fix gradio (#871)

Author: gushiqiao

app/gradio_demo.py

@@ -22,7 +22,7 @@ model_path=/models/
 
 # Server configuration
 server_name="0.0.0.0"
-server_port=8033
+server_port=8036
 
 # Output directory configuration
 output_dir="./outputs"
@@ -115,14 +115,9 @@ if [[ ! -d "$model_path" ]]; then
     exit 1
 fi
 
-# Select demo file based on language
-if [[ "$lang" == "zh" ]]; then
-    demo_file="gradio_demo_zh.py"
-    echo "🌏 Using Chinese interface"
-else
-    demo_file="gradio_demo.py"
-    echo "🌏 Using English interface"
-fi
+# 使用新的统一入口文件
+demo_file="gradio_demo.py"
+echo "🌏 Using unified interface (language: $lang)"
 
 # Check if demo file exists
 if [[ ! -f "$demo_file" ]]; then
@@ -168,11 +163,20 @@ echo "🔄 First startup may take several minutes to load resources..."
 echo "=========================================="
 
 # Start Python demo
+if [[ "$demo_file" == "gradio_demo.py" ]]; then
+    python $demo_file \
+        --model_path "$model_path" \
+        --server_name "$server_name" \
+        --server_port "$server_port" \
+        --output_dir "$output_dir" \
+        --lang "$lang"
+else
 python $demo_file \
     --model_path "$model_path" \
     --server_name "$server_name" \
     --server_port "$server_port" \
     --output_dir "$output_dir"
+fi
 
 # Display final system resource usage
 echo ""

app/gradio_demo_zh.py

@@ -0,0 +1,199 @@
+"""国际化支持模块"""
+
+import os
+
+# 默认语言
+DEFAULT_LANG = os.getenv("GRADIO_LANG", "zh")
+
+# 翻译字典
+TRANSLATIONS = {
+    "zh": {
+        "title": "🎬 LightX2V 图片/视频生成器",
+        "model_config": "🗂️ 模型配置",
+        "model_config_hint": "💡 **提示**：请确保以下每个模型选项至少有一个已下载✅的模型可用，否则可能无法正常生成视频。",
+        "fp8_not_supported": "⚠️ **您的设备不支持fp8推理**，已自动隐藏包含fp8的模型选项。",
+        "model_type": "模型类型",
+        "model_type_info": "Wan2.2 需要分别指定高噪模型和低噪模型; Qwen-Image-Edit-2511 用于图片编辑(i2i); Qwen-Image-2512 用于文本生成图片(t2i); Z-Image-Turbo 用于文本生成图片(t2i)",
+        "qwen3_encoder": "📝 Qwen3 编码器",
+        "scheduler": "⏱️ 调度器",
+        "qwen25vl_encoder": "📝 Qwen25-VL 编码器",
+        "task_type": "任务类型",
+        "task_type_info": "I2V: 图生视频, T2V: 文生视频, T2I: 文生图, I2I: 图片编辑",
+        "download_source": "📥 下载源",
+        "download_source_info": "选择模型下载源",
+        "diffusion_model": "🎨 Diffusion模型",
+        "high_noise_model": "🔊 高噪模型",
+        "low_noise_model": "🔇 低噪模型",
+        "text_encoder": "📝 文本编码器",
+        "text_encoder_tokenizer": "📝 文本编码器 Tokenizer",
+        "image_encoder": "🖼️ 图像编码器",
+        "image_encoder_tokenizer": "🖼️ 图像编码器 Tokenizer",
+        "vae": "🎞️ VAE编码/解码器",
+        "attention_operator": "⚡ 注意力算子",
+        "attention_operator_info": "使用适当的注意力算子加速推理",
+        "quant_operator": "⚡矩阵乘法算子",
+        "quant_operator_info": "选择低精度矩阵乘法算子以加速推理",
+        "input_params": "📥 输入参数",
+        "input_image": "输入图像（可拖入多张图片）",
+        "image_preview": "已上传的图片预览",
+        "image_path": "图片路径",
+        "prompt": "提示词",
+        "prompt_placeholder": "描述视频/图片内容...",
+        "negative_prompt": "负向提示词",
+        "negative_prompt_placeholder": "不希望出现在视频/图片中的内容...",
+        "max_resolution": "最大分辨率",
+        "max_resolution_info": "如果显存不足，可调低分辨率",
+        "random_seed": "随机种子",
+        "infer_steps": "推理步数",
+        "infer_steps_distill": "蒸馏模型推理步数默认为4。",
+        "infer_steps_info": "视频生成的推理步数。增加步数可能提高质量但降低速度。",
+        "sample_shift": "分布偏移",
+        "sample_shift_info": "控制样本分布偏移的程度。值越大表示偏移越明显。",
+        "cfg_scale": "CFG缩放因子",
+        "cfg_scale_info": "控制提示词的影响强度。值越高，提示词的影响越大。当值为1时，自动禁用CFG。",
+        "enable_cfg": "启用无分类器引导",
+        "fps": "每秒帧数(FPS)",
+        "fps_info": "视频的每秒帧数。较高的FPS会产生更流畅的视频。",
+        "num_frames": "总帧数",
+        "num_frames_info": "视频中的总帧数。更多帧数会产生更长的视频。",
+        "video_duration": "视频时长（秒）",
+        "video_duration_info": "视频的时长（秒）。实际帧数 = 时长 × FPS。",
+        "output_path": "输出视频路径",
+        "output_path_info": "必须包含.mp4扩展名。如果留空或使用默认值，将自动生成唯一文件名。",
+        "output_image_path": "输出图片路径",
+        "output_image_path_info": "必须包含.png扩展名。如果留空或使用默认值，将自动生成唯一文件名。",
+        "output_result": "📤 生成的结果",
+        "output_image": "输出图片",
+        "generate_video": "🎬 生成视频",
+        "generate_image": "🖼️ 生成图片",
+        "infer_steps_image_info": "图片编辑的推理步数，默认为8。",
+        "aspect_ratio": "宽高比",
+        "aspect_ratio_info": "选择生成图片的宽高比",
+        "model_config_hint_image": "💡 **提示**：请确保以下每个模型选项至少有一个已下载✅的模型可用，否则可能无法正常生成图片。",
+        "download": "📥 下载",
+        "downloaded": "✅ 已下载",
+        "not_downloaded": "❌ 未下载",
+        "download_complete": "✅ {model_name} 下载完成",
+        "download_start": "开始从 {source} 下载 {model_name}...",
+        "please_select_model": "请先选择模型",
+        "loading_models": "正在加载 Hugging Face 模型列表缓存...",
+        "models_loaded": "模型列表缓存加载完成",
+        "use_lora": "使用 LoRA",
+        "lora": "🎨 LoRA",
+        "lora_info": "选择要使用的 LoRA 模型",
+        "lora_strength": "LoRA 强度",
+        "lora_strength_info": "控制 LoRA 的影响强度，范围 0-10",
+        "high_noise_lora": "🔊 高噪模型 LoRA",
+        "high_noise_lora_info": "选择高噪模型使用的 LoRA",
+        "high_noise_lora_strength": "高噪模型 LoRA 强度",
+        "high_noise_lora_strength_info": "控制高噪模型 LoRA 的影响强度，范围 0-10",
+        "low_noise_lora": "🔇 低噪模型 LoRA",
+        "low_noise_lora_info": "选择低噪模型使用的 LoRA",
+        "low_noise_lora_strength": "低噪模型 LoRA 强度",
+        "low_noise_lora_strength_info": "控制低噪模型 LoRA 的影响强度，范围 0-10",
+    },
+    "en": {
+        "title": "🎬 LightX2V Image/Video Generator",
+        "model_config": "🗂️ Model Configuration",
+        "model_config_hint": "💡 **Tip**: Please ensure at least one downloaded ✅ model is available for each model option below, otherwise video generation may fail.",
+        "fp8_not_supported": "⚠️ **Your device does not support fp8 inference**, fp8 model options have been automatically hidden.",
+        "model_type": "Model Type",
+        "model_type_info": "Wan2.2 requires separate high-noise and low-noise models; Qwen-Image-Edit-2511 is for image editing (i2i); Qwen-Image-2512 is for text-to-image (t2i); Z-Image-Turbo is for text-to-image (t2i)",
+        "qwen3_encoder": "📝 Qwen3 Encoder",
+        "scheduler": "⏱️ Scheduler",
+        "qwen25vl_encoder": "📝 Qwen25-VL Encoder",
+        "task_type": "Task Type",
+        "task_type_info": "I2V: Image-to-Video, T2V: Text-to-Video, T2I: Text-to-Image, I2I: Image Editing",
+        "download_source": "📥 Download Source",
+        "download_source_info": "Select model download source",
+        "diffusion_model": "🎨 Diffusion Model",
+        "high_noise_model": "🔊 High Noise Model",
+        "low_noise_model": "🔇 Low Noise Model",
+        "text_encoder": "📝 Text Encoder",
+        "text_encoder_tokenizer": "📝 Text Encoder Tokenizer",
+        "image_encoder": "🖼️ Image Encoder",
+        "image_encoder_tokenizer": "🖼️ Image Encoder Tokenizer",
+        "vae": "🎞️ VAE Encoder/Decoder",
+        "attention_operator": "⚡ Attention Operator",
+        "attention_operator_info": "Use appropriate attention operator to accelerate inference",
+        "quant_operator": "⚡ Matrix Multiplication Operator",
+        "quant_operator_info": "Select low-precision matrix multiplication operator to accelerate inference",
+        "input_params": "📥 Input Parameters",
+        "input_image": "Input Image (drag multiple images)",
+        "image_preview": "Uploaded Image Preview",
+        "image_path": "Image Path",
+        "prompt": "Prompt",
+        "prompt_placeholder": "Describe video/image content...",
+        "negative_prompt": "Negative Prompt",
+        "negative_prompt_placeholder": "Content you don't want in the video/image...",
+        "max_resolution": "Max Resolution",
+        "max_resolution_info": "Reduce resolution if VRAM is insufficient",
+        "random_seed": "Random Seed",
+        "infer_steps": "Inference Steps",
+        "infer_steps_distill": "Distill model inference steps default to 4.",
+        "infer_steps_info": "Number of inference steps for video generation. More steps may improve quality but reduce speed.",
+        "sample_shift": "Sample Shift",
+        "sample_shift_info": "Control the degree of sample distribution shift. Higher values indicate more obvious shift.",
+        "cfg_scale": "CFG Scale",
+        "cfg_scale_info": "Control the influence strength of prompts. Higher values mean stronger prompt influence. When value is 1, CFG is automatically disabled.",
+        "enable_cfg": "Enable Classifier-Free Guidance",
+        "fps": "Frames Per Second (FPS)",
+        "fps_info": "Frames per second of the video. Higher FPS produces smoother videos.",
+        "num_frames": "Total Frames",
+        "num_frames_info": "Total number of frames in the video. More frames produce longer videos.",
+        "video_duration": "Video Duration (seconds)",
+        "video_duration_info": "Duration of the video in seconds. Actual frames = duration × FPS.",
+        "output_path": "Output Video Path",
+        "output_path_info": "Must include .mp4 extension. If left empty or using default value, a unique filename will be automatically generated.",
+        "output_image_path": "Output Image Path",
+        "output_image_path_info": "Must include .png extension. If left empty or using default value, a unique filename will be automatically generated.",
+        "output_result": "📤 Generated Result",
+        "output_image": "Output Image",
+        "generate_video": "🎬 Generate Video",
+        "generate_image": "🖼️ Generate Image",
+        "infer_steps_image_info": "Number of inference steps for image editing, default is 8.",
+        "aspect_ratio": "Aspect Ratio",
+        "aspect_ratio_info": "Select the aspect ratio for generated images",
+        "model_config_hint_image": "💡 **Tip**: Please ensure at least one downloaded ✅ model is available for each model option below, otherwise image generation may fail.",
+        "download": "📥 Download",
+        "downloaded": "✅ Downloaded",
+        "not_downloaded": "❌ Not Downloaded",
+        "download_complete": "✅ {model_name} download complete",
+        "download_start": "Starting to download {model_name} from {source}...",
+        "please_select_model": "Please select a model first",
+        "loading_models": "Loading Hugging Face model list cache...",
+        "models_loaded": "Model list cache loaded",
+        "use_lora": "Use LoRA",
+        "lora": "🎨 LoRA",
+        "lora_info": "Select LoRA model to use",
+        "lora_strength": "LoRA Strength",
+        "lora_strength_info": "Control LoRA influence strength, range 0-10",
+        "high_noise_lora": "🔊 High Noise Model LoRA",
+        "high_noise_lora_info": "Select high noise model LoRA to use",
+        "high_noise_lora_strength": "High Noise Model LoRA Strength",
+        "high_noise_lora_strength_info": "Control high noise model LoRA influence strength, range 0-10",
+        "low_noise_lora": "🔇 Low Noise Model LoRA",
+        "low_noise_lora_info": "Select low noise model LoRA to use",
+        "low_noise_lora_strength": "Low Noise Model LoRA Strength",
+        "low_noise_lora_strength_info": "Control low noise model LoRA influence strength, range 0-10",
+    },
+}
+
+
+def t(key: str, lang: str = None) -> str:
+    """获取翻译文本"""
+    if lang is None:
+        lang = DEFAULT_LANG
+
+    if lang not in TRANSLATIONS:
+        lang = "zh"
+
+    return TRANSLATIONS[lang].get(key, key)
+
+
+def set_language(lang: str):
+    """设置语言"""
+    global DEFAULT_LANG
+    if lang in TRANSLATIONS:
+        DEFAULT_LANG = lang
+        os.environ["GRADIO_LANG"] = lang