diff --git a/.gitignore b/.gitignore
index 13ecfc75b02df61296a4e0b7118c605debba1661..5c1640ab87605d3618d4e1b78cbf565e9b580cea 100644
--- a/.gitignore
+++ b/.gitignore
@@ -2,4 +2,5 @@ app1.py
 app2.py
 demo_utils1.py
 tmp
-models
\ No newline at end of file
+models
+stablediffusionapi
\ No newline at end of file
diff --git a/app1_a.py b/app1_a.py
new file mode 100644
index 0000000000000000000000000000000000000000..1f4fe30453d8d3ac24af857f546a89c192df4e8e
--- /dev/null
+++ b/app1_a.py
@@ -0,0 +1,386 @@
+import os
+import gradio as gr
+import numpy as np
+from enum import Enum
+import db_examples
+import cv2
+
+
+from demo_utils1 import *
+
+from misc_utils.train_utils import unit_test_create_model
+from misc_utils.image_utils import save_tensor_to_gif, save_tensor_to_images
+import os
+from PIL import Image
+import torch
+import torchvision
+from torchvision import transforms
+from einops import rearrange
+import imageio
+import time
+
+from torchvision.transforms import functional as F
+from torch.hub import download_url_to_file
+
+import os
+
+# 推理设置
+from pl_trainer.inference.inference import InferenceIP2PVideo
+from tqdm import tqdm
+
+
+# if not os.path.exists(filename):
+#     original_path = os.getcwd()
+#     base_path = './models'
+#     os.makedirs(base_path, exist_ok=True)
+
+#     # 直接在代码中写入 Token（注意安全风险）
+#     GIT_TOKEN = "955b8ea91095840b76fe38b90a088c200d4c813c"
+#     repo_url = f"https://YeFang:{GIT_TOKEN}@code.openxlab.org.cn/YeFang/RIV_models.git"
+
+#     try:
+#         if os.system(f'git clone {repo_url} {base_path}') != 0:
+#             raise RuntimeError("Git 克隆失败")
+#         os.chdir(base_path)
+#         if os.system('git lfs pull') != 0:
+#             raise RuntimeError("Git LFS 拉取失败")
+#     finally:
+#         os.chdir(original_path)
+
+def tensor_to_pil_image(x):
+    """
+    将 4D PyTorch 张量转换为 PIL 图像。
+    """
+    x = x.float()  # 确保张量类型为 float
+    grid_img = torchvision.utils.make_grid(x, nrow=4).permute(1, 2, 0).detach().cpu().numpy()
+    grid_img = (grid_img * 255).clip(0, 255).astype("uint8")  # 将 [0, 1] 范围转换为 [0, 255]
+    return Image.fromarray(grid_img)
+
+def frame_to_batch(x):
+    """
+    将帧维度转换为批次维度。
+    """
+    return rearrange(x, 'b f c h w -> (b f) c h w')
+
+def clip_image(x, min=0., max=1.):
+    """
+    将图像张量裁剪到指定的最小和最大值。
+    """
+    return torch.clamp(x, min=min, max=max)
+
+def unnormalize(x):
+    """
+    将张量范围从 [-1, 1] 转换到 [0, 1]。
+    """
+    return (x + 1) / 2
+
+
+# 读取图像文件
+def read_images_from_directory(directory, num_frames=16):
+    images = []
+    for i in range(num_frames):
+        img_path = os.path.join(directory, f'{i:04d}.png')
+        img = imageio.imread(img_path)
+        images.append(torch.tensor(img).permute(2, 0, 1))  # Convert to Tensor (C, H, W)
+    return images
+
+def load_and_process_images(folder_path):
+    """
+    读取文件夹中的所有图片，将它们转换为 [-1, 1] 范围的张量并返回一个 4D 张量。
+    """
+    processed_images = []
+    transform = transforms.Compose([
+        transforms.ToTensor(),
+        transforms.Lambda(lambda x: x * 2 - 1)  # 将 [0, 1] 转换为 [-1, 1]
+    ])
+    for filename in sorted(os.listdir(folder_path)):
+        if filename.endswith(".png"):
+            img_path = os.path.join(folder_path, filename)
+            image = Image.open(img_path).convert("RGB")
+            processed_image = transform(image)
+            processed_images.append(processed_image)
+    return torch.stack(processed_images)  # 返回 4D 张量
+
+def load_and_process_video(video_path, num_frames=16, crop_size=512):
+    """
+    读取视频文件中的前 num_frames 帧，将每一帧转换为 [-1, 1] 范围的张量，
+    并进行中心裁剪至 crop_size x crop_size，返回一个 4D 张量。
+    """
+    processed_frames = []
+    transform = transforms.Compose([
+        transforms.CenterCrop(crop_size),       # 中心裁剪
+        transforms.ToTensor(),
+        transforms.Lambda(lambda x: x * 2 - 1)  # 将 [0, 1] 转换为 [-1, 1]
+    ])
+    
+    # 使用 OpenCV 读取视频
+    cap = cv2.VideoCapture(video_path)
+    
+    if not cap.isOpened():
+        raise ValueError(f"无法打开视频文件: {video_path}")
+
+    frame_count = 0
+
+    while frame_count < num_frames:
+        ret, frame = cap.read()
+        if not ret:
+            break  # 视频帧读取完毕或视频帧不足
+        
+        # 转换为 RGB 格式
+        frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+        image = Image.fromarray(frame)
+        
+        # 应用转换
+        processed_frame = transform(image)
+        processed_frames.append(processed_frame)
+        
+        frame_count += 1
+
+    cap.release()  # 释放视频资源
+
+    if len(processed_frames) < num_frames:
+        raise ValueError(f"视频帧不足 {num_frames} 帧，仅找到 {len(processed_frames)} 帧。")
+
+    return torch.stack(processed_frames)  # 返回 4D 张量 (帧数, 通道数, 高度, 宽度)
+
+
+def clear_cache(output_path):
+    if os.path.exists(output_path):
+        os.remove(output_path)
+    return None
+
+
+#! 加载模型
+# 配置路径和加载模型
+config_path = 'configs/instruct_v2v_ic_gradio.yaml'
+diffusion_model = unit_test_create_model(config_path)
+diffusion_model = diffusion_model.to('cuda')
+
+# 加载模型检查点
+# ckpt_path = 'models/relvid_mm_sd15_fbc_unet.pth' #! change
+# ckpt_path = 'tmp/pytorch_model.bin'
+# 下载文件
+
+os.makedirs('models', exist_ok=True)
+model_path = "models/relvid_mm_sd15_fbc_unet.pth"
+
+if not os.path.exists(model_path):
+    download_url_to_file(url='https://huggingface.co/aleafy/RelightVid/resolve/main/relvid_mm_sd15_fbc_unet.pth', dst=model_path)
+
+
+ckpt = torch.load(model_path, map_location='cpu')
+diffusion_model.load_state_dict(ckpt, strict=False)
+
+
+# import pdb; pdb.set_trace()
+
+# 更改全局临时目录
+new_tmp_dir = "./demo/gradio_bg"
+os.makedirs(new_tmp_dir, exist_ok=True)
+
+# import pdb; pdb.set_trace()
+
+def save_video_from_frames(image_pred, save_pth, fps=8):
+    """
+    将 image_pred 中的帧保存为视频文件。
+
+    参数:
+    - image_pred: Tensor，形状为 (1, 16, 3, 512, 512)
+    - save_pth: 保存视频的路径，例如 "output_video.mp4"
+    - fps: 视频的帧率
+    """
+    # 视频参数
+    num_frames = image_pred.shape[1]
+    frame_height, frame_width = 512, 512  # 目标尺寸
+    fourcc = cv2.VideoWriter_fourcc(*'mp4v')  # 使用 mp4 编码格式
+
+    # 创建 VideoWriter 对象
+    out = cv2.VideoWriter(save_pth, fourcc, fps, (frame_width, frame_height))
+
+    for i in range(num_frames):
+        # 反归一化 + 转换为 0-255 范围
+        pred_frame = clip_image(unnormalize(image_pred[0][i].unsqueeze(0))) * 255
+        pred_frame_resized = pred_frame.squeeze(0).detach().cpu()  # (3, 512, 512)
+        pred_frame_resized = pred_frame_resized.permute(1, 2, 0).numpy().astype("uint8")  # (512, 512, 3)
+
+        # Resize 到 256x256
+        pred_frame_resized = cv2.resize(pred_frame_resized, (frame_width, frame_height))
+
+        # 将 RGB 转为 BGR（因为 OpenCV 使用 BGR 格式）
+        pred_frame_bgr = cv2.cvtColor(pred_frame_resized, cv2.COLOR_RGB2BGR)
+
+        # 写入帧到视频
+        out.write(pred_frame_bgr)
+
+    # 释放 VideoWriter 资源
+    out.release()
+    print(f"视频已保存至 {save_pth}")
+
+
+inf_pipe = InferenceIP2PVideo(
+        diffusion_model.unet, 
+        scheduler='ddpm',
+        num_ddim_steps=20
+    )
+
+# 伪函数占位（生成空白视频）
+def dummy_process(input_fg, input_bg):
+    # import pdb; pdb.set_trace()
+
+    diffusion_model.to(torch.float16)
+    fg_tensor = load_and_process_video(input_fg).cuda().unsqueeze(0).to(dtype=torch.float16)
+    bg_tensor = load_and_process_video(input_bg).cuda().unsqueeze(0).to(dtype=torch.float16) # (1, 16, 4, 64, 64)
+
+    cond_fg_tensor = diffusion_model.encode_image_to_latent(fg_tensor)  # (1, 16, 4, 64, 64)
+    cond_bg_tensor = diffusion_model.encode_image_to_latent(bg_tensor)
+    cond_tensor = torch.cat((cond_fg_tensor, cond_bg_tensor), dim=2)
+
+    # 初始化潜变量
+    init_latent = torch.randn_like(cond_fg_tensor)    
+
+    EDIT_PROMPT = 'change the background'
+    VIDEO_CFG = 1.2
+    TEXT_CFG = 7.5
+    text_cond = diffusion_model.encode_text([EDIT_PROMPT])  # (1, 77, 768)
+    text_uncond = diffusion_model.encode_text([''])
+    # to float16
+    print('------------to float 16----------------')
+    init_latent, text_cond, text_uncond, cond_tensor = (
+        init_latent.to(dtype=torch.float16),
+        text_cond.to(dtype=torch.float16),
+        text_uncond.to(dtype=torch.float16),
+        cond_tensor.to(dtype=torch.float16)
+    )
+    inf_pipe.unet.to(torch.float16)
+    latent_pred = inf_pipe(
+        latent=init_latent,
+        text_cond=text_cond,
+        text_uncond=text_uncond,
+        img_cond=cond_tensor,
+        text_cfg=TEXT_CFG,
+        img_cfg=VIDEO_CFG,
+    )['latent']
+    
+
+    image_pred = diffusion_model.decode_latent_to_image(latent_pred) # (1,16,3,512,512)
+    output_path = os.path.join(new_tmp_dir, f"output_{int(time.time())}.mp4")
+    # clear_cache(output_path)
+    
+    save_video_from_frames(image_pred, output_path)
+    # import pdb; pdb.set_trace()
+    # fps = 8
+    # frames = []
+    # for i in range(16):
+    #     pred_frame = clip_image(unnormalize(image_pred[0][i].unsqueeze(0))) * 255
+    #     pred_frame_resized = pred_frame.squeeze(0).detach().cpu() #(3,512,512)
+    #     pred_frame_resized = pred_frame_resized.permute(1, 2, 0).detach().cpu().numpy().astype("uint8") #(512,512,3) np
+    #     Image.fromarray(pred_frame_resized).save(save_pth)
+
+    # # 生成一个简单的黑色视频作为示例
+    # output_path = os.path.join(new_tmp_dir, "output.mp4")
+    # fourcc = cv2.VideoWriter_fourcc(*'mp4v')
+    # out = cv2.VideoWriter(output_path, fourcc, 20.0, (512, 512))
+
+    # for _ in range(60):  # 生成 3 秒的视频（20fps）
+    #     frame = np.zeros((512, 512, 3), dtype=np.uint8)
+    #     out.write(frame)
+    # out.release()
+    torch.cuda.empty_cache()
+
+    return output_path
+
+# 枚举类用于背景选择
+class BGSource(Enum):
+    UPLOAD = "Use Background Video"
+    UPLOAD_FLIP = "Use Flipped Background Video"
+    UPLOAD_REVERSE = "Use Reversed Background Video"
+
+
+# Quick prompts 示例
+quick_prompts = [
+    'beautiful woman',
+    'handsome man',
+    'beautiful woman, cinematic lighting',
+    'handsome man, cinematic lighting',
+    'beautiful woman, natural lighting',
+    'handsome man, natural lighting',
+    'beautiful woman, neo punk lighting, cyberpunk',
+    'handsome man, neo punk lighting, cyberpunk',
+]
+quick_prompts = [[x] for x in quick_prompts]
+
+# Gradio UI 结构
+block = gr.Blocks().queue()
+with block:
+    with gr.Row():
+        gr.Markdown("## IC-Light (Relighting with Foreground and Background Video Condition)")
+
+    with gr.Row():
+        with gr.Column():
+            with gr.Row():
+                input_fg = gr.Video(label="Foreground Video", height=370, width=370, visible=True)
+                input_bg = gr.Video(label="Background Video", height=370, width=370, visible=True)
+
+            prompt = gr.Textbox(label="Prompt")
+            bg_source = gr.Radio(choices=[e.value for e in BGSource],
+                                 value=BGSource.UPLOAD.value,
+                                 label="Background Source", type='value')
+
+            example_prompts = gr.Dataset(samples=quick_prompts, label='Prompt Quick List', components=[prompt])
+            bg_gallery = gr.Gallery(height=450, object_fit='contain', label='Background Quick List', value=db_examples.bg_samples, columns=5, allow_preview=False)
+            relight_button = gr.Button(value="Relight")
+
+            with gr.Group():
+                with gr.Row():
+                    num_samples = gr.Slider(label="Videos", minimum=1, maximum=12, value=1, step=1)
+                    seed = gr.Number(label="Seed", value=12345, precision=0)
+                with gr.Row():
+                    video_width = gr.Slider(label="Video Width", minimum=256, maximum=1024, value=512, step=64)
+                    video_height = gr.Slider(label="Video Height", minimum=256, maximum=1024, value=640, step=64)
+
+            with gr.Accordion("Advanced options", open=False):
+                steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=20, step=1)
+                cfg = gr.Slider(label="CFG Scale", minimum=1.0, maximum=32.0, value=7.0, step=0.01)
+                highres_scale = gr.Slider(label="Highres Scale", minimum=1.0, maximum=3.0, value=1.5, step=0.01)
+                highres_denoise = gr.Slider(label="Highres Denoise", minimum=0.1, maximum=0.9, value=0.5, step=0.01)
+                a_prompt = gr.Textbox(label="Added Prompt", value='best quality')
+                n_prompt = gr.Textbox(label="Negative Prompt", value='lowres, bad anatomy, bad hands, cropped, worst quality')
+                normal_button = gr.Button(value="Compute Normal (4x Slower)")
+
+        with gr.Column():
+            result_video = gr.Video(label='Output Video', height=600, width=600, visible=True)
+            fg_gallery = gr.Gallery(width=600, object_fit='contain', label='Foreground Quick List', value=db_examples.bg_samples, columns=4, allow_preview=False)
+
+    # 输入列表
+    # ips = [input_fg, input_bg, prompt, video_width, video_height, num_samples, seed, steps, a_prompt, n_prompt, cfg, highres_scale, highres_denoise, bg_source]
+    ips = [input_fg, input_bg]
+
+    # 按钮绑定处理函数
+    # relight_button.click(fn=lambda: None, inputs=[], outputs=[result_video])
+
+    relight_button.click(fn=dummy_process, inputs=ips, outputs=[result_video])
+    
+    normal_button.click(fn=dummy_process, inputs=ips, outputs=[result_video])
+
+    # 背景库选择
+    def bg_gallery_selected(gal, evt: gr.SelectData):
+        # import pdb; pdb.set_trace()
+        # img_path = gal[evt.index][0]
+        img_path = db_examples.bg_samples[evt.index]
+        video_path = img_path.replace('frames/0000.png', 'cropped_video.mp4')
+        return video_path
+
+    bg_gallery.select(bg_gallery_selected, inputs=bg_gallery, outputs=input_bg)
+
+    # 示例
+    # dummy_video_for_outputs = gr.Video(visible=False, label='Result')
+    gr.Examples(
+        fn=lambda *args: args[-1],
+        examples=db_examples.background_conditioned_examples,
+        inputs=[input_fg, input_bg, prompt, bg_source, video_width, video_height, seed, result_video],
+        outputs=[result_video],
+        run_on_click=True, examples_per_page=1024
+    )
+
+# 启动 Gradio 应用
+block.launch(server_name='0.0.0.0', server_port=10003, share=True)
diff --git a/app1_bf.py b/app1_bf.py
new file mode 100644
index 0000000000000000000000000000000000000000..cf2144dd436072eba74e9b4a9b2e85954051caaa
--- /dev/null
+++ b/app1_bf.py
@@ -0,0 +1,388 @@
+import os
+import gradio as gr
+import numpy as np
+from enum import Enum
+import db_examples
+import cv2
+
+
+from demo_utils1 import *
+
+from misc_utils.train_utils import unit_test_create_model
+from misc_utils.image_utils import save_tensor_to_gif, save_tensor_to_images
+import os
+from PIL import Image
+import torch
+import torchvision
+from torchvision import transforms
+from einops import rearrange
+import imageio
+import time
+
+from torchvision.transforms import functional as F
+from torch.hub import download_url_to_file
+
+import os
+
+# 推理设置
+from pl_trainer.inference.inference import InferenceIP2PVideo
+from tqdm import tqdm
+
+
+# if not os.path.exists(filename):
+#     original_path = os.getcwd()
+#     base_path = './models'
+#     os.makedirs(base_path, exist_ok=True)
+
+#     # 直接在代码中写入 Token（注意安全风险）
+#     GIT_TOKEN = "955b8ea91095840b76fe38b90a088c200d4c813c"
+#     repo_url = f"https://YeFang:{GIT_TOKEN}@code.openxlab.org.cn/YeFang/RIV_models.git"
+
+#     try:
+#         if os.system(f'git clone {repo_url} {base_path}') != 0:
+#             raise RuntimeError("Git 克隆失败")
+#         os.chdir(base_path)
+#         if os.system('git lfs pull') != 0:
+#             raise RuntimeError("Git LFS 拉取失败")
+#     finally:
+#         os.chdir(original_path)
+
+def tensor_to_pil_image(x):
+    """
+    将 4D PyTorch 张量转换为 PIL 图像。
+    """
+    x = x.float()  # 确保张量类型为 float
+    grid_img = torchvision.utils.make_grid(x, nrow=4).permute(1, 2, 0).detach().cpu().numpy()
+    grid_img = (grid_img * 255).clip(0, 255).astype("uint8")  # 将 [0, 1] 范围转换为 [0, 255]
+    return Image.fromarray(grid_img)
+
+def frame_to_batch(x):
+    """
+    将帧维度转换为批次维度。
+    """
+    return rearrange(x, 'b f c h w -> (b f) c h w')
+
+def clip_image(x, min=0., max=1.):
+    """
+    将图像张量裁剪到指定的最小和最大值。
+    """
+    return torch.clamp(x, min=min, max=max)
+
+def unnormalize(x):
+    """
+    将张量范围从 [-1, 1] 转换到 [0, 1]。
+    """
+    return (x + 1) / 2
+
+
+# 读取图像文件
+def read_images_from_directory(directory, num_frames=16):
+    images = []
+    for i in range(num_frames):
+        img_path = os.path.join(directory, f'{i:04d}.png')
+        img = imageio.imread(img_path)
+        images.append(torch.tensor(img).permute(2, 0, 1))  # Convert to Tensor (C, H, W)
+    return images
+
+def load_and_process_images(folder_path):
+    """
+    读取文件夹中的所有图片，将它们转换为 [-1, 1] 范围的张量并返回一个 4D 张量。
+    """
+    processed_images = []
+    transform = transforms.Compose([
+        transforms.ToTensor(),
+        transforms.Lambda(lambda x: x * 2 - 1)  # 将 [0, 1] 转换为 [-1, 1]
+    ])
+    for filename in sorted(os.listdir(folder_path)):
+        if filename.endswith(".png"):
+            img_path = os.path.join(folder_path, filename)
+            image = Image.open(img_path).convert("RGB")
+            processed_image = transform(image)
+            processed_images.append(processed_image)
+    return torch.stack(processed_images)  # 返回 4D 张量
+
+def load_and_process_video(video_path, num_frames=16, crop_size=512):
+    """
+    读取视频文件中的前 num_frames 帧，将每一帧转换为 [-1, 1] 范围的张量，
+    并进行中心裁剪至 crop_size x crop_size，返回一个 4D 张量。
+    """
+    processed_frames = []
+    transform = transforms.Compose([
+        transforms.CenterCrop(crop_size),       # 中心裁剪
+        transforms.ToTensor(),
+        transforms.Lambda(lambda x: x * 2 - 1)  # 将 [0, 1] 转换为 [-1, 1]
+    ])
+    
+    # 使用 OpenCV 读取视频
+    cap = cv2.VideoCapture(video_path)
+    
+    if not cap.isOpened():
+        raise ValueError(f"无法打开视频文件: {video_path}")
+
+    frame_count = 0
+
+    while frame_count < num_frames:
+        ret, frame = cap.read()
+        if not ret:
+            break  # 视频帧读取完毕或视频帧不足
+        
+        # 转换为 RGB 格式
+        frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+        image = Image.fromarray(frame)
+        
+        # 应用转换
+        processed_frame = transform(image)
+        processed_frames.append(processed_frame)
+        
+        frame_count += 1
+
+    cap.release()  # 释放视频资源
+
+    if len(processed_frames) < num_frames:
+        raise ValueError(f"视频帧不足 {num_frames} 帧，仅找到 {len(processed_frames)} 帧。")
+
+    return torch.stack(processed_frames)  # 返回 4D 张量 (帧数, 通道数, 高度, 宽度)
+
+
+def clear_cache(output_path):
+    if os.path.exists(output_path):
+        os.remove(output_path)
+    return None
+
+
+#! 加载模型
+# 配置路径和加载模型
+config_path = 'configs/instruct_v2v_ic_gradio.yaml'
+diffusion_model = unit_test_create_model(config_path)
+diffusion_model = diffusion_model.to('cuda')
+
+# 加载模型检查点
+# ckpt_path = 'models/relvid_mm_sd15_fbc_unet.pth' #! change
+# ckpt_path = 'tmp/pytorch_model.bin'
+# 下载文件
+
+os.makedirs('models', exist_ok=True)
+model_path = "models/relvid_mm_sd15_fbc_unet.pth"
+
+if not os.path.exists(model_path):
+    download_url_to_file(url='https://huggingface.co/aleafy/RelightVid/resolve/main/relvid_mm_sd15_fbc_unet.pth', dst=model_path)
+
+
+ckpt = torch.load(model_path, map_location='cpu')
+diffusion_model.load_state_dict(ckpt, strict=False)
+
+
+# import pdb; pdb.set_trace()
+
+# 更改全局临时目录
+new_tmp_dir = "./demo/gradio_bg"
+os.makedirs(new_tmp_dir, exist_ok=True)
+
+# import pdb; pdb.set_trace()
+
+def save_video_from_frames(image_pred, save_pth, fps=8):
+    """
+    将 image_pred 中的帧保存为视频文件。
+
+    参数:
+    - image_pred: Tensor，形状为 (1, 16, 3, 512, 512)
+    - save_pth: 保存视频的路径，例如 "output_video.mp4"
+    - fps: 视频的帧率
+    """
+    # 视频参数
+    num_frames = image_pred.shape[1]
+    frame_height, frame_width = 512, 512  # 目标尺寸
+    fourcc = cv2.VideoWriter_fourcc(*'mp4v')  # 使用 mp4 编码格式
+
+    # 创建 VideoWriter 对象
+    out = cv2.VideoWriter(save_pth, fourcc, fps, (frame_width, frame_height))
+
+    for i in range(num_frames):
+        # 反归一化 + 转换为 0-255 范围
+        pred_frame = clip_image(unnormalize(image_pred[0][i].unsqueeze(0))) * 255
+        pred_frame_resized = pred_frame.squeeze(0).detach().cpu()  # (3, 512, 512)
+        pred_frame_resized = pred_frame_resized.permute(1, 2, 0).numpy().astype("uint8")  # (512, 512, 3)
+
+        # Resize 到 256x256
+        pred_frame_resized = cv2.resize(pred_frame_resized, (frame_width, frame_height))
+
+        # 将 RGB 转为 BGR（因为 OpenCV 使用 BGR 格式）
+        pred_frame_bgr = cv2.cvtColor(pred_frame_resized, cv2.COLOR_RGB2BGR)
+
+        # 写入帧到视频
+        out.write(pred_frame_bgr)
+
+    # 释放 VideoWriter 资源
+    out.release()
+    print(f"视频已保存至 {save_pth}")
+
+
+inf_pipe = InferenceIP2PVideo(
+        diffusion_model.unet, 
+        scheduler='ddpm',
+        num_ddim_steps=20
+    )
+
+# 伪函数占位（生成空白视频）
+def dummy_process(input_fg, input_bg):
+    # import pdb; pdb.set_trace()
+
+    diffusion_model.to(torch.float16)
+    fg_tensor = load_and_process_video(input_fg).cuda().unsqueeze(0).to(dtype=torch.float16)
+    bg_tensor = load_and_process_video(input_bg).cuda().unsqueeze(0).to(dtype=torch.float16) # (1, 16, 4, 64, 64)
+
+    cond_fg_tensor = diffusion_model.encode_image_to_latent(fg_tensor)  # (1, 16, 4, 64, 64)
+    cond_bg_tensor = diffusion_model.encode_image_to_latent(bg_tensor)
+    cond_tensor = torch.cat((cond_fg_tensor, cond_bg_tensor), dim=2)
+
+    # 初始化潜变量
+    init_latent = torch.randn_like(cond_fg_tensor)    
+
+    EDIT_PROMPT = 'change the background'
+    VIDEO_CFG = 1.2
+    TEXT_CFG = 7.5
+    text_cond = diffusion_model.encode_text([EDIT_PROMPT])  # (1, 77, 768)
+    text_uncond = diffusion_model.encode_text([''])
+    # to float16
+    print('------------to float 16----------------')
+    init_latent, text_cond, text_uncond, cond_tensor = (
+        init_latent.to(dtype=torch.float16),
+        text_cond.to(dtype=torch.float16),
+        text_uncond.to(dtype=torch.float16),
+        cond_tensor.to(dtype=torch.float16)
+    )
+    inf_pipe.unet.to(torch.float16)
+    latent_pred = inf_pipe(
+        latent=init_latent,
+        text_cond=text_cond,
+        text_uncond=text_uncond,
+        img_cond=cond_tensor,
+        text_cfg=TEXT_CFG,
+        img_cfg=VIDEO_CFG,
+    )['latent']
+    
+
+    image_pred = diffusion_model.decode_latent_to_image(latent_pred) # (1,16,3,512,512)
+    output_path = os.path.join(new_tmp_dir, f"output_{int(time.time())}.mp4")
+    # clear_cache(output_path)
+    
+    save_video_from_frames(image_pred, output_path)
+    # import pdb; pdb.set_trace()
+    # fps = 8
+    # frames = []
+    # for i in range(16):
+    #     pred_frame = clip_image(unnormalize(image_pred[0][i].unsqueeze(0))) * 255
+    #     pred_frame_resized = pred_frame.squeeze(0).detach().cpu() #(3,512,512)
+    #     pred_frame_resized = pred_frame_resized.permute(1, 2, 0).detach().cpu().numpy().astype("uint8") #(512,512,3) np
+    #     Image.fromarray(pred_frame_resized).save(save_pth)
+
+    # # 生成一个简单的黑色视频作为示例
+    # output_path = os.path.join(new_tmp_dir, "output.mp4")
+    # fourcc = cv2.VideoWriter_fourcc(*'mp4v')
+    # out = cv2.VideoWriter(output_path, fourcc, 20.0, (512, 512))
+
+    # for _ in range(60):  # 生成 3 秒的视频（20fps）
+    #     frame = np.zeros((512, 512, 3), dtype=np.uint8)
+    #     out.write(frame)
+    # out.release()
+    torch.cuda.empty_cache()
+
+    return output_path
+
+# 枚举类用于背景选择
+class BGSource(Enum):
+    UPLOAD = "Use Background Video"
+    UPLOAD_FLIP = "Use Flipped Background Video"
+    LEFT = "Left Light"
+    RIGHT = "Right Light"
+    TOP = "Top Light"
+    BOTTOM = "Bottom Light"
+    GREY = "Ambient"
+
+# Quick prompts 示例
+quick_prompts = [
+    'beautiful woman',
+    'handsome man',
+    'beautiful woman, cinematic lighting',
+    'handsome man, cinematic lighting',
+    'beautiful woman, natural lighting',
+    'handsome man, natural lighting',
+    'beautiful woman, neo punk lighting, cyberpunk',
+    'handsome man, neo punk lighting, cyberpunk',
+]
+quick_prompts = [[x] for x in quick_prompts]
+
+# Gradio UI 结构
+block = gr.Blocks().queue()
+with block:
+    with gr.Row():
+        gr.Markdown("## IC-Light (Relighting with Foreground and Background Video Condition)")
+
+    with gr.Row():
+        with gr.Column():
+            with gr.Row():
+                input_fg = gr.Video(label="Foreground Video", height=370, width=370, visible=True)
+                input_bg = gr.Video(label="Background Video", height=370, width=370, visible=True)
+
+            prompt = gr.Textbox(label="Prompt")
+            bg_source = gr.Radio(choices=[e.value for e in BGSource],
+                                 value=BGSource.UPLOAD.value,
+                                 label="Background Source", type='value')
+
+            example_prompts = gr.Dataset(samples=quick_prompts, label='Prompt Quick List', components=[prompt])
+            bg_gallery = gr.Gallery(height=450, object_fit='contain', label='Background Quick List', value=db_examples.bg_samples, columns=5, allow_preview=False)
+            relight_button = gr.Button(value="Relight")
+
+            with gr.Group():
+                with gr.Row():
+                    num_samples = gr.Slider(label="Videos", minimum=1, maximum=12, value=1, step=1)
+                    seed = gr.Number(label="Seed", value=12345, precision=0)
+                with gr.Row():
+                    video_width = gr.Slider(label="Video Width", minimum=256, maximum=1024, value=512, step=64)
+                    video_height = gr.Slider(label="Video Height", minimum=256, maximum=1024, value=640, step=64)
+
+            with gr.Accordion("Advanced options", open=False):
+                steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=20, step=1)
+                cfg = gr.Slider(label="CFG Scale", minimum=1.0, maximum=32.0, value=7.0, step=0.01)
+                highres_scale = gr.Slider(label="Highres Scale", minimum=1.0, maximum=3.0, value=1.5, step=0.01)
+                highres_denoise = gr.Slider(label="Highres Denoise", minimum=0.1, maximum=0.9, value=0.5, step=0.01)
+                a_prompt = gr.Textbox(label="Added Prompt", value='best quality')
+                n_prompt = gr.Textbox(label="Negative Prompt", value='lowres, bad anatomy, bad hands, cropped, worst quality')
+                normal_button = gr.Button(value="Compute Normal (4x Slower)")
+
+        with gr.Column():
+            result_video = gr.Video(label='Output Video', height=600, width=600, visible=True)
+
+    # 输入列表
+    # ips = [input_fg, input_bg, prompt, video_width, video_height, num_samples, seed, steps, a_prompt, n_prompt, cfg, highres_scale, highres_denoise, bg_source]
+    ips = [input_fg, input_bg]
+
+    # 按钮绑定处理函数
+    # relight_button.click(fn=lambda: None, inputs=[], outputs=[result_video])
+
+    relight_button.click(fn=dummy_process, inputs=ips, outputs=[result_video])
+    
+    normal_button.click(fn=dummy_process, inputs=ips, outputs=[result_video])
+
+    # 背景库选择
+    def bg_gallery_selected(gal, evt: gr.SelectData):
+        # import pdb; pdb.set_trace()
+        # img_path = gal[evt.index][0]
+        img_path = db_examples.bg_samples[evt.index]
+        video_path = img_path.replace('frames/0000.png', 'cropped_video.mp4')
+        return video_path
+
+    bg_gallery.select(bg_gallery_selected, inputs=bg_gallery, outputs=input_bg)
+
+    # 示例
+    # dummy_video_for_outputs = gr.Video(visible=False, label='Result')
+    gr.Examples(
+        fn=lambda *args: args[-1],
+        examples=db_examples.background_conditioned_examples,
+        inputs=[input_fg, input_bg, prompt, bg_source, video_width, video_height, seed, result_video],
+        outputs=[result_video],
+        run_on_click=True, examples_per_page=1024
+    )
+
+# 启动 Gradio 应用
+block.launch(server_name='0.0.0.0', server_port=10002, share=True)
diff --git a/app1_bf2.py b/app1_bf2.py
new file mode 100644
index 0000000000000000000000000000000000000000..e9410f9a295d2d46bd8440b0539ab5ce8acf0822
--- /dev/null
+++ b/app1_bf2.py
@@ -0,0 +1,388 @@
+import os
+import gradio as gr
+import numpy as np
+from enum import Enum
+import db_examples
+import cv2
+
+
+from demo_utils1 import *
+
+from misc_utils.train_utils import unit_test_create_model
+from misc_utils.image_utils import save_tensor_to_gif, save_tensor_to_images
+import os
+from PIL import Image
+import torch
+import torchvision
+from torchvision import transforms
+from einops import rearrange
+import imageio
+import time
+
+from torchvision.transforms import functional as F
+from torch.hub import download_url_to_file
+
+import os
+
+# 推理设置
+from pl_trainer.inference.inference import InferenceIP2PVideo
+from tqdm import tqdm
+
+
+# if not os.path.exists(filename):
+#     original_path = os.getcwd()
+#     base_path = './models'
+#     os.makedirs(base_path, exist_ok=True)
+
+#     # 直接在代码中写入 Token（注意安全风险）
+#     GIT_TOKEN = "955b8ea91095840b76fe38b90a088c200d4c813c"
+#     repo_url = f"https://YeFang:{GIT_TOKEN}@code.openxlab.org.cn/YeFang/RIV_models.git"
+
+#     try:
+#         if os.system(f'git clone {repo_url} {base_path}') != 0:
+#             raise RuntimeError("Git 克隆失败")
+#         os.chdir(base_path)
+#         if os.system('git lfs pull') != 0:
+#             raise RuntimeError("Git LFS 拉取失败")
+#     finally:
+#         os.chdir(original_path)
+
+def tensor_to_pil_image(x):
+    """
+    将 4D PyTorch 张量转换为 PIL 图像。
+    """
+    x = x.float()  # 确保张量类型为 float
+    grid_img = torchvision.utils.make_grid(x, nrow=4).permute(1, 2, 0).detach().cpu().numpy()
+    grid_img = (grid_img * 255).clip(0, 255).astype("uint8")  # 将 [0, 1] 范围转换为 [0, 255]
+    return Image.fromarray(grid_img)
+
+def frame_to_batch(x):
+    """
+    将帧维度转换为批次维度。
+    """
+    return rearrange(x, 'b f c h w -> (b f) c h w')
+
+def clip_image(x, min=0., max=1.):
+    """
+    将图像张量裁剪到指定的最小和最大值。
+    """
+    return torch.clamp(x, min=min, max=max)
+
+def unnormalize(x):
+    """
+    将张量范围从 [-1, 1] 转换到 [0, 1]。
+    """
+    return (x + 1) / 2
+
+
+# 读取图像文件
+def read_images_from_directory(directory, num_frames=16):
+    images = []
+    for i in range(num_frames):
+        img_path = os.path.join(directory, f'{i:04d}.png')
+        img = imageio.imread(img_path)
+        images.append(torch.tensor(img).permute(2, 0, 1))  # Convert to Tensor (C, H, W)
+    return images
+
+def load_and_process_images(folder_path):
+    """
+    读取文件夹中的所有图片，将它们转换为 [-1, 1] 范围的张量并返回一个 4D 张量。
+    """
+    processed_images = []
+    transform = transforms.Compose([
+        transforms.ToTensor(),
+        transforms.Lambda(lambda x: x * 2 - 1)  # 将 [0, 1] 转换为 [-1, 1]
+    ])
+    for filename in sorted(os.listdir(folder_path)):
+        if filename.endswith(".png"):
+            img_path = os.path.join(folder_path, filename)
+            image = Image.open(img_path).convert("RGB")
+            processed_image = transform(image)
+            processed_images.append(processed_image)
+    return torch.stack(processed_images)  # 返回 4D 张量
+
+def load_and_process_video(video_path, num_frames=16, crop_size=512):
+    """
+    读取视频文件中的前 num_frames 帧，将每一帧转换为 [-1, 1] 范围的张量，
+    并进行中心裁剪至 crop_size x crop_size，返回一个 4D 张量。
+    """
+    processed_frames = []
+    transform = transforms.Compose([
+        transforms.CenterCrop(crop_size),       # 中心裁剪
+        transforms.ToTensor(),
+        transforms.Lambda(lambda x: x * 2 - 1)  # 将 [0, 1] 转换为 [-1, 1]
+    ])
+    
+    # 使用 OpenCV 读取视频
+    cap = cv2.VideoCapture(video_path)
+    
+    if not cap.isOpened():
+        raise ValueError(f"无法打开视频文件: {video_path}")
+
+    frame_count = 0
+
+    while frame_count < num_frames:
+        ret, frame = cap.read()
+        if not ret:
+            break  # 视频帧读取完毕或视频帧不足
+        
+        # 转换为 RGB 格式
+        frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+        image = Image.fromarray(frame)
+        
+        # 应用转换
+        processed_frame = transform(image)
+        processed_frames.append(processed_frame)
+        
+        frame_count += 1
+
+    cap.release()  # 释放视频资源
+
+    if len(processed_frames) < num_frames:
+        raise ValueError(f"视频帧不足 {num_frames} 帧，仅找到 {len(processed_frames)} 帧。")
+
+    return torch.stack(processed_frames)  # 返回 4D 张量 (帧数, 通道数, 高度, 宽度)
+
+
+def clear_cache(output_path):
+    if os.path.exists(output_path):
+        os.remove(output_path)
+    return None
+
+
+#! 加载模型
+# 配置路径和加载模型
+config_path = 'configs/instruct_v2v_ic_gradio.yaml'
+diffusion_model = unit_test_create_model(config_path)
+diffusion_model = diffusion_model.to('cuda')
+
+# 加载模型检查点
+# ckpt_path = 'models/relvid_mm_sd15_fbc_unet.pth' #! change
+# ckpt_path = 'tmp/pytorch_model.bin'
+# 下载文件
+
+os.makedirs('models', exist_ok=True)
+model_path = "models/relvid_mm_sd15_fbc_unet.pth"
+
+if not os.path.exists(model_path):
+    download_url_to_file(url='https://huggingface.co/aleafy/RelightVid/resolve/main/relvid_mm_sd15_fbc_unet.pth', dst=model_path)
+
+
+ckpt = torch.load(model_path, map_location='cpu')
+diffusion_model.load_state_dict(ckpt, strict=False)
+
+
+# import pdb; pdb.set_trace()
+
+# 更改全局临时目录
+new_tmp_dir = "./demo/gradio_bg"
+os.makedirs(new_tmp_dir, exist_ok=True)
+
+# import pdb; pdb.set_trace()
+
+def save_video_from_frames(image_pred, save_pth, fps=8):
+    """
+    将 image_pred 中的帧保存为视频文件。
+
+    参数:
+    - image_pred: Tensor，形状为 (1, 16, 3, 512, 512)
+    - save_pth: 保存视频的路径，例如 "output_video.mp4"
+    - fps: 视频的帧率
+    """
+    # 视频参数
+    num_frames = image_pred.shape[1]
+    frame_height, frame_width = 512, 512  # 目标尺寸
+    fourcc = cv2.VideoWriter_fourcc(*'mp4v')  # 使用 mp4 编码格式
+
+    # 创建 VideoWriter 对象
+    out = cv2.VideoWriter(save_pth, fourcc, fps, (frame_width, frame_height))
+
+    for i in range(num_frames):
+        # 反归一化 + 转换为 0-255 范围
+        pred_frame = clip_image(unnormalize(image_pred[0][i].unsqueeze(0))) * 255
+        pred_frame_resized = pred_frame.squeeze(0).detach().cpu()  # (3, 512, 512)
+        pred_frame_resized = pred_frame_resized.permute(1, 2, 0).numpy().astype("uint8")  # (512, 512, 3)
+
+        # Resize 到 256x256
+        pred_frame_resized = cv2.resize(pred_frame_resized, (frame_width, frame_height))
+
+        # 将 RGB 转为 BGR（因为 OpenCV 使用 BGR 格式）
+        pred_frame_bgr = cv2.cvtColor(pred_frame_resized, cv2.COLOR_RGB2BGR)
+
+        # 写入帧到视频
+        out.write(pred_frame_bgr)
+
+    # 释放 VideoWriter 资源
+    out.release()
+    print(f"视频已保存至 {save_pth}")
+
+
+inf_pipe = InferenceIP2PVideo(
+        diffusion_model.unet, 
+        scheduler='ddpm',
+        num_ddim_steps=20
+    )
+
+# 伪函数占位（生成空白视频）
+def dummy_process(input_fg, input_bg):
+    # import pdb; pdb.set_trace()
+
+    diffusion_model.to(torch.float16)
+    fg_tensor = load_and_process_video(input_fg).cuda().unsqueeze(0).to(dtype=torch.float16)
+    bg_tensor = load_and_process_video(input_bg).cuda().unsqueeze(0).to(dtype=torch.float16) # (1, 16, 4, 64, 64)
+
+    cond_fg_tensor = diffusion_model.encode_image_to_latent(fg_tensor)  # (1, 16, 4, 64, 64)
+    cond_bg_tensor = diffusion_model.encode_image_to_latent(bg_tensor)
+    cond_tensor = torch.cat((cond_fg_tensor, cond_bg_tensor), dim=2)
+
+    # 初始化潜变量
+    init_latent = torch.randn_like(cond_fg_tensor)    
+
+    EDIT_PROMPT = 'change the background'
+    VIDEO_CFG = 1.2
+    TEXT_CFG = 7.5
+    text_cond = diffusion_model.encode_text([EDIT_PROMPT])  # (1, 77, 768)
+    text_uncond = diffusion_model.encode_text([''])
+    # to float16
+    print('------------to float 16----------------')
+    init_latent, text_cond, text_uncond, cond_tensor = (
+        init_latent.to(dtype=torch.float16),
+        text_cond.to(dtype=torch.float16),
+        text_uncond.to(dtype=torch.float16),
+        cond_tensor.to(dtype=torch.float16)
+    )
+    inf_pipe.unet.to(torch.float16)
+    latent_pred = inf_pipe(
+        latent=init_latent,
+        text_cond=text_cond,
+        text_uncond=text_uncond,
+        img_cond=cond_tensor,
+        text_cfg=TEXT_CFG,
+        img_cfg=VIDEO_CFG,
+    )['latent']
+    
+
+    image_pred = diffusion_model.decode_latent_to_image(latent_pred) # (1,16,3,512,512)
+    output_path = os.path.join(new_tmp_dir, f"output_{int(time.time())}.mp4")
+    # clear_cache(output_path)
+    
+    save_video_from_frames(image_pred, output_path)
+    # import pdb; pdb.set_trace()
+    # fps = 8
+    # frames = []
+    # for i in range(16):
+    #     pred_frame = clip_image(unnormalize(image_pred[0][i].unsqueeze(0))) * 255
+    #     pred_frame_resized = pred_frame.squeeze(0).detach().cpu() #(3,512,512)
+    #     pred_frame_resized = pred_frame_resized.permute(1, 2, 0).detach().cpu().numpy().astype("uint8") #(512,512,3) np
+    #     Image.fromarray(pred_frame_resized).save(save_pth)
+
+    # # 生成一个简单的黑色视频作为示例
+    # output_path = os.path.join(new_tmp_dir, "output.mp4")
+    # fourcc = cv2.VideoWriter_fourcc(*'mp4v')
+    # out = cv2.VideoWriter(output_path, fourcc, 20.0, (512, 512))
+
+    # for _ in range(60):  # 生成 3 秒的视频（20fps）
+    #     frame = np.zeros((512, 512, 3), dtype=np.uint8)
+    #     out.write(frame)
+    # out.release()
+    torch.cuda.empty_cache()
+
+    return output_path
+
+# 枚举类用于背景选择
+class BGSource(Enum):
+    UPLOAD = "Use Background Video"
+    UPLOAD_FLIP = "Use Flipped Background Video"
+    UPLOAD_REVERSE = "Use Reversed Background Video"
+
+# Quick prompts 示例
+quick_prompts = [
+    'beautiful woman',
+    'handsome man',
+    'beautiful woman, cinematic lighting',
+    'handsome man, cinematic lighting',
+    'beautiful woman, natural lighting',
+    'handsome man, natural lighting',
+    'beautiful woman, neo punk lighting, cyberpunk',
+    'handsome man, neo punk lighting, cyberpunk',
+]
+quick_prompts = [[x] for x in quick_prompts]
+
+# Gradio UI 结构
+block = gr.Blocks().queue()
+with block:
+    with gr.Row():
+        gr.Markdown("## IC-Light (Relighting with Foreground and Background Video Condition)")
+
+    with gr.Row():
+        with gr.Column():
+            input_fg = gr.Video(label="Foreground Video", height=450, visible=True)
+        with gr.Column():
+            input_bg = gr.Video(label="Background Video", height=450, visible=True)
+        with gr.Column():
+            result_video = gr.Video(label='Output Video', height=450, visible=True)
+
+    with gr.Row():
+        with gr.Column():
+            prompt = gr.Textbox(label="Prompt")
+            bg_source = gr.Radio(choices=[e.value for e in BGSource],
+                                    value=BGSource.UPLOAD.value,
+                                    label="Background Source", type='value')
+
+            example_prompts = gr.Dataset(samples=quick_prompts, label='Prompt Quick List', components=[prompt])
+            bg_gallery = gr.Gallery(height=450, object_fit='contain', label='Background Quick List', value=db_examples.bg_samples, columns=5, allow_preview=False)
+            relight_button = gr.Button(value="Relight")
+
+            with gr.Group():
+                with gr.Row():
+                    num_samples = gr.Slider(label="Videos", minimum=1, maximum=12, value=1, step=1)
+                    seed = gr.Number(label="Seed", value=12345, precision=0)
+                with gr.Row():
+                    video_width = gr.Slider(label="Video Width", minimum=256, maximum=1024, value=512, step=64)
+                    video_height = gr.Slider(label="Video Height", minimum=256, maximum=1024, value=640, step=64)
+
+            
+        with gr.Column():
+            with gr.Accordion("Advanced options", open=False):
+                steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=20, step=1)
+                cfg = gr.Slider(label="CFG Scale", minimum=1.0, maximum=32.0, value=7.0, step=0.01)
+                highres_scale = gr.Slider(label="Highres Scale", minimum=1.0, maximum=3.0, value=1.5, step=0.01)
+                highres_denoise = gr.Slider(label="Highres Denoise", minimum=0.1, maximum=0.9, value=0.5, step=0.01)
+                a_prompt = gr.Textbox(label="Added Prompt", value='best quality')
+                n_prompt = gr.Textbox(label="Negative Prompt", value='lowres, bad anatomy, bad hands, cropped, worst quality')
+                normal_button = gr.Button(value="Compute Normal (4x Slower)")
+            
+
+    # 输入列表
+    # ips = [input_fg, input_bg, prompt, video_width, video_height, num_samples, seed, steps, a_prompt, n_prompt, cfg, highres_scale, highres_denoise, bg_source]
+    ips = [input_fg, input_bg]
+
+    # 按钮绑定处理函数
+    # relight_button.click(fn=lambda: None, inputs=[], outputs=[result_video])
+
+    relight_button.click(fn=dummy_process, inputs=ips, outputs=[result_video])
+    
+    normal_button.click(fn=dummy_process, inputs=ips, outputs=[result_video])
+
+    # 背景库选择
+    def bg_gallery_selected(gal, evt: gr.SelectData):
+        # import pdb; pdb.set_trace()
+        # img_path = gal[evt.index][0]
+        img_path = db_examples.bg_samples[evt.index]
+        video_path = img_path.replace('frames/0000.png', 'cropped_video.mp4')
+        return video_path
+
+    bg_gallery.select(bg_gallery_selected, inputs=bg_gallery, outputs=input_bg)
+
+    # 示例
+    # dummy_video_for_outputs = gr.Video(visible=False, label='Result')
+    gr.Examples(
+        fn=lambda *args: args[-1],
+        examples=db_examples.background_conditioned_examples,
+        inputs=[input_fg, input_bg, prompt, bg_source, video_width, video_height, seed, result_video],
+        outputs=[result_video],
+        run_on_click=True, examples_per_page=1024
+    )
+
+# 启动 Gradio 应用
+block.launch(server_name='0.0.0.0', server_port=10002, share=True)
diff --git a/app1_rg3.py b/app1_rg3.py
new file mode 100644
index 0000000000000000000000000000000000000000..e0b3c1761eaa491c804cba1c16a78c1d29d32f7a
--- /dev/null
+++ b/app1_rg3.py
@@ -0,0 +1,483 @@
+import os
+import gradio as gr
+import numpy as np
+from enum import Enum
+import db_examples
+import cv2
+
+
+from demo_utils1 import *
+
+from misc_utils.train_utils import unit_test_create_model
+from misc_utils.image_utils import save_tensor_to_gif, save_tensor_to_images
+import os
+from PIL import Image
+import torch
+import torchvision
+from torchvision import transforms
+from einops import rearrange
+import imageio
+import time
+
+from torchvision.transforms import functional as F
+from torch.hub import download_url_to_file
+
+import os
+
+# 推理设置
+from pl_trainer.inference.inference import InferenceIP2PVideo
+from tqdm import tqdm
+
+
+# if not os.path.exists(filename):
+#     original_path = os.getcwd()
+#     base_path = './models'
+#     os.makedirs(base_path, exist_ok=True)
+
+#     # 直接在代码中写入 Token（注意安全风险）
+#     GIT_TOKEN = "955b8ea91095840b76fe38b90a088c200d4c813c"
+#     repo_url = f"https://YeFang:{GIT_TOKEN}@code.openxlab.org.cn/YeFang/RIV_models.git"
+
+#     try:
+#         if os.system(f'git clone {repo_url} {base_path}') != 0:
+#             raise RuntimeError("Git 克隆失败")
+#         os.chdir(base_path)
+#         if os.system('git lfs pull') != 0:
+#             raise RuntimeError("Git LFS 拉取失败")
+#     finally:
+#         os.chdir(original_path)
+
+def tensor_to_pil_image(x):
+    """
+    将 4D PyTorch 张量转换为 PIL 图像。
+    """
+    x = x.float()  # 确保张量类型为 float
+    grid_img = torchvision.utils.make_grid(x, nrow=4).permute(1, 2, 0).detach().cpu().numpy()
+    grid_img = (grid_img * 255).clip(0, 255).astype("uint8")  # 将 [0, 1] 范围转换为 [0, 255]
+    return Image.fromarray(grid_img)
+
+def frame_to_batch(x):
+    """
+    将帧维度转换为批次维度。
+    """
+    return rearrange(x, 'b f c h w -> (b f) c h w')
+
+def clip_image(x, min=0., max=1.):
+    """
+    将图像张量裁剪到指定的最小和最大值。
+    """
+    return torch.clamp(x, min=min, max=max)
+
+def unnormalize(x):
+    """
+    将张量范围从 [-1, 1] 转换到 [0, 1]。
+    """
+    return (x + 1) / 2
+
+
+# 读取图像文件
+def read_images_from_directory(directory, num_frames=16):
+    images = []
+    for i in range(num_frames):
+        img_path = os.path.join(directory, f'{i:04d}.png')
+        img = imageio.imread(img_path)
+        images.append(torch.tensor(img).permute(2, 0, 1))  # Convert to Tensor (C, H, W)
+    return images
+
+def load_and_process_images(folder_path):
+    """
+    读取文件夹中的所有图片，将它们转换为 [-1, 1] 范围的张量并返回一个 4D 张量。
+    """
+    processed_images = []
+    transform = transforms.Compose([
+        transforms.ToTensor(),
+        transforms.Lambda(lambda x: x * 2 - 1)  # 将 [0, 1] 转换为 [-1, 1]
+    ])
+    for filename in sorted(os.listdir(folder_path)):
+        if filename.endswith(".png"):
+            img_path = os.path.join(folder_path, filename)
+            image = Image.open(img_path).convert("RGB")
+            processed_image = transform(image)
+            processed_images.append(processed_image)
+    return torch.stack(processed_images)  # 返回 4D 张量
+
+def load_and_process_video(video_path, num_frames=16, crop_size=512):
+    """
+    读取视频文件中的前 num_frames 帧，将每一帧转换为 [-1, 1] 范围的张量，
+    并进行中心裁剪至 crop_size x crop_size，返回一个 4D 张量。
+    """
+    processed_frames = []
+    transform = transforms.Compose([
+        transforms.CenterCrop(crop_size),       # 中心裁剪
+        transforms.ToTensor(),
+        transforms.Lambda(lambda x: x * 2 - 1)  # 将 [0, 1] 转换为 [-1, 1]
+    ])
+    
+    # 使用 OpenCV 读取视频
+    cap = cv2.VideoCapture(video_path)
+    
+    if not cap.isOpened():
+        raise ValueError(f"无法打开视频文件: {video_path}")
+
+    frame_count = 0
+
+    while frame_count < num_frames:
+        ret, frame = cap.read()
+        if not ret:
+            break  # 视频帧读取完毕或视频帧不足
+        
+        # 转换为 RGB 格式
+        frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+        image = Image.fromarray(frame)
+        
+        # 应用转换
+        processed_frame = transform(image)
+        processed_frames.append(processed_frame)
+        
+        frame_count += 1
+
+    cap.release()  # 释放视频资源
+
+    if len(processed_frames) < num_frames:
+        raise ValueError(f"视频帧不足 {num_frames} 帧，仅找到 {len(processed_frames)} 帧。")
+
+    return torch.stack(processed_frames)  # 返回 4D 张量 (帧数, 通道数, 高度, 宽度)
+
+
+def clear_cache(output_path):
+    if os.path.exists(output_path):
+        os.remove(output_path)
+    return None
+
+
+#! 加载模型
+# 配置路径和加载模型
+config_path = 'configs/instruct_v2v_ic_gradio.yaml'
+diffusion_model = unit_test_create_model(config_path)
+diffusion_model = diffusion_model.to('cuda')
+
+# 加载模型检查点
+# ckpt_path = 'models/relvid_mm_sd15_fbc_unet.pth' #! change
+# ckpt_path = 'tmp/pytorch_model.bin'
+# 下载文件
+
+os.makedirs('models', exist_ok=True)
+model_path = "models/relvid_mm_sd15_fbc_unet.pth"
+
+if not os.path.exists(model_path):
+    download_url_to_file(url='https://huggingface.co/aleafy/RelightVid/resolve/main/relvid_mm_sd15_fbc_unet.pth', dst=model_path)
+
+
+ckpt = torch.load(model_path, map_location='cpu')
+diffusion_model.load_state_dict(ckpt, strict=False)
+
+
+# import pdb; pdb.set_trace()
+
+# 更改全局临时目录
+new_tmp_dir = "./demo/gradio_bg"
+os.makedirs(new_tmp_dir, exist_ok=True)
+
+# import pdb; pdb.set_trace()
+
+def save_video_from_frames(image_pred, save_pth, fps=8):
+    """
+    将 image_pred 中的帧保存为视频文件。
+
+    参数:
+    - image_pred: Tensor，形状为 (1, 16, 3, 512, 512)
+    - save_pth: 保存视频的路径，例如 "output_video.mp4"
+    - fps: 视频的帧率
+    """
+    # 视频参数
+    num_frames = image_pred.shape[1]
+    frame_height, frame_width = 512, 512  # 目标尺寸
+    fourcc = cv2.VideoWriter_fourcc(*'mp4v')  # 使用 mp4 编码格式
+
+    # 创建 VideoWriter 对象
+    out = cv2.VideoWriter(save_pth, fourcc, fps, (frame_width, frame_height))
+
+    for i in range(num_frames):
+        # 反归一化 + 转换为 0-255 范围
+        pred_frame = clip_image(unnormalize(image_pred[0][i].unsqueeze(0))) * 255
+        pred_frame_resized = pred_frame.squeeze(0).detach().cpu()  # (3, 512, 512)
+        pred_frame_resized = pred_frame_resized.permute(1, 2, 0).numpy().astype("uint8")  # (512, 512, 3)
+
+        # Resize 到 256x256
+        pred_frame_resized = cv2.resize(pred_frame_resized, (frame_width, frame_height))
+
+        # 将 RGB 转为 BGR（因为 OpenCV 使用 BGR 格式）
+        pred_frame_bgr = cv2.cvtColor(pred_frame_resized, cv2.COLOR_RGB2BGR)
+
+        # 写入帧到视频
+        out.write(pred_frame_bgr)
+
+    # 释放 VideoWriter 资源
+    out.release()
+    print(f"视频已保存至 {save_pth}")
+
+
+inf_pipe = InferenceIP2PVideo(
+        diffusion_model.unet, 
+        scheduler='ddpm',
+        num_ddim_steps=20
+    )
+
+
+def process_example(*args):
+    v_index = args[0]
+    select_e = db_examples.background_conditioned_examples[int(v_index)-1]
+    input_fg_path = select_e[1]
+    input_bg_path = select_e[2]
+    result_video_path = select_e[-1]
+    # input_fg_img = args[1]  # 第 0 个参数
+    # input_bg_img = args[2]  # 第 1 个参数
+    # result_video_img = args[-1]  # 最后一个参数
+    
+    input_fg = input_fg_path.replace("frames/0000.png", "cropped_video.mp4")
+    input_bg = input_bg_path.replace("frames/0000.png", "cropped_video.mp4")
+    result_video = result_video_path.replace(".png", ".mp4")
+    
+    return input_fg, input_bg, result_video
+
+
+
+# 伪函数占位（生成空白视频）
+def dummy_process(input_fg, input_bg, prompt):
+    # import pdb; pdb.set_trace()
+
+    diffusion_model.to(torch.float16)
+    fg_tensor = load_and_process_video(input_fg).cuda().unsqueeze(0).to(dtype=torch.float16)
+    bg_tensor = load_and_process_video(input_bg).cuda().unsqueeze(0).to(dtype=torch.float16) # (1, 16, 4, 64, 64)
+
+    cond_fg_tensor = diffusion_model.encode_image_to_latent(fg_tensor)  # (1, 16, 4, 64, 64)
+    cond_bg_tensor = diffusion_model.encode_image_to_latent(bg_tensor)
+    cond_tensor = torch.cat((cond_fg_tensor, cond_bg_tensor), dim=2)
+
+    # 初始化潜变量
+    init_latent = torch.randn_like(cond_fg_tensor)    
+
+    # EDIT_PROMPT = 'change the background'
+    EDIT_PROMPT = prompt
+    VIDEO_CFG = 1.2
+    TEXT_CFG = 7.5
+    text_cond = diffusion_model.encode_text([EDIT_PROMPT])  # (1, 77, 768)
+    text_uncond = diffusion_model.encode_text([''])
+    # to float16
+    print('------------to float 16----------------')
+    init_latent, text_cond, text_uncond, cond_tensor = (
+        init_latent.to(dtype=torch.float16),
+        text_cond.to(dtype=torch.float16),
+        text_uncond.to(dtype=torch.float16),
+        cond_tensor.to(dtype=torch.float16)
+    )
+    inf_pipe.unet.to(torch.float16)
+    latent_pred = inf_pipe(
+        latent=init_latent,
+        text_cond=text_cond,
+        text_uncond=text_uncond,
+        img_cond=cond_tensor,
+        text_cfg=TEXT_CFG,
+        img_cfg=VIDEO_CFG,
+    )['latent']
+    
+
+    image_pred = diffusion_model.decode_latent_to_image(latent_pred) # (1,16,3,512,512)
+    output_path = os.path.join(new_tmp_dir, f"output_{int(time.time())}.mp4")
+    # clear_cache(output_path)
+    
+    save_video_from_frames(image_pred, output_path)
+    # import pdb; pdb.set_trace()
+    # fps = 8
+    # frames = []
+    # for i in range(16):
+    #     pred_frame = clip_image(unnormalize(image_pred[0][i].unsqueeze(0))) * 255
+    #     pred_frame_resized = pred_frame.squeeze(0).detach().cpu() #(3,512,512)
+    #     pred_frame_resized = pred_frame_resized.permute(1, 2, 0).detach().cpu().numpy().astype("uint8") #(512,512,3) np
+    #     Image.fromarray(pred_frame_resized).save(save_pth)
+
+    # # 生成一个简单的黑色视频作为示例
+    # output_path = os.path.join(new_tmp_dir, "output.mp4")
+    # fourcc = cv2.VideoWriter_fourcc(*'mp4v')
+    # out = cv2.VideoWriter(output_path, fourcc, 20.0, (512, 512))
+
+    # for _ in range(60):  # 生成 3 秒的视频（20fps）
+    #     frame = np.zeros((512, 512, 3), dtype=np.uint8)
+    #     out.write(frame)
+    # out.release()
+    torch.cuda.empty_cache()
+
+    return output_path
+
+# 枚举类用于背景选择
+class BGSource(Enum):
+    UPLOAD = "Use Background Video"
+    UPLOAD_FLIP = "Use Flipped Background Video"
+    UPLOAD_REVERSE = "Use Reversed Background Video"
+
+
+# Quick prompts 示例
+# quick_prompts = [
+#     'beautiful woman, fantasy setting',
+#     'beautiful woman, neon dynamic lighting',
+#     'man in suit, tunel lighting',
+#     'animated mouse, aesthetic lighting',
+#     'robot warrior, a sunset background',
+#     'yellow cat, reflective wet beach',
+#     'camera, dock, calm sunset',
+#     'astronaut, dim lighting',
+#     'astronaut, colorful balloons',
+#     'astronaut, desert landscape'
+# ]
+
+# quick_prompts = [
+#     'beautiful woman',
+#     'handsome man',
+#     'beautiful woman, cinematic lighting',
+#     'handsome man, cinematic lighting',
+#     'beautiful woman, natural lighting',
+#     'handsome man, natural lighting',
+#     'beautiful woman, neo punk lighting, cyberpunk',
+#     'handsome man, neo punk lighting, cyberpunk',
+# ]
+
+
+quick_prompts = [
+    'beautiful woman',
+    'handsome man',
+    'beautiful woman, cinematic lighting',
+    'handsome man, cinematic lighting',
+    'beautiful woman, natural lighting',
+    'handsome man, natural lighting',
+    'beautiful woman, warm lighting',
+    'handsome man, soft lighting',
+    'change the background lighting',
+]
+
+
+quick_prompts = [[x] for x in quick_prompts]
+
+# css = """
+# #foreground-gallery {
+#     width: 700 !important; /* 限制最大宽度 */
+#     max-width: 700px !important; /* 避免它自动变宽 */
+#     flex: none !important; /* 让它不自动扩展 */
+# }
+# """
+
+# Gradio UI 结构
+block = gr.Blocks().queue()
+with block:
+    with gr.Row():
+        # gr.Markdown("## RelightVid (Relighting with Foreground and Background Video Condition)")
+        gr.Markdown("# 💡RelightVid  \n### Relighting with Foreground and Background Video Condition")
+
+    with gr.Row():
+        with gr.Column():
+            with gr.Row():
+                input_fg = gr.Video(label="Foreground Video", height=380, width=420, visible=True)
+                input_bg = gr.Video(label="Background Video", height=380, width=420, visible=True)
+            
+            segment_button = gr.Button(value="Video Segmentation")
+            with gr.Accordion("Segmentation Options", open=False):
+                # 如果用户不使用 point_prompt，而是直接提供坐标，则使用 x, y
+                with gr.Row():
+                    x_coord = gr.Slider(label="X Coordinate (Point Prompt Ratio)", minimum=0.0, maximum=1.0, value=0.5, step=0.01)
+                    y_coord = gr.Slider(label="Y Coordinate (Point Prompt Ratio)", minimum=0.0, maximum=1.0, value=0.5, step=0.01)
+                        
+
+            fg_gallery = gr.Gallery(height=150, object_fit='contain', label='Foreground Quick List', value=db_examples.fg_samples, columns=5, allow_preview=False)
+            bg_gallery = gr.Gallery(height=450, object_fit='contain', label='Background Quick List', value=db_examples.bg_samples, columns=5, allow_preview=False)
+            
+
+            with gr.Group():
+            #     with gr.Row():
+            #         num_samples = gr.Slider(label="Videos", minimum=1, maximum=12, value=1, step=1)
+            #         seed = gr.Number(label="Seed", value=12345, precision=0)
+                with gr.Row():
+                    video_width = gr.Slider(label="Video Width", minimum=256, maximum=1024, value=512, step=64, visible=False)
+                    video_height = gr.Slider(label="Video Height", minimum=256, maximum=1024, value=512, step=64, visible=False)
+
+            # with gr.Accordion("Advanced options", open=False):
+            #     steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=20, step=1)
+            #     cfg = gr.Slider(label="CFG Scale", minimum=1.0, maximum=32.0, value=7.0, step=0.01)
+            #     highres_scale = gr.Slider(label="Highres Scale", minimum=1.0, maximum=3.0, value=1.5, step=0.01)
+            #     highres_denoise = gr.Slider(label="Highres Denoise", minimum=0.1, maximum=0.9, value=0.5, step=0.01)
+            #     a_prompt = gr.Textbox(label="Added Prompt", value='best quality')
+            #     n_prompt = gr.Textbox(label="Negative Prompt", value='lowres, bad anatomy, bad hands, cropped, worst quality')
+            #     normal_button = gr.Button(value="Compute Normal (4x Slower)")
+
+        with gr.Column():
+            result_video = gr.Video(label='Output Video', height=700, width=700, visible=True)
+
+            prompt = gr.Textbox(label="Prompt")
+            bg_source = gr.Radio(choices=[e.value for e in BGSource],
+                                value=BGSource.UPLOAD.value,
+                                label="Background Source", type='value')
+
+            example_prompts = gr.Dataset(samples=quick_prompts, label='Prompt Quick List', components=[prompt])
+            relight_button = gr.Button(value="Relight")
+            # fg_gallery = gr.Gallery(witdth=400, object_fit='contain', label='Foreground Quick List', value=db_examples.bg_samples, columns=4, allow_preview=False)
+            # fg_gallery = gr.Gallery(
+            #     height=380, 
+            #     object_fit='contain', 
+            #     label='Foreground Quick List', 
+            #     value=db_examples.fg_samples, 
+            #     columns=4, 
+            #     allow_preview=False,
+            #     elem_id="foreground-gallery"  # 👈 添加 elem_id
+            # )
+
+
+    # 输入列表
+    # ips = [input_fg, input_bg, prompt, video_width, video_height, num_samples, seed, steps, a_prompt, n_prompt, cfg, highres_scale, highres_denoise, bg_source]
+    ips = [input_fg, input_bg, prompt]
+
+    # 按钮绑定处理函数
+    # relight_button.click(fn=lambda: None, inputs=[], outputs=[result_video])
+
+    relight_button.click(fn=dummy_process, inputs=ips, outputs=[result_video])
+    
+    # normal_button.click(fn=dummy_process, inputs=ips, outputs=[result_video])
+
+    # 背景库选择
+    def bg_gallery_selected(gal, evt: gr.SelectData):
+        # import pdb; pdb.set_trace()
+        # img_path = gal[evt.index][0]
+        img_path = db_examples.bg_samples[evt.index]
+        video_path = img_path.replace('frames/0000.png', 'cropped_video.mp4')
+        return video_path
+
+    bg_gallery.select(bg_gallery_selected, inputs=bg_gallery, outputs=input_bg)
+
+    def fg_gallery_selected(gal, evt: gr.SelectData):
+        # import pdb; pdb.set_trace()
+        # img_path = gal[evt.index][0]
+        img_path = db_examples.fg_samples[evt.index]
+        video_path = img_path.replace('frames/0000.png', 'cropped_video.mp4')
+        return video_path
+
+    fg_gallery.select(fg_gallery_selected, inputs=fg_gallery, outputs=input_fg)
+
+    input_fg_img = gr.Image(label="Foreground Video", visible=False)
+    input_bg_img = gr.Image(label="Background Video", visible=False)
+    result_video_img = gr.Image(label="Output Video", visible=False)
+
+    v_index = gr.Textbox(label="ID", visible=False)
+    example_prompts.click(lambda x: x[0], inputs=example_prompts, outputs=prompt, show_progress=False, queue=False)
+
+    # 示例
+    # dummy_video_for_outputs = gr.Video(visible=False, label='Result')
+    gr.Examples(
+        # fn=lambda *args: args[-1],
+        fn=process_example,
+        examples=db_examples.background_conditioned_examples,
+        # inputs=[v_index, input_fg_img, input_bg_img, prompt, bg_source, video_width, video_height, result_video_img],
+        inputs=[v_index, input_fg_img, input_bg_img, prompt, bg_source, result_video_img],
+        outputs=[input_fg, input_bg, result_video],
+        run_on_click=True, examples_per_page=1024
+    )
+
+# 启动 Gradio 应用
+# block.launch(server_name='0.0.0.0', server_port=10002, share=True)
+block.launch(share=True)
diff --git a/app_bf.py b/app_bf.py
new file mode 100644
index 0000000000000000000000000000000000000000..0b2a5be48a75bb770edd94cea1a4f2ffaf27e430
--- /dev/null
+++ b/app_bf.py
@@ -0,0 +1,391 @@
+import os
+import gradio as gr
+import numpy as np
+from enum import Enum
+import db_examples
+import cv2
+
+import spaces
+
+from demo_utils1 import *
+
+from misc_utils.train_utils import unit_test_create_model
+from misc_utils.image_utils import save_tensor_to_gif, save_tensor_to_images
+import os
+from PIL import Image
+import torch
+import torchvision
+from torchvision import transforms
+from einops import rearrange
+import imageio
+import time
+
+from torchvision.transforms import functional as F
+from torch.hub import download_url_to_file
+
+import os
+
+# 推理设置
+from pl_trainer.inference.inference import InferenceIP2PVideo
+from tqdm import tqdm
+
+
+# if not os.path.exists(filename):
+#     original_path = os.getcwd()
+#     base_path = './models'
+#     os.makedirs(base_path, exist_ok=True)
+
+#     # 直接在代码中写入 Token（注意安全风险）
+#     GIT_TOKEN = "955b8ea91095840b76fe38b90a088c200d4c813c"
+#     repo_url = f"https://YeFang:{GIT_TOKEN}@code.openxlab.org.cn/YeFang/RIV_models.git"
+
+#     try:
+#         if os.system(f'git clone {repo_url} {base_path}') != 0:
+#             raise RuntimeError("Git 克隆失败")
+#         os.chdir(base_path)
+#         if os.system('git lfs pull') != 0:
+#             raise RuntimeError("Git LFS 拉取失败")
+#     finally:
+#         os.chdir(original_path)
+
+def tensor_to_pil_image(x):
+    """
+    将 4D PyTorch 张量转换为 PIL 图像。
+    """
+    x = x.float()  # 确保张量类型为 float
+    grid_img = torchvision.utils.make_grid(x, nrow=4).permute(1, 2, 0).detach().cpu().numpy()
+    grid_img = (grid_img * 255).clip(0, 255).astype("uint8")  # 将 [0, 1] 范围转换为 [0, 255]
+    return Image.fromarray(grid_img)
+
+def frame_to_batch(x):
+    """
+    将帧维度转换为批次维度。
+    """
+    return rearrange(x, 'b f c h w -> (b f) c h w')
+
+def clip_image(x, min=0., max=1.):
+    """
+    将图像张量裁剪到指定的最小和最大值。
+    """
+    return torch.clamp(x, min=min, max=max)
+
+def unnormalize(x):
+    """
+    将张量范围从 [-1, 1] 转换到 [0, 1]。
+    """
+    return (x + 1) / 2
+
+
+# 读取图像文件
+def read_images_from_directory(directory, num_frames=16):
+    images = []
+    for i in range(num_frames):
+        img_path = os.path.join(directory, f'{i:04d}.png')
+        img = imageio.imread(img_path)
+        images.append(torch.tensor(img).permute(2, 0, 1))  # Convert to Tensor (C, H, W)
+    return images
+
+def load_and_process_images(folder_path):
+    """
+    读取文件夹中的所有图片，将它们转换为 [-1, 1] 范围的张量并返回一个 4D 张量。
+    """
+    processed_images = []
+    transform = transforms.Compose([
+        transforms.ToTensor(),
+        transforms.Lambda(lambda x: x * 2 - 1)  # 将 [0, 1] 转换为 [-1, 1]
+    ])
+    for filename in sorted(os.listdir(folder_path)):
+        if filename.endswith(".png"):
+            img_path = os.path.join(folder_path, filename)
+            image = Image.open(img_path).convert("RGB")
+            processed_image = transform(image)
+            processed_images.append(processed_image)
+    return torch.stack(processed_images)  # 返回 4D 张量
+
+def load_and_process_video(video_path, num_frames=16, crop_size=512):
+    """
+    读取视频文件中的前 num_frames 帧，将每一帧转换为 [-1, 1] 范围的张量，
+    并进行中心裁剪至 crop_size x crop_size，返回一个 4D 张量。
+    """
+    processed_frames = []
+    transform = transforms.Compose([
+        transforms.CenterCrop(crop_size),       # 中心裁剪
+        transforms.ToTensor(),
+        transforms.Lambda(lambda x: x * 2 - 1)  # 将 [0, 1] 转换为 [-1, 1]
+    ])
+    
+    # 使用 OpenCV 读取视频
+    cap = cv2.VideoCapture(video_path)
+    
+    if not cap.isOpened():
+        raise ValueError(f"无法打开视频文件: {video_path}")
+
+    frame_count = 0
+
+    while frame_count < num_frames:
+        ret, frame = cap.read()
+        if not ret:
+            break  # 视频帧读取完毕或视频帧不足
+        
+        # 转换为 RGB 格式
+        frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+        image = Image.fromarray(frame)
+        
+        # 应用转换
+        processed_frame = transform(image)
+        processed_frames.append(processed_frame)
+        
+        frame_count += 1
+
+    cap.release()  # 释放视频资源
+
+    if len(processed_frames) < num_frames:
+        raise ValueError(f"视频帧不足 {num_frames} 帧，仅找到 {len(processed_frames)} 帧。")
+
+    return torch.stack(processed_frames)  # 返回 4D 张量 (帧数, 通道数, 高度, 宽度)
+
+
+def clear_cache(output_path):
+    if os.path.exists(output_path):
+        os.remove(output_path)
+    return None
+
+
+#! 加载模型
+# 配置路径和加载模型
+config_path = 'configs/instruct_v2v_ic_gradio.yaml'
+diffusion_model = unit_test_create_model(config_path)
+diffusion_model = diffusion_model.to('cuda')
+
+# 加载模型检查点
+# ckpt_path = 'models/relvid_mm_sd15_fbc_unet.pth' #! change
+# ckpt_path = 'tmp/pytorch_model.bin'
+# 下载文件
+
+os.makedirs('models', exist_ok=True)
+model_path = "models/relvid_mm_sd15_fbc_unet.pth"
+
+if not os.path.exists(model_path):
+    download_url_to_file(url='https://huggingface.co/aleafy/RelightVid/resolve/main/relvid_mm_sd15_fbc_unet.pth', dst=model_path)
+
+
+ckpt = torch.load(model_path, map_location='cpu')
+diffusion_model.load_state_dict(ckpt, strict=False)
+
+
+# import pdb; pdb.set_trace()
+
+# 更改全局临时目录
+new_tmp_dir = "./demo/gradio_bg"
+os.makedirs(new_tmp_dir, exist_ok=True)
+
+# import pdb; pdb.set_trace()
+
+def save_video_from_frames(image_pred, save_pth, fps=8):
+    """
+    将 image_pred 中的帧保存为视频文件。
+
+    参数:
+    - image_pred: Tensor，形状为 (1, 16, 3, 512, 512)
+    - save_pth: 保存视频的路径，例如 "output_video.mp4"
+    - fps: 视频的帧率
+    """
+    # 视频参数
+    num_frames = image_pred.shape[1]
+    frame_height, frame_width = 512, 512  # 目标尺寸
+    fourcc = cv2.VideoWriter_fourcc(*'mp4v')  # 使用 mp4 编码格式
+
+    # 创建 VideoWriter 对象
+    out = cv2.VideoWriter(save_pth, fourcc, fps, (frame_width, frame_height))
+
+    for i in range(num_frames):
+        # 反归一化 + 转换为 0-255 范围
+        pred_frame = clip_image(unnormalize(image_pred[0][i].unsqueeze(0))) * 255
+        pred_frame_resized = pred_frame.squeeze(0).detach().cpu()  # (3, 512, 512)
+        pred_frame_resized = pred_frame_resized.permute(1, 2, 0).numpy().astype("uint8")  # (512, 512, 3)
+
+        # Resize 到 256x256
+        pred_frame_resized = cv2.resize(pred_frame_resized, (frame_width, frame_height))
+
+        # 将 RGB 转为 BGR（因为 OpenCV 使用 BGR 格式）
+        pred_frame_bgr = cv2.cvtColor(pred_frame_resized, cv2.COLOR_RGB2BGR)
+
+        # 写入帧到视频
+        out.write(pred_frame_bgr)
+
+    # 释放 VideoWriter 资源
+    out.release()
+    print(f"视频已保存至 {save_pth}")
+
+
+inf_pipe = InferenceIP2PVideo(
+        diffusion_model.unet, 
+        scheduler='ddpm',
+        num_ddim_steps=20
+    )
+
+# 伪函数占位（生成空白视频）
+@spaces.GPU
+def dummy_process(input_fg, input_bg):
+    # import pdb; pdb.set_trace()
+
+    diffusion_model.to(torch.float16)
+    fg_tensor = load_and_process_video(input_fg).cuda().unsqueeze(0).to(dtype=torch.float16)
+    bg_tensor = load_and_process_video(input_bg).cuda().unsqueeze(0).to(dtype=torch.float16) # (1, 16, 4, 64, 64)
+
+    cond_fg_tensor = diffusion_model.encode_image_to_latent(fg_tensor)  # (1, 16, 4, 64, 64)
+    cond_bg_tensor = diffusion_model.encode_image_to_latent(bg_tensor)
+    cond_tensor = torch.cat((cond_fg_tensor, cond_bg_tensor), dim=2)
+
+    # 初始化潜变量
+    init_latent = torch.randn_like(cond_fg_tensor)    
+
+    EDIT_PROMPT = 'change the background'
+    VIDEO_CFG = 1.2
+    TEXT_CFG = 7.5
+    text_cond = diffusion_model.encode_text([EDIT_PROMPT])  # (1, 77, 768)
+    text_uncond = diffusion_model.encode_text([''])
+    # to float16
+    print('------------to float 16----------------')
+    init_latent, text_cond, text_uncond, cond_tensor = (
+        init_latent.to(dtype=torch.float16),
+        text_cond.to(dtype=torch.float16),
+        text_uncond.to(dtype=torch.float16),
+        cond_tensor.to(dtype=torch.float16)
+    )
+    inf_pipe.unet.to(torch.float16)
+    latent_pred = inf_pipe(
+        latent=init_latent,
+        text_cond=text_cond,
+        text_uncond=text_uncond,
+        img_cond=cond_tensor,
+        text_cfg=TEXT_CFG,
+        img_cfg=VIDEO_CFG,
+    )['latent']
+    
+
+    image_pred = diffusion_model.decode_latent_to_image(latent_pred) # (1,16,3,512,512)
+    output_path = os.path.join(new_tmp_dir, f"output_{int(time.time())}.mp4")
+    # clear_cache(output_path)
+    
+    save_video_from_frames(image_pred, output_path)
+    # import pdb; pdb.set_trace()
+    # fps = 8
+    # frames = []
+    # for i in range(16):
+    #     pred_frame = clip_image(unnormalize(image_pred[0][i].unsqueeze(0))) * 255
+    #     pred_frame_resized = pred_frame.squeeze(0).detach().cpu() #(3,512,512)
+    #     pred_frame_resized = pred_frame_resized.permute(1, 2, 0).detach().cpu().numpy().astype("uint8") #(512,512,3) np
+    #     Image.fromarray(pred_frame_resized).save(save_pth)
+
+    # # 生成一个简单的黑色视频作为示例
+    # output_path = os.path.join(new_tmp_dir, "output.mp4")
+    # fourcc = cv2.VideoWriter_fourcc(*'mp4v')
+    # out = cv2.VideoWriter(output_path, fourcc, 20.0, (512, 512))
+
+    # for _ in range(60):  # 生成 3 秒的视频（20fps）
+    #     frame = np.zeros((512, 512, 3), dtype=np.uint8)
+    #     out.write(frame)
+    # out.release()
+    torch.cuda.empty_cache()
+
+    return output_path
+
+# 枚举类用于背景选择
+class BGSource(Enum):
+    UPLOAD = "Use Background Video"
+    UPLOAD_FLIP = "Use Flipped Background Video"
+    LEFT = "Left Light"
+    RIGHT = "Right Light"
+    TOP = "Top Light"
+    BOTTOM = "Bottom Light"
+    GREY = "Ambient"
+
+# Quick prompts 示例
+quick_prompts = [
+    'beautiful woman',
+    'handsome man',
+    'beautiful woman, cinematic lighting',
+    'handsome man, cinematic lighting',
+    'beautiful woman, natural lighting',
+    'handsome man, natural lighting',
+    'beautiful woman, neo punk lighting, cyberpunk',
+    'handsome man, neo punk lighting, cyberpunk',
+]
+quick_prompts = [[x] for x in quick_prompts]
+
+# Gradio UI 结构
+block = gr.Blocks().queue()
+with block:
+    with gr.Row():
+        gr.Markdown("## IC-Light (Relighting with Foreground and Background Video Condition)")
+
+    with gr.Row():
+        with gr.Column():
+            with gr.Row():
+                input_fg = gr.Video(label="Foreground Video", height=370, width=370, visible=True)
+                input_bg = gr.Video(label="Background Video", height=370, width=370, visible=True)
+
+            prompt = gr.Textbox(label="Prompt")
+            bg_source = gr.Radio(choices=[e.value for e in BGSource],
+                                 value=BGSource.UPLOAD.value,
+                                 label="Background Source", type='value')
+
+            example_prompts = gr.Dataset(samples=quick_prompts, label='Prompt Quick List', components=[prompt])
+            bg_gallery = gr.Gallery(height=450, object_fit='contain', label='Background Quick List', value=db_examples.bg_samples, columns=5, allow_preview=False)
+            relight_button = gr.Button(value="Relight")
+
+            with gr.Group():
+                with gr.Row():
+                    num_samples = gr.Slider(label="Videos", minimum=1, maximum=12, value=1, step=1)
+                    seed = gr.Number(label="Seed", value=12345, precision=0)
+                with gr.Row():
+                    video_width = gr.Slider(label="Video Width", minimum=256, maximum=1024, value=512, step=64)
+                    video_height = gr.Slider(label="Video Height", minimum=256, maximum=1024, value=640, step=64)
+
+            with gr.Accordion("Advanced options", open=False):
+                steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=20, step=1)
+                cfg = gr.Slider(label="CFG Scale", minimum=1.0, maximum=32.0, value=7.0, step=0.01)
+                highres_scale = gr.Slider(label="Highres Scale", minimum=1.0, maximum=3.0, value=1.5, step=0.01)
+                highres_denoise = gr.Slider(label="Highres Denoise", minimum=0.1, maximum=0.9, value=0.5, step=0.01)
+                a_prompt = gr.Textbox(label="Added Prompt", value='best quality')
+                n_prompt = gr.Textbox(label="Negative Prompt", value='lowres, bad anatomy, bad hands, cropped, worst quality')
+                normal_button = gr.Button(value="Compute Normal (4x Slower)")
+
+        with gr.Column():
+            result_video = gr.Video(label='Output Video', height=600, width=600, visible=True)
+
+    # 输入列表
+    # ips = [input_fg, input_bg, prompt, video_width, video_height, num_samples, seed, steps, a_prompt, n_prompt, cfg, highres_scale, highres_denoise, bg_source]
+    ips = [input_fg, input_bg]
+
+    # 按钮绑定处理函数
+    # relight_button.click(fn=lambda: None, inputs=[], outputs=[result_video])
+
+    relight_button.click(fn=dummy_process, inputs=ips, outputs=[result_video])
+    
+    normal_button.click(fn=dummy_process, inputs=ips, outputs=[result_video])
+
+    # 背景库选择
+    def bg_gallery_selected(gal, evt: gr.SelectData):
+        # import pdb; pdb.set_trace()
+        # img_path = gal[evt.index][0]
+        img_path = db_examples.bg_samples[evt.index]
+        video_path = img_path.replace('frames/0000.png', 'cropped_video.mp4')
+        return video_path
+
+    bg_gallery.select(bg_gallery_selected, inputs=bg_gallery, outputs=input_bg)
+
+    # 示例
+    # dummy_video_for_outputs = gr.Video(visible=False, label='Result')
+    gr.Examples(
+        fn=lambda *args: args[-1],
+        examples=db_examples.background_conditioned_examples,
+        inputs=[input_fg, input_bg, prompt, bg_source, video_width, video_height, seed, result_video],
+        outputs=[result_video],
+        run_on_click=True, examples_per_page=1024
+    )
+
+# 启动 Gradio 应用
+# block.launch(server_name='0.0.0.0', server_port=10002, share=True)
+block.launch()
diff --git a/db_examples_bf.py b/db_examples_bf.py
new file mode 100644
index 0000000000000000000000000000000000000000..9db6869b5be93abb611cc2f22c680227bd68a81c
--- /dev/null
+++ b/db_examples_bf.py
@@ -0,0 +1,260 @@
+
+bg_samples = [
+    'demo/clean_bg_extracted/22/frames/0000.png',
+    'demo/clean_bg_extracted/23/frames/0000.png',
+    'demo/clean_bg_extracted/27/frames/0000.png',
+    'demo/clean_bg_extracted/33/frames/0000.png',
+    'demo/clean_bg_extracted/47/frames/0000.png',
+    'demo/clean_bg_extracted/39/frames/0000.png',
+    'demo/clean_bg_extracted/59/frames/0000.png',
+    'demo/clean_bg_extracted/55/frames/0000.png',
+    'demo/clean_bg_extracted/58/frames/0000.png',
+    'demo/clean_bg_extracted/57/frames/0000.png', #42
+    'demo/clean_bg_extracted/8/frames/0000.png',
+    'demo/clean_bg_extracted/9/frames/0000.png',
+    'demo/clean_bg_extracted/10/frames/0000.png',
+    'demo/clean_bg_extracted/14/frames/0000.png',
+    'demo/clean_bg_extracted/62/frames/0000.png'
+] # 准备大概 15 个 background视频
+
+fg_samples = [
+    'demo/clean_fg_extracted/14/frames/0000.png',
+    'demo/clean_fg_extracted/15/frames/0000.png',
+    'demo/clean_fg_extracted/18/frames/0000.png',
+    'demo/clean_fg_extracted/9/frames/0000.png',
+    'demo/clean_fg_extracted/22/frames/0000.png',
+    # 'demo/clean_bg_extracted/39/frames/0000.png',
+    # 'demo/clean_bg_extracted/59/frames/0000.png',
+    # 'demo/clean_bg_extracted/55/frames/0000.png',
+    # 'demo/clean_bg_extracted/58/frames/0000.png',
+    # 'demo/clean_bg_extracted/57/frames/0000.png', #42
+    # 'demo/clean_bg_extracted/8/frames/0000.png',
+    # 'demo/clean_bg_extracted/9/frames/0000.png',
+    # 'demo/clean_bg_extracted/10/frames/0000.png',
+    # 'demo/clean_bg_extracted/14/frames/0000.png',
+    # 'demo/clean_bg_extracted/62/frames/0000.png'
+] # 准备大概 15 个 background视频
+
+
+background_conditioned_examples = [
+    [
+        "demo/clean_fg_extracted/14/cropped_video.mp4",
+        "demo/clean_bg_extracted/22/cropped_video.mp4",
+        "beautiful woman, cinematic lighting",
+        "Use Background Video",
+        512,
+        512,
+        "static_fg_sync_bg_visualization_fy/14_22_100fps.mp4",
+    ],
+    [
+        "demo/clean_fg_extracted/14/cropped_video.mp4",
+        "demo/clean_bg_extracted/55/cropped_video.mp4",
+        "beautiful woman, cinematic lighting",
+        "Use Background Video",
+        512,
+        512,
+        "static_fg_sync_bg_visualization_fy/14_55_100fps.mp4",
+    ],
+    [
+        "demo/clean_fg_extracted/15/cropped_video.mp4",
+        "demo/clean_bg_extracted/27/cropped_video.mp4",
+        "beautiful woman, cinematic lighting",
+        "Use Background Video",
+        512,
+        512,
+        
+        "static_fg_sync_bg_visualization_fy/15_27_100fps.mp4",
+    ],
+    [
+        "demo/clean_fg_extracted/18/cropped_video.mp4",
+        "demo/clean_bg_extracted/23/cropped_video.mp4",
+        "beautiful woman, cinematic lighting",
+        "Use Background Video",
+        512,
+        512,
+        
+        "static_fg_sync_bg_visualization_fy/18_23_100fps.mp4",
+    ],
+    # [
+    #     "demo/clean_fg_extracted/18/cropped_video.mp4",
+    #     "demo/clean_bg_extracted/33/cropped_video.mp4",
+    #     "beautiful woman, cinematic lighting",
+    #     "Use Background Video",
+    #     512,
+    #     512,
+    #     
+    #     "static_fg_sync_bg_visualization_fy/18_33_100fps.mp4",
+    # ],
+    [
+        "demo/clean_fg_extracted/22/cropped_video.mp4",
+        "demo/clean_bg_extracted/39/cropped_video.mp4",
+        "beautiful woman, cinematic lighting",
+        "Use Background Video",
+        512,
+        512,
+        
+        "static_fg_sync_bg_visualization_fy/22_39_100fps.mp4",
+    ],
+    # [
+    #     "demo/clean_fg_extracted/22/cropped_video.mp4",
+    #     "demo/clean_bg_extracted/59/cropped_video.mp4",
+    #     "beautiful woman, cinematic lighting",
+    #     "Use Background Video",
+    #     512,
+    #     512,
+    #     
+    #     "static_fg_sync_bg_visualization_fy/22_59_100fps.mp4",
+    # ],
+    [
+        "demo/clean_fg_extracted/9/cropped_video.mp4",
+        "demo/clean_bg_extracted/8/cropped_video.mp4",
+        "beautiful woman, cinematic lighting",
+        "Use Background Video",
+        512,
+        512,
+        
+        "static_fg_sync_bg_visualization_fy/9_8_100fps.mp4",
+    ],
+    [
+        "demo/clean_fg_extracted/9/cropped_video.mp4",
+        "demo/clean_bg_extracted/9/cropped_video.mp4",
+        "beautiful woman, cinematic lighting",
+        "Use Background Video",
+        512,
+        512,
+        
+        "static_fg_sync_bg_visualization_fy/9_9_100fps.mp4",
+    ],
+    [
+        "demo/clean_fg_extracted/9/cropped_video.mp4",
+        "demo/clean_bg_extracted/10/cropped_video.mp4",
+        "beautiful woman, cinematic lighting",
+        "Use Background Video",
+        512,
+        512,
+        
+        "static_fg_sync_bg_visualization_fy/9_10_100fps.mp4",
+    ],
+    # [
+    #     "demo/clean_fg_extracted/9/cropped_video.mp4",
+    #     "demo/clean_bg_extracted/14/cropped_video.mp4",
+    #     "beautiful woman, cinematic lighting",
+    #     "Use Background Video",
+    #     512,
+    #     512,
+    #     
+    #     "static_fg_sync_bg_visualization_fy/9_14_100fps.mp4",
+    # ],
+
+]
+# background_conditioned_examples = [
+#     [
+#         "demo/clean_fg_extracted/14/cropped_video.mp4",
+#         "demo/clean_bg_extracted/22/cropped_video.mp4",
+#         "beautiful woman, cinematic lighting",
+#         "Use Background Video",
+#         512,
+#         512,
+#         "static_fg_sync_bg_visualization_fy/14_22_100fps.mp4",
+#     ],
+#     [
+#         "demo/clean_fg_extracted/14/cropped_video.mp4",
+#         "demo/clean_bg_extracted/55/cropped_video.mp4",
+#         "beautiful woman, cinematic lighting",
+#         "Use Background Video",
+#         512,
+#         512,
+#         "static_fg_sync_bg_visualization_fy/14_55_100fps.mp4",
+#     ],
+#     [
+#         "demo/clean_fg_extracted/15/cropped_video.mp4",
+#         "demo/clean_bg_extracted/27/cropped_video.mp4",
+#         "beautiful woman, cinematic lighting",
+#         "Use Background Video",
+#         512,
+#         512,
+        
+#         "static_fg_sync_bg_visualization_fy/15_27_100fps.mp4",
+#     ],
+#     [
+#         "demo/clean_fg_extracted/18/cropped_video.mp4",
+#         "demo/clean_bg_extracted/23/cropped_video.mp4",
+#         "beautiful woman, cinematic lighting",
+#         "Use Background Video",
+#         512,
+#         512,
+        
+#         "static_fg_sync_bg_visualization_fy/18_23_100fps.mp4",
+#     ],
+#     # [
+#     #     "demo/clean_fg_extracted/18/cropped_video.mp4",
+#     #     "demo/clean_bg_extracted/33/cropped_video.mp4",
+#     #     "beautiful woman, cinematic lighting",
+#     #     "Use Background Video",
+#     #     512,
+#     #     512,
+#     #     
+#     #     "static_fg_sync_bg_visualization_fy/18_33_100fps.mp4",
+#     # ],
+#     [
+#         "demo/clean_fg_extracted/22/cropped_video.mp4",
+#         "demo/clean_bg_extracted/39/cropped_video.mp4",
+#         "beautiful woman, cinematic lighting",
+#         "Use Background Video",
+#         512,
+#         512,
+        
+#         "static_fg_sync_bg_visualization_fy/22_39_100fps.mp4",
+#     ],
+#     # [
+#     #     "demo/clean_fg_extracted/22/cropped_video.mp4",
+#     #     "demo/clean_bg_extracted/59/cropped_video.mp4",
+#     #     "beautiful woman, cinematic lighting",
+#     #     "Use Background Video",
+#     #     512,
+#     #     512,
+#     #     
+#     #     "static_fg_sync_bg_visualization_fy/22_59_100fps.mp4",
+#     # ],
+#     [
+#         "demo/clean_fg_extracted/9/cropped_video.mp4",
+#         "demo/clean_bg_extracted/8/cropped_video.mp4",
+#         "beautiful woman, cinematic lighting",
+#         "Use Background Video",
+#         512,
+#         512,
+        
+#         "static_fg_sync_bg_visualization_fy/9_8_100fps.mp4",
+#     ],
+#     [
+#         "demo/clean_fg_extracted/9/cropped_video.mp4",
+#         "demo/clean_bg_extracted/9/cropped_video.mp4",
+#         "beautiful woman, cinematic lighting",
+#         "Use Background Video",
+#         512,
+#         512,
+        
+#         "static_fg_sync_bg_visualization_fy/9_9_100fps.mp4",
+#     ],
+#     [
+#         "demo/clean_fg_extracted/9/cropped_video.mp4",
+#         "demo/clean_bg_extracted/10/cropped_video.mp4",
+#         "beautiful woman, cinematic lighting",
+#         "Use Background Video",
+#         512,
+#         512,
+        
+#         "static_fg_sync_bg_visualization_fy/9_10_100fps.mp4",
+#     ],
+#     # [
+#     #     "demo/clean_fg_extracted/9/cropped_video.mp4",
+#     #     "demo/clean_bg_extracted/14/cropped_video.mp4",
+#     #     "beautiful woman, cinematic lighting",
+#     #     "Use Background Video",
+#     #     512,
+#     #     512,
+#     #     
+#     #     "static_fg_sync_bg_visualization_fy/9_14_100fps.mp4",
+#     # ],
+
+# ]
diff --git a/demo/clean_bg_extracted/0/cropped_video.mp4 b/demo/clean_bg_extracted/0/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..fd0532ee297ea58ab37f0a91928272aa58340d72
Binary files /dev/null and b/demo/clean_bg_extracted/0/cropped_video.mp4 differ
diff --git a/demo/clean_bg_extracted/0/frames/0000.png b/demo/clean_bg_extracted/0/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..f0f14cd8f9a1845221e4ba4c534b2dbaea6ba968
Binary files /dev/null and b/demo/clean_bg_extracted/0/frames/0000.png differ
diff --git a/demo/clean_bg_extracted/1/cropped_video.mp4 b/demo/clean_bg_extracted/1/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..701fb967feeeff89b0171ce14ae29348f7a2f2a8
Binary files /dev/null and b/demo/clean_bg_extracted/1/cropped_video.mp4 differ
diff --git a/demo/clean_bg_extracted/1/frames/0000.png b/demo/clean_bg_extracted/1/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..a46b0d2154341361c633f20333298aca2f4d2781
Binary files /dev/null and b/demo/clean_bg_extracted/1/frames/0000.png differ
diff --git a/demo/clean_bg_extracted/2/cropped_video.mp4 b/demo/clean_bg_extracted/2/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..9872390e0e51c31ca8d62853c7f439826bc1de01
Binary files /dev/null and b/demo/clean_bg_extracted/2/cropped_video.mp4 differ
diff --git a/demo/clean_bg_extracted/2/frames/0000.png b/demo/clean_bg_extracted/2/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..8fb6571ff86a438b58ac58d7984b8b0cb181c824
Binary files /dev/null and b/demo/clean_bg_extracted/2/frames/0000.png differ
diff --git a/demo/clean_bg_extracted/47/cropped_video.mp4 b/demo/clean_bg_extracted/47/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..886b25aaff041af358ba5f5b34b8855c9b3502dd
Binary files /dev/null and b/demo/clean_bg_extracted/47/cropped_video.mp4 differ
diff --git a/demo/clean_bg_extracted/57/cropped_video.mp4 b/demo/clean_bg_extracted/57/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..f40aebed3999210df943882de4e8b7eefa37542e
Binary files /dev/null and b/demo/clean_bg_extracted/57/cropped_video.mp4 differ
diff --git a/demo/clean_bg_extracted/58/cropped_video.mp4 b/demo/clean_bg_extracted/58/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..9d62f352a05715c59e9d40e85e30aa6941822660
Binary files /dev/null and b/demo/clean_bg_extracted/58/cropped_video.mp4 differ
diff --git a/demo/clean_bg_extracted/62/cropped_video.mp4 b/demo/clean_bg_extracted/62/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..eb09b6ee2faa87b6d17cd5bf5d53b5741b7b2944
Binary files /dev/null and b/demo/clean_bg_extracted/62/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/0/cropped_video.mp4 b/demo/clean_fg_extracted/0/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..7c763fc944168a62b07d9a22984f585852002b43
Binary files /dev/null and b/demo/clean_fg_extracted/0/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/0/frames/0000.png b/demo/clean_fg_extracted/0/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..bf7b2a36003febcd14a425f963e90e310e879e4a
Binary files /dev/null and b/demo/clean_fg_extracted/0/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/1/cropped_video.mp4 b/demo/clean_fg_extracted/1/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..70404dea498df2383bfc9ac18bfd5ac66f5c4acf
Binary files /dev/null and b/demo/clean_fg_extracted/1/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/1/frames/0000.png b/demo/clean_fg_extracted/1/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..a22fe0662d4414da11f5c84c42398c6df67ff6ff
Binary files /dev/null and b/demo/clean_fg_extracted/1/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/10/cropped_video.mp4 b/demo/clean_fg_extracted/10/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..071a1e6448bee93bc58b259b5ddc16a4f1d62e3c
Binary files /dev/null and b/demo/clean_fg_extracted/10/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/10/frames/0000.png b/demo/clean_fg_extracted/10/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..daf620563b12d130ffe74495aa392f94ffbe5315
Binary files /dev/null and b/demo/clean_fg_extracted/10/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/11/cropped_video.mp4 b/demo/clean_fg_extracted/11/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..fe5ca721c66ffca8e1fcac57bf6e0eca9dda35e3
Binary files /dev/null and b/demo/clean_fg_extracted/11/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/11/frames/0000.png b/demo/clean_fg_extracted/11/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..89771cd8588b1ebbe0ebe299797ad94ce4a5dbce
Binary files /dev/null and b/demo/clean_fg_extracted/11/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/12/cropped_video.mp4 b/demo/clean_fg_extracted/12/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..c0ea971ad53ea2bdd28fb704b413bea655f7c1e0
Binary files /dev/null and b/demo/clean_fg_extracted/12/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/12/frames/0000.png b/demo/clean_fg_extracted/12/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..8f3b9cac63f695596981e841034c0f725d0a68d8
Binary files /dev/null and b/demo/clean_fg_extracted/12/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/13/cropped_video.mp4 b/demo/clean_fg_extracted/13/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..a9f622da2f1afdad539916e7c813bed27203d380
Binary files /dev/null and b/demo/clean_fg_extracted/13/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/13/frames/0000.png b/demo/clean_fg_extracted/13/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..20d4b00d9ec000c1a1b6b480a637d61bd8d6b9d2
Binary files /dev/null and b/demo/clean_fg_extracted/13/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/14/frames/0000.png b/demo/clean_fg_extracted/14/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..9978d6a98c821942d01a8c63f6d990f3fa675a08
Binary files /dev/null and b/demo/clean_fg_extracted/14/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/14/frames/0000_rmbg.png b/demo/clean_fg_extracted/14/frames/0000_rmbg.png
new file mode 100644
index 0000000000000000000000000000000000000000..4fb2964f195b999ff3f2b16eaa888eff7a35958d
Binary files /dev/null and b/demo/clean_fg_extracted/14/frames/0000_rmbg.png differ
diff --git a/demo/clean_fg_extracted/15/frames/0000.png b/demo/clean_fg_extracted/15/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..6210c2e112da409ffcc49d8304d6152fc1b9bfaf
Binary files /dev/null and b/demo/clean_fg_extracted/15/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/15/frames/0000_rmbg.png b/demo/clean_fg_extracted/15/frames/0000_rmbg.png
new file mode 100644
index 0000000000000000000000000000000000000000..9fc594f5fd536221bbafbfa7850780367a0ee8a4
Binary files /dev/null and b/demo/clean_fg_extracted/15/frames/0000_rmbg.png differ
diff --git a/demo/clean_fg_extracted/16/cropped_video.mp4 b/demo/clean_fg_extracted/16/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..e6a23da6cb4dd4289e6992dfa775dedbaba74877
Binary files /dev/null and b/demo/clean_fg_extracted/16/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/16/frames/0000.png b/demo/clean_fg_extracted/16/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..73e9ee0be09db05174892bd0537c4711d46948ca
Binary files /dev/null and b/demo/clean_fg_extracted/16/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/17/cropped_video.mp4 b/demo/clean_fg_extracted/17/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..3572ee769b6dd8ff18838ef738167736b7b06569
Binary files /dev/null and b/demo/clean_fg_extracted/17/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/17/frames/0000.png b/demo/clean_fg_extracted/17/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..982fdc23ebdb44efc57f9b61c0cf56d6dbb87c4b
Binary files /dev/null and b/demo/clean_fg_extracted/17/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/18/frames/0000.png b/demo/clean_fg_extracted/18/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..506c30eb9c08eedaa96c08202c8184a60bce82fb
Binary files /dev/null and b/demo/clean_fg_extracted/18/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/18/frames/0000_rmbg.png b/demo/clean_fg_extracted/18/frames/0000_rmbg.png
new file mode 100644
index 0000000000000000000000000000000000000000..ed95c82ca3d72202feaa87f3fd2927b6669b9416
Binary files /dev/null and b/demo/clean_fg_extracted/18/frames/0000_rmbg.png differ
diff --git a/demo/clean_fg_extracted/2/cropped_video.mp4 b/demo/clean_fg_extracted/2/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..79e3047a91002d6c944b93f0303604802ffcd6d4
Binary files /dev/null and b/demo/clean_fg_extracted/2/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/2/frames/0000.png b/demo/clean_fg_extracted/2/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..e512d320e03f7c4c68c67b6a2aa5ffdd0bf95002
Binary files /dev/null and b/demo/clean_fg_extracted/2/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/22/frames/0000.png b/demo/clean_fg_extracted/22/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..40b625a6ffc17ddea92cae39b18c8a28e87b7942
Binary files /dev/null and b/demo/clean_fg_extracted/22/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/22/frames/0000_rmbg.png b/demo/clean_fg_extracted/22/frames/0000_rmbg.png
new file mode 100644
index 0000000000000000000000000000000000000000..14c16e2caa41609215e1d524a459b9cddde2d16f
Binary files /dev/null and b/demo/clean_fg_extracted/22/frames/0000_rmbg.png differ
diff --git a/demo/clean_fg_extracted/3/cropped_video.mp4 b/demo/clean_fg_extracted/3/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..765b420086c337769018a06cc94dda9bd37096f7
Binary files /dev/null and b/demo/clean_fg_extracted/3/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/3/frames/0000.png b/demo/clean_fg_extracted/3/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..af2d7d6e317b4239ad9dbe47d11ee5e85b2735cc
Binary files /dev/null and b/demo/clean_fg_extracted/3/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/4/cropped_video.mp4 b/demo/clean_fg_extracted/4/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..9e8375b3410af711ea595a99431017c33056d19c
Binary files /dev/null and b/demo/clean_fg_extracted/4/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/4/frames/0000.png b/demo/clean_fg_extracted/4/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..8b7435c6aa2a2831b0de1267a02275fe31daae64
Binary files /dev/null and b/demo/clean_fg_extracted/4/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/5/cropped_video.mp4 b/demo/clean_fg_extracted/5/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..f30f6a5e9ad60002aa933ae57ca90059886c7cbe
Binary files /dev/null and b/demo/clean_fg_extracted/5/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/5/frames/0000.png b/demo/clean_fg_extracted/5/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..7dc55de892ce49133a5e519c1fc8686124758143
Binary files /dev/null and b/demo/clean_fg_extracted/5/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/6/3.mp4 b/demo/clean_fg_extracted/6/3.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..037d6180e54a6133798ed077b88103ad483e276e
Binary files /dev/null and b/demo/clean_fg_extracted/6/3.mp4 differ
diff --git a/demo/clean_fg_extracted/6/frames/0000.png b/demo/clean_fg_extracted/6/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..00407245453ad919c14575ec13bde0aa2c1adc8a
Binary files /dev/null and b/demo/clean_fg_extracted/6/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/7/cropped_video.mp4 b/demo/clean_fg_extracted/7/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..8a040b0e9f29f26ca8700a1002d5465d6f8f1a13
Binary files /dev/null and b/demo/clean_fg_extracted/7/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/7/frames/0000.png b/demo/clean_fg_extracted/7/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..708fabf5615c0b750c00e75b873d2b7fda1f734d
Binary files /dev/null and b/demo/clean_fg_extracted/7/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/8/cropped_video.mp4 b/demo/clean_fg_extracted/8/cropped_video.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..07fd7cdf8c57bc630f943553d032ebd4347a51b9
Binary files /dev/null and b/demo/clean_fg_extracted/8/cropped_video.mp4 differ
diff --git a/demo/clean_fg_extracted/8/frames/0000.png b/demo/clean_fg_extracted/8/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..b95f4e000a8465a5449f7249046afe74673a607f
Binary files /dev/null and b/demo/clean_fg_extracted/8/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/9/frames/0000.png b/demo/clean_fg_extracted/9/frames/0000.png
new file mode 100644
index 0000000000000000000000000000000000000000..6fe18ace7b4d828b4293372d4ef2f5d4477ba249
Binary files /dev/null and b/demo/clean_fg_extracted/9/frames/0000.png differ
diff --git a/demo/clean_fg_extracted/9/frames/0000_rmbg.png b/demo/clean_fg_extracted/9/frames/0000_rmbg.png
new file mode 100644
index 0000000000000000000000000000000000000000..b1ef8f019f2b615dd459f9df3acc9deb5e319d9c
Binary files /dev/null and b/demo/clean_fg_extracted/9/frames/0000_rmbg.png differ
diff --git a/static_fg_sync_bg_visualization_fy/14_22_100fps.png b/static_fg_sync_bg_visualization_fy/14_22_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..50d20596e80439ee9fa6ada6035f59dd96e92343
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/14_22_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/14_55_100fps.png b/static_fg_sync_bg_visualization_fy/14_55_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..222e8a14f171aa2f443967ccc939656521b5161a
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/14_55_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/15_27_100fps.png b/static_fg_sync_bg_visualization_fy/15_27_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..85e37635185d3dd5bb5e7570e5909552fc9a4424
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/15_27_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/16_1_100fps.mp4 b/static_fg_sync_bg_visualization_fy/16_1_100fps.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..ce96396689e0d39670ad8702775a1d147710c5d8
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/16_1_100fps.mp4 differ
diff --git a/static_fg_sync_bg_visualization_fy/16_1_100fps.png b/static_fg_sync_bg_visualization_fy/16_1_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..8a5f92db22b9a6fb01bebf58e69d8240634e8975
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/16_1_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/17_0_100fps.mp4 b/static_fg_sync_bg_visualization_fy/17_0_100fps.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..64ede1a2c8c5898c8033731bff44009ed0a9c44a
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/17_0_100fps.mp4 differ
diff --git a/static_fg_sync_bg_visualization_fy/17_0_100fps.png b/static_fg_sync_bg_visualization_fy/17_0_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..8495e71327a7f1e5b976282ee64f1bc33dbb6e64
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/17_0_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/18_23_100fps.png b/static_fg_sync_bg_visualization_fy/18_23_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..f7d98f9bb5f2a9703236b7d267458491b9b855d2
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/18_23_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/18_33_100fps.png b/static_fg_sync_bg_visualization_fy/18_33_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..4031f95ece8207590f2f2dedd0a06900f5a2a8a0
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/18_33_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/22_39_100fps.png b/static_fg_sync_bg_visualization_fy/22_39_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..b4ffb9e4f7141905f671be755e0042bd07964c97
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/22_39_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/22_59_100fps.png b/static_fg_sync_bg_visualization_fy/22_59_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..9603ee290c0990fb7da873d55ef8ed03e5ce82de
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/22_59_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/9_10_100fps.png b/static_fg_sync_bg_visualization_fy/9_10_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..d93f4dce898508b1812086ca6117bc2192387bd3
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/9_10_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/9_14_100fps.png b/static_fg_sync_bg_visualization_fy/9_14_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..332bdf7260dfc2a14793ad4ff9b27f14b9d33e26
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/9_14_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/9_8_100fps.png b/static_fg_sync_bg_visualization_fy/9_8_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..71fd2098f2b6ce723180f63b391b0542b5f61171
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/9_8_100fps.png differ
diff --git a/static_fg_sync_bg_visualization_fy/9_9_100fps.png b/static_fg_sync_bg_visualization_fy/9_9_100fps.png
new file mode 100644
index 0000000000000000000000000000000000000000..fcf3b6b98adbcc7cba1c128e7ef7b380f805d999
Binary files /dev/null and b/static_fg_sync_bg_visualization_fy/9_9_100fps.png differ