, ""
+        )
+
+        # 强制使用动态目录，忽略底层原始逻辑
+        dyn_dir = get_dynamic_output_path()
+        output_path = dyn_dir / f"generation_{uuid.uuid4().hex[:8]}.mp4"
+
+        try:
+            self._text.prepare_text_encoding(enhanced_prompt, enhance_prompt=False)
+            # 调整为 32 的倍数（LTX 要求）
+            height = round(height / 32) * 32
+            width = round(width / 32) * 32
+
+            if audio_path:
+                # A2V pipeline 参数
+                gen_kwargs = {
+                    "prompt": enhanced_prompt,
+                    "negative_prompt": neg_prompt,
+                    "seed": seed,
+                    "height": height,
+                    "width": width,
+                    "num_frames": num_frames,
+                    "frame_rate": fps,
+                    "num_inference_steps": num_inference_steps,
+                    "images": images_inputs,
+                    "audio_path": audio_path,
+                    "audio_start_time": 0.0,
+                    "audio_max_duration": None,
+                    "output_path": str(output_path),
+                }
+            else:
+                # Fast pipeline 参数
+                gen_kwargs = {
+                    "prompt": enhanced_prompt,
+                    "seed": seed,
+                    "height": height,
+                    "width": width,
+                    "num_frames": num_frames,
+                    "frame_rate": fps,
+                    "images": images_inputs,
+                    "output_path": str(output_path),
+                }
+
+            pipeline_state.pipeline.generate(**gen_kwargs)
+
+            # 标记完成
+            self._generation.complete_generation(str(output_path))
+            return str(output_path)
+        finally:
+            self._text.clear_api_embeddings()
+            for p in temp_paths:
+                if os.path.exists(p):
+                    os.unlink(p)
+            self._start_frame_path = None
+            self._end_frame_path = None
+
+    VideoGenerationHandler.generate = patched_generate
+    VideoGenerationHandler.generate_video = patched_generate_video
+
+    # 2. 增强视频超分功能 (Monkey Patch LTXRetakePipeline)
+    _orig_ltx_retake_run = LTXRetakePipeline._run
+
+    def patched_ltx_retake_run(
+        self, video_path, prompt, start_time, end_time, seed, **kwargs
+    ):
+        # 拦截并修改目标宽高
+        target_w = getattr(self, "_target_width", None)
+        target_h = getattr(self, "_target_height", None)
+        target_strength = getattr(self, "_target_strength", 0.7)
+        is_upscale = target_w is not None and target_h is not None
+
+        import ltx_pipelines.utils.media_io as media_io
+        import services.retake_pipeline.ltx_retake_pipeline as lrp
+        import ltx_pipelines.utils.samplers as samplers
+        import ltx_pipelines.utils.helpers as helpers
+
+        _orig_get_meta = media_io.get_videostream_metadata
+        _orig_lrp_get_meta = getattr(lrp, "get_videostream_metadata", _orig_get_meta)
+        _orig_euler_loop = samplers.euler_denoising_loop
+        _orig_noise_video = helpers.noise_video_state
+
+        fps, num_frames, src_w, src_h = _orig_get_meta(video_path)
+
+        if is_upscale:
+            print(
+                f">>> 启动超分内核: {src_w}x{src_h} -> {target_w}x{target_h} (强度: {target_strength})"
+            )
+
+            # 1. 注入分辨率
+            def get_meta_patched(path):
+                return fps, num_frames, target_w, target_h
+
+            media_io.get_videostream_metadata = get_meta_patched
+            lrp.get_videostream_metadata = get_meta_patched
+
+            # 2. 注入起始噪声 (SDEdit 核心：加噪到指定强度)
+            def noise_video_patched(*args, **kwargs_inner):
+                kwargs_inner["noise_scale"] = target_strength
+                return _orig_noise_video(*args, **kwargs_inner)
+
+            helpers.noise_video_state = noise_video_patched
+
+            # 3. 注入采样起点 (从对应噪声位开始去噪)
+            def patched_euler_loop(
+                sigmas, video_state, audio_state, stepper, denoise_fn
+            ):
+                full_len = len(sigmas)
+                skip_idx = 0
+                for i, s in enumerate(sigmas):
+                    if s <= target_strength:
+                        skip_idx = i
+                        break
+                skip_idx = min(skip_idx, full_len - 2)
+                new_sigmas = sigmas[skip_idx:]
+                print(
+                    f">>> 采样拦截成功: 原步数 {full_len}, 现步数 {len(new_sigmas)}, 起始强度 {new_sigmas[0].item():.2f}"
+                )
+                return _orig_euler_loop(
+                    new_sigmas, video_state, audio_state, stepper, denoise_fn
+                )
+
+            samplers.euler_denoising_loop = patched_euler_loop
+
+            kwargs["regenerate_video"] = False
+            kwargs["regenerate_audio"] = False
+
+            try:
+                return _orig_ltx_retake_run(
+                    self, video_path, prompt, start_time, end_time, seed, **kwargs
+                )
+            finally:
+                media_io.get_videostream_metadata = _orig_get_meta
+                lrp.get_videostream_metadata = _orig_lrp_get_meta
+                samplers.euler_denoising_loop = _orig_euler_loop
+                helpers.noise_video_state = _orig_noise_video
+
+        return _orig_ltx_retake_run(
+            self, video_path, prompt, start_time, end_time, seed, **kwargs
+        )
+
+        return _orig_ltx_retake_run(
+            self, video_path, prompt, start_time, end_time, seed, **kwargs
+        )
+
+    LTXRetakePipeline._run = patched_ltx_retake_run
+
+    # --- 最终视频超分接口实现 ---
+    @app.post("/api/system/upscale-video")
+    async def route_upscale_video(request: Request):
+        try:
+            import uuid
+            import os
+            from datetime import datetime
+            from ltx_pipelines.utils.media_io import get_videostream_metadata
+            from ltx_core.types import SpatioTemporalScaleFactors
+
+            data = await request.json()
+            video_path = data.get("video_path")
+            target_res = data.get("resolution", "1080p")
+            prompt = data.get("prompt", "high quality, detailed, 4k")
+            strength = data.get("strength", 0.7)  # 获取前端传来的重绘幅度
+
+            if not video_path or not os.path.exists(video_path):
+                return JSONResponse(
+                    status_code=400, content={"error": "Invalid video path"}
+                )
+
+            # 计算目标宽高 (必须是 32 的倍数)
+            res_map = {"1080p": (1920, 1088), "720p": (1280, 704), "544p": (960, 544)}
+            target_w, target_h = res_map.get(target_res, (1920, 1088))
+
+            fps, num_frames, _, _ = get_videostream_metadata(video_path)
+
+            # 校验帧数 8k+1，如果不符则自动调整
+            scale = SpatioTemporalScaleFactors.default()
+            if (num_frames - 1) % scale.time != 0:
+                # 计算需要调整到的最近的有效帧数 (8k+1)
+                # 找到最接近的8k+1帧数
+                target_k = (num_frames - 1) // scale.time
+                # 选择最接近的k值：向下或向上取整
+                current_k = (num_frames - 1) // scale.time
+                current_remainder = (num_frames - 1) % scale.time
+
+                # 比较向上和向下取整哪个更接近
+                down_k = current_k
+                up_k = current_k + 1
+
+                # 向下取整的帧数
+                down_frames = down_k * scale.time + 1
+                # 向上取整的帧数
+                up_frames = up_k * scale.time + 1
+
+                # 选择差异最小的
+                if abs(num_frames - down_frames) <= abs(num_frames - up_frames):
+                    adjusted_frames = down_frames
+                else:
+                    adjusted_frames = up_frames
+
+                print(
+                    f">>> 帧数调整: {num_frames} -> {adjusted_frames} (符合 8k+1 规则)"
+                )
+
+                # 调整视频帧数 - 截断多余的帧或填充黑帧
+                adjusted_video_path = None
+                try:
+                    import cv2
+                    import numpy as np
+                    import tempfile
+
+                    # 使用cv2读取视频
+                    cap = cv2.VideoCapture(video_path)
+                    if not cap.isOpened():
+                        raise Exception("无法打开视频文件")
+
+                    frames = []
+                    while True:
+                        ret, frame = cap.read()
+                        if not ret:
+                            break
+                        frames.append(frame)
+                    cap.release()
+
+                    original_frame_count = len(frames)
+
+                    if adjusted_frames < original_frame_count:
+                        # 截断多余的帧
+                        frames = frames[:adjusted_frames]
+                        print(
+                            f">>> 已截断视频: {original_frame_count} -> {len(frames)} 帧"
+                        )
+                    else:
+                        # 填充黑帧 (复制最后一帧)
+                        last_frame = frames[-1] if frames else None
+                        if last_frame is not None:
+                            h, w = last_frame.shape[:2]
+                            black_frame = np.zeros((h, w, 3), dtype=np.uint8)
+                            while len(frames) < adjusted_frames:
+                                frames.append(black_frame.copy())
+                        print(
+                            f">>> 已填充视频: {original_frame_count} -> {len(frames)} 帧"
+                        )
+
+                    # 保存调整后的视频到临时文件
+                    adjusted_video_fd = tempfile.NamedTemporaryFile(
+                        suffix=".mp4", delete=False
+                    )
+                    adjusted_video_path = adjusted_video_fd.name
+                    adjusted_video_fd.close()
+
+                    fourcc = cv2.VideoWriter_fourcc(*"mp4v")
+                    out = cv2.VideoWriter(
+                        adjusted_video_path,
+                        fourcc,
+                        fps,
+                        (frames[0].shape[1], frames[0].shape[0]),
+                    )
+                    for frame in frames:
+                        out.write(frame)
+                    out.release()
+
+                    video_path = adjusted_video_path
+                    num_frames = adjusted_frames
+                    print(
+                        f">>> 视频帧数调整完成: {original_frame_count} -> {num_frames}"
+                    )
+
+                except ImportError:
+                    # cv2不可用，尝试使用LTX内置方法
+                    try:
+                        from ltx_pipelines.utils.media_io import (
+                            read_video_stream,
+                            write_video_stream,
+                        )
+                        import numpy as np
+
+                        frames, audio_data = read_video_stream(video_path, fps)
+                        original_frame_count = len(frames)
+
+                        if adjusted_frames < original_frame_count:
+                            frames = frames[:adjusted_frames]
+                        else:
+                            while len(frames) < adjusted_frames:
+                                frames = np.concatenate([frames, frames[-1:]], axis=0)
+
+                        import tempfile
+
+                        adjusted_video_fd = tempfile.NamedTemporaryFile(
+                            suffix=".mp4", delete=False
+                        )
+                        adjusted_video_path = adjusted_video_fd.name
+                        adjusted_video_fd.close()
+
+                        write_video_stream(adjusted_video_path, frames, fps)
+                        video_path = adjusted_video_path
+                        num_frames = adjusted_frames
+                        print(
+                            f">>> 视频帧数调整完成: {original_frame_count} -> {num_frames}"
+                        )
+
+                    except Exception as e2:
+                        print(f">>> 视频帧数自动调整失败: {e2}")
+                        return JSONResponse(
+                            status_code=400,
+                            content={
+                                "error": f"视频帧数({num_frames})不符合 8k+1 规则，且自动调整失败。请手动将视频帧数调整为 8k+1 格式（如 9, 17, 25, 33, 41, 49, 57, 65, 73, 81, 89, 97, 105 等）。"
+                            },
+                        )
+                except Exception as e:
+                    print(f">>> 视频帧数自动调整失败: {e}")
+                    return JSONResponse(
+                        status_code=400,
+                        content={
+                            "error": f"视频帧数({num_frames})不符合 8k+1 规则，且自动调整失败。请手动将视频帧数调整为 8k+1 格式（如 9, 17, 25, 33, 41, 49, 57, 65, 73, 81, 89, 97, 105 等）。"
+                        },
+                    )
+
+            # 1. 加载模型
+            pipeline_state = handler.pipelines.load_retake_pipeline(distilled=True)
+
+            # 3. 启动任务
+            generation_id = uuid.uuid4().hex[:8]
+            handler.generation.start_generation(generation_id)
+
+            # 核心修正：确保文件保存在动态的输出目录
+            save_dir = get_dynamic_output_path()
+            filename = f"upscale_{datetime.now().strftime('%Y%m%d_%H%M%S')}_{generation_id}.mp4"
+            full_output_path = save_dir / filename
+
+            # 3. 执行真正的超分逻辑
+            try:
+                # 注入目标分辨率和重绘幅度
+                pipeline_state.pipeline._target_width = target_w
+                pipeline_state.pipeline._target_height = target_h
+                pipeline_state.pipeline._target_strength = strength
+
+                def do_generate():
+                    pipeline_state.pipeline.generate(
+                        video_path=str(video_path),
+                        prompt=prompt,
+                        start_time=0.0,
+                        end_time=float(num_frames / fps),
+                        seed=int(time.time()) % 2147483647,
+                        output_path=str(full_output_path),
+                        distilled=True,
+                        regenerate_video=True,
+                        regenerate_audio=False,
+                    )
+
+                # 重要修复：放到线程池运行，避免阻塞主循环导致前端拿不到显存数据
+                from starlette.concurrency import run_in_threadpool
+
+                await run_in_threadpool(do_generate)
+
+                handler.generation.complete_generation(str(full_output_path))
+                return {"status": "complete", "video_path": filename}
+            except Exception as e:
+                # OOM 异常逃逸修复：强制返回友好的异常信息
+                try:
+                    handler.generation.cancel_generation()
+                except Exception:
+                    pass
+                if hasattr(handler.generation, "_generation_id"):
+                    handler.generation._generation_id = None
+                if hasattr(handler.generation, "_is_generating"):
+                    handler.generation._is_generating = False
+
+                error_msg = str(e)
+                if "CUDA out of memory" in error_msg:
+                    error_msg = "🚨 显存不足 (OOM)：视频时长过长或目标分辨率超出了当前显卡的承载极限，请降低目标分辨率重试！"
+                raise RuntimeError(error_msg) from e
+            finally:
+                if hasattr(pipeline_state.pipeline, "_target_width"):
+                    del pipeline_state.pipeline._target_width
+                if hasattr(pipeline_state.pipeline, "_target_height"):
+                    del pipeline_state.pipeline._target_height
+                if hasattr(pipeline_state.pipeline, "_target_strength"):
+                    del pipeline_state.pipeline._target_strength
+                import gc
+
+                gc.collect()
+                if (
+                    getattr(torch, "cuda", None) is not None
+                    and torch.cuda.is_available()
+                ):
+                    torch.cuda.empty_cache()
+
+        except Exception as e:
+            import traceback
+
+            traceback.print_exc()
+            return JSONResponse(status_code=500, content={"error": str(e)})
+
+    # ------------------
+
+    @app.post("/api/system/upload-image")
+    async def route_upload_image(request: Request):
+        try:
+            import uuid
+            import base64
+
+            # 接收 JSON 而不是 Multipart，绕过 python-multipart 缺失问题
+            data = await request.json()
+            b64_data = data.get("image")
+            filename = data.get("filename", "image.png")
+
+            if not b64_data:
+                return JSONResponse(
+                    status_code=400, content={"error": "No image data provided"}
+                )
+
+            # 处理 base64 头部 (例如 data:image/png;base64,...)
+            if "," in b64_data:
+                b64_data = b64_data.split(",")[1]
+
+            image_bytes = base64.b64decode(b64_data)
+
+            # 确保上传目录存在
+            upload_dir = get_dynamic_output_path() / "uploads"
+            upload_dir.mkdir(parents=True, exist_ok=True)
+
+            safe_filename = "".join([c for c in filename if c.isalnum() or c in "._-"])
+            file_path = upload_dir / f"up_{uuid.uuid4().hex[:6]}_{safe_filename}"
+
+            with file_path.open("wb") as buffer:
+                buffer.write(image_bytes)
+
+            return {"status": "success", "path": str(file_path)}
+        except Exception as e:
+            import traceback
+
+            error_msg = f"{type(e).__name__}: {str(e)}\n{traceback.format_exc()}"
+            print(f"Upload error: {error_msg}")
+            return JSONResponse(
+                status_code=500, content={"error": str(e), "detail": error_msg}
+            )
+
+    # ------------------
+
+    @app.get("/api/system/history")
+    async def route_get_history(request: Request):
+        try:
+            import os
+
+            page = int(request.query_params.get("page", 1))
+            limit = int(request.query_params.get("limit", 20))
+
+            history = []
+            dyn_path = get_dynamic_output_path()
+            if dyn_path.exists():
+                for filename in os.listdir(dyn_path):
+                    if filename == "uploads":
+                        continue
+                    full_path = dyn_path / filename
+                    if full_path.is_file() and filename.lower().endswith(
+                        (".mp4", ".png", ".jpg", ".webp")
+                    ):
+                        mtime = os.path.getmtime(full_path)
+                        history.append(
+                            {
+                                "filename": filename,
+                                "type": "video"
+                                if filename.lower().endswith(".mp4")
+                                else "image",
+                                "mtime": mtime,
+                                "fullpath": str(full_path),
+                            }
+                        )
+            history.sort(key=lambda x: x["mtime"], reverse=True)
+
+            total_items = len(history)
+            total_pages = (total_items + limit - 1) // limit
+            start_idx = (page - 1) * limit
+            end_idx = start_idx + limit
+
+            return {
+                "status": "success",
+                "history": history[start_idx:end_idx],
+                "total_pages": total_pages,
+                "current_page": page,
+                "total_items": total_items,
+            }
+        except Exception as e:
+            return JSONResponse(status_code=500, content={"error": str(e)})
+
+    @app.post("/api/system/delete-file")
+    async def route_delete_file(request: Request):
+        try:
+            import os
+
+            data = await request.json()
+            filename = data.get("filename", "")
+
+            if not filename:
+                return JSONResponse(
+                    status_code=400, content={"error": "Filename is required"}
+                )
+
+            dyn_path = get_dynamic_output_path()
+            file_path = dyn_path / filename
+
+            if file_path.exists() and file_path.is_file():
+                file_path.unlink()
+                return {"status": "success", "message": "File deleted"}
+            else:
+                return JSONResponse(
+                    status_code=404, content={"error": "File not found"}
+                )
+        except Exception as e:
+            return JSONResponse(status_code=500, content={"error": str(e)})
+
+    # 路由注册
+    app.include_router(health_router)
+    app.include_router(generation_router)
+    app.include_router(models_router)
+    app.include_router(settings_router)
+    app.include_router(image_gen_router)
+    app.include_router(suggest_gap_prompt_router)
+    app.include_router(retake_router)
+    app.include_router(ic_lora_router)
+    app.include_router(runtime_policy_router)
+
+    # --- [安全补丁] 状态栏显示修复 ---
+
+    # --- 最终状态栏修复补丁: 只要服务运行且 GPU 没死，就视为就绪 ---
+    from handlers.health_handler import HealthHandler
+
+    if not hasattr(HealthHandler, "_fixed_v2"):
+        _orig_get_health = HealthHandler.get_health
+
+        def patched_health_v2(self):
+            resp = _orig_get_health(self)
+            # 解析：如果后端逻辑还在判断模型未加载，我们检查一下核心状态
+            # 如果系统没有崩溃，我们就强制标记为已加载，让前端允许交互
+            if not resp.models_loaded:
+                # 我们认为只要 API 能通，底层状态服务(state)只要存在，就视为由于异步加载引起的暂时性 False
+                # 直接返回 True，前端会显示"待机就绪"
+                resp.models_loaded = True
+            return resp
+
+        HealthHandler.get_health = patched_health_v2
+        HealthHandler._fixed_v2 = True
+    # ------------------------------------------------------------
+
+    # --- 修复显存采集指针：使得显存监控永远对准当前选定工作的 GPU ---
+    from services.gpu_info.gpu_info_impl import GpuInfoImpl
+
+    if not hasattr(GpuInfoImpl, "_fixed_vram_patch"):
+        _orig_get_gpu_info = GpuInfoImpl.get_gpu_info
+
+        def patched_get_gpu_info(self):
+            import torch
+
+            if self.get_cuda_available():
+                idx = 0
+                if (
+                    hasattr(handler.config.device, "index")
+                    and handler.config.device.index is not None
+                ):
+                    idx = handler.config.device.index
+                try:
+                    import pynvml
+
+                    pynvml.nvmlInit()
+                    handle = pynvml.nvmlDeviceGetHandleByIndex(idx)
+                    raw_name = pynvml.nvmlDeviceGetName(handle)
+                    name = (
+                        raw_name.decode("utf-8", errors="replace")
+                        if isinstance(raw_name, bytes)
+                        else str(raw_name)
+                    )
+                    memory = pynvml.nvmlDeviceGetMemoryInfo(handle)
+                    pynvml.nvmlShutdown()
+                    return {
+                        "name": f"{name} [ID: {idx}]",
+                        "vram": memory.total // (1024 * 1024),
+                        "vramUsed": memory.used // (1024 * 1024),
+                    }
+                except Exception:
+                    pass
+            return _orig_get_gpu_info(self)
+
+        GpuInfoImpl.get_gpu_info = patched_get_gpu_info
+        GpuInfoImpl._fixed_vram_patch = True
+
+    return app

LTX2.3/patches/handlers/video_generation_handler.py

@@ -0,0 +1,673 @@
+"""Video generation orchestration handler."""
+
+from __future__ import annotations
+
+import logging
+import os
+import tempfile
+import time
+import uuid
+from datetime import datetime
+from pathlib import Path
+from threading import RLock
+from typing import TYPE_CHECKING
+
+from PIL import Image
+
+from api_types import (
+    GenerateVideoRequest,
+    GenerateVideoResponse,
+    ImageConditioningInput,
+    VideoCameraMotion,
+)
+from _routes._errors import HTTPError
+from handlers.base import StateHandlerBase
+from handlers.generation_handler import GenerationHandler
+from handlers.pipelines_handler import PipelinesHandler
+from handlers.text_handler import TextHandler
+from runtime_config.model_download_specs import resolve_model_path
+from server_utils.media_validation import (
+    normalize_optional_path,
+    validate_audio_file,
+    validate_image_file,
+)
+from services.interfaces import LTXAPIClient
+from state.app_state_types import AppState
+from state.app_settings import should_video_generate_with_ltx_api
+
+if TYPE_CHECKING:
+    from runtime_config.runtime_config import RuntimeConfig
+
+logger = logging.getLogger(__name__)
+
+FORCED_API_MODEL_MAP: dict[str, str] = {
+    "fast": "ltx-2-3-fast",
+    "pro": "ltx-2-3-pro",
+}
+FORCED_API_RESOLUTION_MAP: dict[str, dict[str, str]] = {
+    "1080p": {"16:9": "1920x1080", "9:16": "1080x1920"},
+    "1440p": {"16:9": "2560x1440", "9:16": "1440x2560"},
+    "2160p": {"16:9": "3840x2160", "9:16": "2160x3840"},
+}
+A2V_FORCED_API_RESOLUTION = "1920x1080"
+FORCED_API_ALLOWED_ASPECT_RATIOS = {"16:9", "9:16"}
+FORCED_API_ALLOWED_FPS = {24, 25, 48, 50}
+
+
+def _get_allowed_durations(model_id: str, resolution_label: str, fps: int) -> set[int]:
+    if model_id == "ltx-2-3-fast" and resolution_label == "1080p" and fps in {24, 25}:
+        return {6, 8, 10, 12, 14, 16, 18, 20}
+    return {6, 8, 10}
+
+
+class VideoGenerationHandler(StateHandlerBase):
+    def __init__(
+        self,
+        state: AppState,
+        lock: RLock,
+        generation_handler: GenerationHandler,
+        pipelines_handler: PipelinesHandler,
+        text_handler: TextHandler,
+        ltx_api_client: LTXAPIClient,
+        config: RuntimeConfig,
+    ) -> None:
+        super().__init__(state, lock, config)
+        self._generation = generation_handler
+        self._pipelines = pipelines_handler
+        self._text = text_handler
+        self._ltx_api_client = ltx_api_client
+
+    def generate(self, req: GenerateVideoRequest) -> GenerateVideoResponse:
+        if should_video_generate_with_ltx_api(
+            force_api_generations=self.config.force_api_generations,
+            settings=self.state.app_settings,
+        ):
+            return self._generate_forced_api(req)
+
+        if self._generation.is_generation_running():
+            raise HTTPError(409, "Generation already in progress")
+
+        resolution = req.resolution
+
+        duration = int(float(req.duration))
+        fps = int(float(req.fps))
+
+        audio_path = normalize_optional_path(req.audioPath)
+        if audio_path:
+            return self._generate_a2v(req, duration, fps, audio_path=audio_path)
+
+        logger.info("Resolution %s - using fast pipeline", resolution)
+
+        RESOLUTION_MAP_16_9: dict[str, tuple[int, int]] = {
+            "540p": (960, 540),
+            "720p": (1280, 720),
+            "1080p": (1920, 1080),
+        }
+
+        def get_16_9_size(res: str) -> tuple[int, int]:
+            return RESOLUTION_MAP_16_9.get(res, (1280, 720))
+
+        def get_9_16_size(res: str) -> tuple[int, int]:
+            w, h = get_16_9_size(res)
+            return h, w
+
+        match req.aspectRatio:
+            case "9:16":
+                width, height = get_9_16_size(resolution)
+            case "16:9":
+                width, height = get_16_9_size(resolution)
+
+        num_frames = self._compute_num_frames(duration, fps)
+
+        image = None
+        image_path = normalize_optional_path(req.imagePath)
+        if image_path:
+            image = self._prepare_image(image_path, width, height)
+            logger.info("Image: %s -> %sx%s", image_path, width, height)
+
+        generation_id = self._make_generation_id()
+        seed = self._resolve_seed()
+
+        try:
+            self._pipelines.load_gpu_pipeline("fast", should_warm=False)
+            self._generation.start_generation(generation_id)
+
+            output_path = self.generate_video(
+                prompt=req.prompt,
+                image=image,
+                height=height,
+                width=width,
+                num_frames=num_frames,
+                fps=fps,
+                seed=seed,
+                camera_motion=req.cameraMotion,
+                negative_prompt=req.negativePrompt,
+            )
+
+            self._generation.complete_generation(output_path)
+            return GenerateVideoResponse(status="complete", video_path=output_path)
+
+        except Exception as e:
+            self._generation.fail_generation(str(e))
+            if "cancelled" in str(e).lower():
+                logger.info("Generation cancelled by user")
+                return GenerateVideoResponse(status="cancelled")
+
+            raise HTTPError(500, str(e)) from e
+
+    def generate_video(
+        self,
+        prompt: str,
+        image: Image.Image | None,
+        height: int,
+        width: int,
+        num_frames: int,
+        fps: float,
+        seed: int,
+        camera_motion: VideoCameraMotion,
+        negative_prompt: str,
+    ) -> str:
+        t_total_start = time.perf_counter()
+        gen_mode = "i2v" if image is not None else "t2v"
+        logger.info(
+            "[%s] Generation started (model=fast, %dx%d, %d frames, %d fps)",
+            gen_mode,
+            width,
+            height,
+            num_frames,
+            int(fps),
+        )
+
+        if self._generation.is_generation_cancelled():
+            raise RuntimeError("Generation was cancelled")
+
+        if not resolve_model_path(
+            self.models_dir, self.config.model_download_specs, "checkpoint"
+        ).exists():
+            raise RuntimeError(
+                "Models not downloaded. Please download the AI models first using the Model Status menu."
+            )
+
+        total_steps = 8
+
+        self._generation.update_progress("loading_model", 5, 0, total_steps)
+        t_load_start = time.perf_counter()
+        pipeline_state = self._pipelines.load_gpu_pipeline("fast", should_warm=False)
+        t_load_end = time.perf_counter()
+        logger.info("[%s] Pipeline load: %.2fs", gen_mode, t_load_end - t_load_start)
+
+        self._generation.update_progress("encoding_text", 10, 0, total_steps)
+
+        enhanced_prompt = prompt + self.config.camera_motion_prompts.get(
+            camera_motion, ""
+        )
+
+        images: list[ImageConditioningInput] = []
+        temp_image_path: str | None = None
+        if image is not None:
+            temp_image_path = tempfile.NamedTemporaryFile(
+                suffix=".png", delete=False
+            ).name
+            image.save(temp_image_path)
+            images = [
+                ImageConditioningInput(path=temp_image_path, frame_idx=0, strength=1.0)
+            ]
+
+        output_path = self._make_output_path()
+
+        try:
+            settings = self.state.app_settings
+            use_api_encoding = not self._text.should_use_local_encoding()
+            if image is not None:
+                enhance = use_api_encoding and settings.prompt_enhancer_enabled_i2v
+            else:
+                enhance = use_api_encoding and settings.prompt_enhancer_enabled_t2v
+
+            encoding_method = "api" if use_api_encoding else "local"
+            t_text_start = time.perf_counter()
+            self._text.prepare_text_encoding(enhanced_prompt, enhance_prompt=enhance)
+            t_text_end = time.perf_counter()
+            logger.info(
+                "[%s] Text encoding (%s): %.2fs",
+                gen_mode,
+                encoding_method,
+                t_text_end - t_text_start,
+            )
+
+            self._generation.update_progress("inference", 15, 0, total_steps)
+
+            height = round(height / 64) * 64
+            width = round(width / 64) * 64
+
+            t_inference_start = time.perf_counter()
+            pipeline_state.pipeline.generate(
+                prompt=enhanced_prompt,
+                seed=seed,
+                height=height,
+                width=width,
+                num_frames=num_frames,
+                frame_rate=fps,
+                images=images,
+                output_path=str(output_path),
+            )
+            t_inference_end = time.perf_counter()
+            logger.info(
+                "[%s] Inference: %.2fs", gen_mode, t_inference_end - t_inference_start
+            )
+
+            if self._generation.is_generation_cancelled():
+                if output_path.exists():
+                    output_path.unlink()
+                raise RuntimeError("Generation was cancelled")
+
+            t_total_end = time.perf_counter()
+            logger.info(
+                "[%s] Total generation: %.2fs (load=%.2fs, text=%.2fs, inference=%.2fs)",
+                gen_mode,
+                t_total_end - t_total_start,
+                t_load_end - t_load_start,
+                t_text_end - t_text_start,
+                t_inference_end - t_inference_start,
+            )
+
+            self._generation.update_progress("complete", 100, total_steps, total_steps)
+            return str(output_path)
+        finally:
+            self._text.clear_api_embeddings()
+            if temp_image_path and os.path.exists(temp_image_path):
+                os.unlink(temp_image_path)
+
+    def _generate_a2v(
+        self, req: GenerateVideoRequest, duration: int, fps: int, *, audio_path: str
+    ) -> GenerateVideoResponse:
+        if req.model != "pro":
+            logger.warning(
+                "A2V local requested with model=%s; A2V always uses pro pipeline",
+                req.model,
+            )
+        validated_audio_path = validate_audio_file(audio_path)
+        audio_path_str = str(validated_audio_path)
+
+        # 支持竖屏和横屏
+        RESOLUTION_MAP: dict[str, tuple[int, int]] = {
+            "540p": (960, 540),
+            "720p": (1280, 720),
+            "1080p": (1920, 1080),
+        }
+
+        base_w, base_h = RESOLUTION_MAP.get(req.resolution, (1280, 720))
+
+        # 根据 aspectRatio 调整分辨率
+        if req.aspectRatio == "9:16":
+            width, height = base_h, base_w  # 竖屏
+        else:
+            width, height = base_w, base_h  # 横屏
+
+        num_frames = self._compute_num_frames(duration, fps)
+
+        image = None
+        temp_image_path: str | None = None
+        image_path = normalize_optional_path(req.imagePath)
+        if image_path:
+            image = self._prepare_image(image_path, width, height)
+
+        # 获取首尾帧
+        start_frame_path = normalize_optional_path(getattr(req, "startFramePath", None))
+        end_frame_path = normalize_optional_path(getattr(req, "endFramePath", None))
+
+        seed = self._resolve_seed()
+
+        generation_id = self._make_generation_id()
+
+        temp_image_paths: list[str] = []
+        try:
+            a2v_state = self._pipelines.load_a2v_pipeline()
+            self._generation.start_generation(generation_id)
+
+            enhanced_prompt = req.prompt + self.config.camera_motion_prompts.get(
+                req.cameraMotion, ""
+            )
+            neg = (
+                req.negativePrompt
+                if req.negativePrompt
+                else self.config.default_negative_prompt
+            )
+
+            images: list[ImageConditioningInput] = []
+            temp_image_paths: list[str] = []
+
+            # 首帧
+            if start_frame_path:
+                start_img = self._prepare_image(start_frame_path, width, height)
+                temp_start_path = tempfile.NamedTemporaryFile(
+                    suffix=".png", delete=False
+                ).name
+                start_img.save(temp_start_path)
+                temp_image_paths.append(temp_start_path)
+                images.append(
+                    ImageConditioningInput(
+                        path=temp_start_path, frame_idx=0, strength=1.0
+                    )
+                )
+
+            # 中间图片（如果有）
+            if image is not None and not start_frame_path:
+                temp_image_path = tempfile.NamedTemporaryFile(
+                    suffix=".png", delete=False
+                ).name
+                image.save(temp_image_path)
+                temp_image_paths.append(temp_image_path)
+                images.append(
+                    ImageConditioningInput(
+                        path=temp_image_path, frame_idx=0, strength=1.0
+                    )
+                )
+
+            # 尾帧
+            if end_frame_path:
+                last_latent_idx = (num_frames - 1) // 8 + 1 - 1
+                end_img = self._prepare_image(end_frame_path, width, height)
+                temp_end_path = tempfile.NamedTemporaryFile(
+                    suffix=".png", delete=False
+                ).name
+                end_img.save(temp_end_path)
+                temp_image_paths.append(temp_end_path)
+                images.append(
+                    ImageConditioningInput(
+                        path=temp_end_path, frame_idx=last_latent_idx, strength=1.0
+                    )
+                )
+
+            output_path = self._make_output_path()
+
+            total_steps = 11  # distilled: 8 steps (stage 1) + 3 steps (stage 2)
+
+            a2v_settings = self.state.app_settings
+            a2v_use_api = not self._text.should_use_local_encoding()
+            if image is not None:
+                a2v_enhance = a2v_use_api and a2v_settings.prompt_enhancer_enabled_i2v
+            else:
+                a2v_enhance = a2v_use_api and a2v_settings.prompt_enhancer_enabled_t2v
+
+            self._generation.update_progress("loading_model", 5, 0, total_steps)
+            self._generation.update_progress("encoding_text", 10, 0, total_steps)
+            self._text.prepare_text_encoding(
+                enhanced_prompt, enhance_prompt=a2v_enhance
+            )
+            self._generation.update_progress("inference", 15, 0, total_steps)
+
+            a2v_state.pipeline.generate(
+                prompt=enhanced_prompt,
+                negative_prompt=neg,
+                seed=seed,
+                height=height,
+                width=width,
+                num_frames=num_frames,
+                frame_rate=fps,
+                num_inference_steps=total_steps,
+                images=images,
+                audio_path=audio_path_str,
+                audio_start_time=0.0,
+                audio_max_duration=None,
+                output_path=str(output_path),
+            )
+
+            if self._generation.is_generation_cancelled():
+                if output_path.exists():
+                    output_path.unlink()
+                raise RuntimeError("Generation was cancelled")
+
+            self._generation.update_progress("complete", 100, total_steps, total_steps)
+            self._generation.complete_generation(str(output_path))
+            return GenerateVideoResponse(status="complete", video_path=str(output_path))
+
+        except Exception as e:
+            self._generation.fail_generation(str(e))
+            if "cancelled" in str(e).lower():
+                logger.info("Generation cancelled by user")
+                return GenerateVideoResponse(status="cancelled")
+            raise HTTPError(500, str(e)) from e
+        finally:
+            self._text.clear_api_embeddings()
+            # 清理所有临时图片
+            for tmp_path in temp_image_paths:
+                if tmp_path and os.path.exists(tmp_path):
+                    try:
+                        os.unlink(tmp_path)
+                    except Exception:
+                        pass
+            if temp_image_path and os.path.exists(temp_image_path):
+                try:
+                    os.unlink(temp_image_path)
+                except Exception:
+                    pass
+
+    def _prepare_image(self, image_path: str, width: int, height: int) -> Image.Image:
+        validated_path = validate_image_file(image_path)
+        try:
+            img = Image.open(validated_path).convert("RGB")
+        except Exception:
+            raise HTTPError(400, f"Invalid image file: {image_path}") from None
+        img_w, img_h = img.size
+        target_ratio = width / height
+        img_ratio = img_w / img_h
+        if img_ratio > target_ratio:
+            new_h = height
+            new_w = int(img_w * (height / img_h))
+        else:
+            new_w = width
+            new_h = int(img_h * (width / img_w))
+        resized = img.resize((new_w, new_h), Image.Resampling.LANCZOS)
+        left = (new_w - width) // 2
+        top = (new_h - height) // 2
+        return resized.crop((left, top, left + width, top + height))
+
+    @staticmethod
+    def _make_generation_id() -> str:
+        return uuid.uuid4().hex[:8]
+
+    @staticmethod
+    def _compute_num_frames(duration: int, fps: int) -> int:
+        n = ((duration * fps) // 8) * 8 + 1
+        return max(n, 9)
+
+    def _resolve_seed(self) -> int:
+        settings = self.state.app_settings
+        if settings.seed_locked:
+            logger.info("Using locked seed: %s", settings.locked_seed)
+            return settings.locked_seed
+        return int(time.time()) % 2147483647
+
+    def _make_output_path(self) -> Path:
+        timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+        return (
+            self.config.outputs_dir
+            / f"ltx2_video_{timestamp}_{self._make_generation_id()}.mp4"
+        )
+
+    def _generate_forced_api(self, req: GenerateVideoRequest) -> GenerateVideoResponse:
+        if self._generation.is_generation_running():
+            raise HTTPError(409, "Generation already in progress")
+
+        generation_id = self._make_generation_id()
+        self._generation.start_api_generation(generation_id)
+
+        audio_path = normalize_optional_path(req.audioPath)
+        image_path = normalize_optional_path(req.imagePath)
+        has_input_audio = bool(audio_path)
+        has_input_image = bool(image_path)
+
+        try:
+            self._generation.update_progress("validating_request", 5, None, None)
+
+            api_key = self.state.app_settings.ltx_api_key.strip()
+            logger.info(
+                "Forced API generation route selected (key_present=%s)", bool(api_key)
+            )
+            if not api_key:
+                raise HTTPError(400, "PRO_API_KEY_REQUIRED")
+
+            requested_model = req.model.strip().lower()
+            api_model_id = FORCED_API_MODEL_MAP.get(requested_model)
+            if api_model_id is None:
+                raise HTTPError(400, "INVALID_FORCED_API_MODEL")
+
+            resolution_label = req.resolution
+            resolution_by_aspect = FORCED_API_RESOLUTION_MAP.get(resolution_label)
+            if resolution_by_aspect is None:
+                raise HTTPError(400, "INVALID_FORCED_API_RESOLUTION")
+
+            aspect_ratio = req.aspectRatio.strip()
+            if aspect_ratio not in FORCED_API_ALLOWED_ASPECT_RATIOS:
+                raise HTTPError(400, "INVALID_FORCED_API_ASPECT_RATIO")
+
+            api_resolution = resolution_by_aspect[aspect_ratio]
+
+            prompt = req.prompt
+
+            if self._generation.is_generation_cancelled():
+                raise RuntimeError("Generation was cancelled")
+
+            if has_input_audio:
+                if requested_model != "pro":
+                    logger.warning(
+                        "A2V requested with model=%s; overriding to 'pro'",
+                        requested_model,
+                    )
+                api_model_id = FORCED_API_MODEL_MAP["pro"]
+                if api_resolution != A2V_FORCED_API_RESOLUTION:
+                    logger.warning(
+                        "A2V requested with resolution=%s; overriding to '%s'",
+                        api_resolution,
+                        A2V_FORCED_API_RESOLUTION,
+                    )
+                api_resolution = A2V_FORCED_API_RESOLUTION
+                validated_audio_path = validate_audio_file(audio_path)
+                validated_image_path: Path | None = None
+                if image_path is not None:
+                    validated_image_path = validate_image_file(image_path)
+
+                self._generation.update_progress("uploading_audio", 20, None, None)
+                audio_uri = self._ltx_api_client.upload_file(
+                    api_key=api_key,
+                    file_path=str(validated_audio_path),
+                )
+                image_uri: str | None = None
+                if validated_image_path is not None:
+                    self._generation.update_progress("uploading_image", 35, None, None)
+                    image_uri = self._ltx_api_client.upload_file(
+                        api_key=api_key,
+                        file_path=str(validated_image_path),
+                    )
+                self._generation.update_progress("inference", 55, None, None)
+                video_bytes = self._ltx_api_client.generate_audio_to_video(
+                    api_key=api_key,
+                    prompt=prompt,
+                    audio_uri=audio_uri,
+                    image_uri=image_uri,
+                    model=api_model_id,
+                    resolution=api_resolution,
+                )
+                self._generation.update_progress("downloading_output", 85, None, None)
+            elif has_input_image:
+                validated_image_path = validate_image_file(image_path)
+
+                duration = self._parse_forced_numeric_field(
+                    req.duration, "INVALID_FORCED_API_DURATION"
+                )
+                fps = self._parse_forced_numeric_field(
+                    req.fps, "INVALID_FORCED_API_FPS"
+                )
+                if fps not in FORCED_API_ALLOWED_FPS:
+                    raise HTTPError(400, "INVALID_FORCED_API_FPS")
+                if duration not in _get_allowed_durations(
+                    api_model_id, resolution_label, fps
+                ):
+                    raise HTTPError(400, "INVALID_FORCED_API_DURATION")
+
+                generate_audio = self._parse_audio_flag(req.audio)
+                self._generation.update_progress("uploading_image", 20, None, None)
+                image_uri = self._ltx_api_client.upload_file(
+                    api_key=api_key,
+                    file_path=str(validated_image_path),
+                )
+                self._generation.update_progress("inference", 55, None, None)
+                video_bytes = self._ltx_api_client.generate_image_to_video(
+                    api_key=api_key,
+                    prompt=prompt,
+                    image_uri=image_uri,
+                    model=api_model_id,
+                    resolution=api_resolution,
+                    duration=float(duration),
+                    fps=float(fps),
+                    generate_audio=generate_audio,
+                    camera_motion=req.cameraMotion,
+                )
+                self._generation.update_progress("downloading_output", 85, None, None)
+            else:
+                duration = self._parse_forced_numeric_field(
+                    req.duration, "INVALID_FORCED_API_DURATION"
+                )
+                fps = self._parse_forced_numeric_field(
+                    req.fps, "INVALID_FORCED_API_FPS"
+                )
+                if fps not in FORCED_API_ALLOWED_FPS:
+                    raise HTTPError(400, "INVALID_FORCED_API_FPS")
+                if duration not in _get_allowed_durations(
+                    api_model_id, resolution_label, fps
+                ):
+                    raise HTTPError(400, "INVALID_FORCED_API_DURATION")
+
+                generate_audio = self._parse_audio_flag(req.audio)
+                self._generation.update_progress("inference", 55, None, None)
+                video_bytes = self._ltx_api_client.generate_text_to_video(
+                    api_key=api_key,
+                    prompt=prompt,
+                    model=api_model_id,
+                    resolution=api_resolution,
+                    duration=float(duration),
+                    fps=float(fps),
+                    generate_audio=generate_audio,
+                    camera_motion=req.cameraMotion,
+                )
+                self._generation.update_progress("downloading_output", 85, None, None)
+
+            if self._generation.is_generation_cancelled():
+                raise RuntimeError("Generation was cancelled")
+
+            output_path = self._write_forced_api_video(video_bytes)
+            if self._generation.is_generation_cancelled():
+                output_path.unlink(missing_ok=True)
+                raise RuntimeError("Generation was cancelled")
+
+            self._generation.update_progress("complete", 100, None, None)
+            self._generation.complete_generation(str(output_path))
+            return GenerateVideoResponse(status="complete", video_path=str(output_path))
+        except HTTPError as e:
+            self._generation.fail_generation(e.detail)
+            raise
+        except Exception as e:
+            self._generation.fail_generation(str(e))
+            if "cancelled" in str(e).lower():
+                logger.info("Generation cancelled by user")
+                return GenerateVideoResponse(status="cancelled")
+            raise HTTPError(500, str(e)) from e
+
+    def _write_forced_api_video(self, video_bytes: bytes) -> Path:
+        output_path = self._make_output_path()
+        output_path.write_bytes(video_bytes)
+        return output_path
+
+    @staticmethod
+    def _parse_forced_numeric_field(raw_value: str, error_detail: str) -> int:
+        try:
+            return int(float(raw_value))
+        except (TypeError, ValueError):
+            raise HTTPError(400, error_detail) from None
+
+    @staticmethod
+    def _parse_audio_flag(audio_value: str | bool) -> bool:
+        if isinstance(audio_value, bool):
+            return audio_value
+        normalized = audio_value.strip().lower()
+        return normalized in {"1", "true", "yes", "on"}

LTX2.3/patches/launcher.py

@@ -0,0 +1,20 @@
+
+import sys
+import os
+
+patch_dir = r"C:\Users\1-xuanran\Desktop\LTX多合一启动器\patches"
+backend_dir = r"C:\Program Files\LTX Desktop\resources\backend"
+
+# 防御性清除：强行剥离所有的默认 backend_dir 引用
+sys.path = [p for p in sys.path if p and os.path.normpath(p) != os.path.normpath(backend_dir)]
+sys.path = [p for p in sys.path if p and p != "." and p != ""]
+
+# 绝对插队注入：优先搜索 PATCHES_DIR
+sys.path.insert(0, patch_dir)
+sys.path.insert(1, backend_dir)
+
+import uvicorn
+from ltx2_server import app
+
+if __name__ == '__main__':
+    uvicorn.run(app, host="0.0.0.0", port=3000, log_level="info", access_log=False)

LTX2.3/run.bat

@@ -0,0 +1,38 @@
+@echo off
+title LTX-2 Cinematic Workstation
+
+echo =========================================================
+echo    LTX-2 Cinematic UI Booting...
+echo =========================================================
+echo.
+
+set "LTX_PY=%USERPROFILE%\AppData\Local\LTXDesktop\python\python.exe"
+set "LTX_UI_URL=http://127.0.0.1:4000/"
+
+if exist "%LTX_PY%" (
+    echo [SUCCESS] LTX Bundled Python environment detected!
+    echo [INFO] Browser will open automatically when UI is ready...
+    start "" powershell -NoProfile -WindowStyle Hidden -Command "$ProgressPreference='SilentlyContinue'; $deadline=(Get-Date).AddSeconds(60); while((Get-Date) -lt $deadline){ try { Invoke-WebRequest -UseBasicParsing '%LTX_UI_URL%' -TimeoutSec 2 | Out-Null; Start-Process '%LTX_UI_URL%'; exit 0 } catch { Start-Sleep -Seconds 1 } }"
+    echo [INFO] Starting workspace natively...
+    echo ---------------------------------------------------------
+    "%LTX_PY%" main.py
+    pause
+    exit /b
+)
+
+python --version >nul 2>&1
+if %errorlevel% equ 0 (
+    echo [WARNING] LTX Bundled Python not found.
+    echo [INFO] Browser will open automatically when UI is ready...
+    start "" powershell -NoProfile -WindowStyle Hidden -Command "$ProgressPreference='SilentlyContinue'; $deadline=(Get-Date).AddSeconds(60); while((Get-Date) -lt $deadline){ try { Invoke-WebRequest -UseBasicParsing '%LTX_UI_URL%' -TimeoutSec 2 | Out-Null; Start-Process '%LTX_UI_URL%'; exit 0 } catch { Start-Sleep -Seconds 1 } }"
+    echo [INFO] Falling back to global Python environment...
+    echo ---------------------------------------------------------
+    python main.py
+    pause
+    exit /b
+)
+
+echo [ERROR] FATAL: No Python interpreter found on this system.
+echo [INFO] Please run install.bat to download and set up Python!
+echo.
+pause