Update app.py

app.py

@@ -4,7 +4,7 @@ import gradio as gr
 from gradio_toggle import Toggle
 import torch
 from huggingface_hub import snapshot_download
-from transformers import CLIPProcessor, CLIPModel
+from transformers import CLIPProcessor, CLIPModel, pipeline
 import random
 from xora.models.autoencoders.causal_video_autoencoder import CausalVideoAutoencoder
 from xora.models.transformers.transformer3d import Transformer3DModel
@@ -26,15 +26,15 @@ import csv
 from datetime import datetime
 from openai import OpenAI
 
+# 한글-영어 번역기 초기화
+translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en")
+
 torch.backends.cuda.matmul.allow_tf32 = False
 torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction = False
 torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = False
 torch.backends.cudnn.allow_tf32 = False
 torch.backends.cudnn.deterministic = False
-#torch.backends.cudnn.benchmark = False
 torch.backends.cuda.preferred_blas_library="cublas"
-#torch.backends.cuda.preferred_linalg_library="cusolver"
-
 torch.set_float32_matmul_precision("highest")
 
 MAX_SEED = np.iinfo(np.int32).max
@@ -45,13 +45,9 @@ openai_api_key = os.getenv("OPENAI_API_KEY")
 client = OpenAI(api_key=openai_api_key)
 
 system_prompt_t2v_path = "assets/system_prompt_t2v.txt"
-system_prompt_i2v_path = "assets/system_prompt_i2v.txt"
 with open(system_prompt_t2v_path, "r") as f:
     system_prompt_t2v = f.read()
 
-with open(system_prompt_i2v_path, "r") as f:
-    system_prompt_i2v = f.read()
-
 # Set model download directory within Hugging Face Spaces
 model_path = "asset"
 
@@ -67,22 +63,20 @@ scheduler_dir = Path(model_path) / "scheduler"
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
-request_log = []  
-
 clip_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32", cache_dir=model_path).to(torch.device("cuda:0"))
 clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32", cache_dir=model_path)
 
-def compute_clip_embedding(text=None, image=None):
-    """
-    Compute CLIP embedding for a given text or image.
-    Args:
-        text (str): Input text prompt.
-        image (PIL.Image): Input image.
-    Returns:
-        list: CLIP embedding as a list of floats.
-    """
-    inputs = clip_processor(text=text, images=image, return_tensors="pt", padding=True).to(device)
-    outputs = clip_model.get_text_features(**inputs) if text else clip_model.get_image_features(**inputs)
+def process_prompt(prompt):
+    # 한글이 포함되어 있는지 확인
+    if any(ord('가') <= ord(char) <= ord('힣') for char in prompt):
+        # 한글을 영어로 번역
+        translated = translator(prompt)[0]['translation_text']
+        return translated
+    return prompt
+
+def compute_clip_embedding(text=None):
+    inputs = clip_processor(text=text, return_tensors="pt", padding=True).to(device)
+    outputs = clip_model.get_text_features(**inputs)
     embedding = outputs.detach().cpu().numpy().flatten().tolist()
     return embedding
 
@@ -110,53 +104,6 @@ def load_scheduler(scheduler_dir):
     scheduler_config = RectifiedFlowScheduler.load_config(scheduler_config_path)
     return RectifiedFlowScheduler.from_config(scheduler_config)
 
-# Helper function for image processing
-def center_crop_and_resize(frame, target_height, target_width):
-    h, w, _ = frame.shape
-    aspect_ratio_target = target_width / target_height
-    aspect_ratio_frame = w / h
-    if aspect_ratio_frame > aspect_ratio_target:
-        new_width = int(h * aspect_ratio_target)
-        x_start = (w - new_width) // 2
-        frame_cropped = frame[:, x_start : x_start + new_width]
-    else:
-        new_height = int(w / aspect_ratio_target)
-        y_start = (h - new_height) // 2
-        frame_cropped = frame[y_start : y_start + new_height, :]
-    frame_resized = cv2.resize(frame_cropped, (target_width, target_height))
-    return frame_resized
-
-def load_image_to_tensor_with_resize(image_path, target_height, target_width):
-    image = Image.open(image_path).convert("RGB")
-    image_np = np.array(image)
-    frame_resized = center_crop_and_resize(image_np, target_height, target_width)
-    frame_tensor = torch.tensor(frame_resized).permute(2, 0, 1).float()
-    frame_tensor = (frame_tensor / 127.5) - 1.0
-    return frame_tensor.unsqueeze(0).unsqueeze(2)
-
-def enhance_prompt_if_enabled(prompt, enhance_toggle, type="t2v"):
-    if not enhance_toggle:
-        print("Enhance toggle is off, Prompt: ", prompt)
-        return prompt
-
-    system_prompt = system_prompt_t2v if type == "t2v" else system_prompt_i2v
-    messages = [
-        {"role": "system", "content": system_prompt},
-        {"role": "user", "content": prompt},
-    ]
-
-    try:
-        response = client.chat.completions.create(
-            model="gpt-4o-mini",
-            messages=messages,
-            max_tokens=200,
-        )
-        print("Enhanced Prompt: ", response.choices[0].message.content.strip())
-        return response.choices[0].message.content.strip()
-    except Exception as e:
-        print(f"Error: {e}")
-        return prompt
-
 # Preset options for resolution and frame configuration
 preset_options = [
     {"label": "1216x704, 41 frames", "width": 1216, "height": 704, "num_frames": 41},
@@ -174,26 +121,8 @@ preset_options = [
     {"label": "768x512, 97 frames", "width": 768, "height": 512, "num_frames": 97},
     {"label": "512x512, 160 frames", "width": 512, "height": 512, "num_frames": 160},
     {"label": "512x512, 200 frames", "width": 512, "height": 512, "num_frames": 200},
-    {"label": "736x480, 113 frames", "width": 736, "height": 480, "num_frames": 113},
-    {"label": "704x480, 121 frames", "width": 704, "height": 480, "num_frames": 121},
-    {"label": "704x448, 129 frames", "width": 704, "height": 448, "num_frames": 129},
-    {"label": "672x448, 137 frames", "width": 672, "height": 448, "num_frames": 137},
-    {"label": "640x416, 153 frames", "width": 640, "height": 416, "num_frames": 153},
-    {"label": "672x384, 161 frames", "width": 672, "height": 384, "num_frames": 161},
-    {"label": "640x384, 169 frames", "width": 640, "height": 384, "num_frames": 169},
-    {"label": "608x384, 177 frames", "width": 608, "height": 384, "num_frames": 177},
-    {"label": "576x384, 185 frames", "width": 576, "height": 384, "num_frames": 185},
-    {"label": "608x352, 193 frames", "width": 608, "height": 352, "num_frames": 193},
-    {"label": "576x352, 201 frames", "width": 576, "height": 352, "num_frames": 201},
-    {"label": "544x352, 209 frames", "width": 544, "height": 352, "num_frames": 209},
-    {"label": "512x352, 225 frames", "width": 512, "height": 352, "num_frames": 225},
-    {"label": "512x352, 233 frames", "width": 512, "height": 352, "num_frames": 233},
-    {"label": "544x320, 241 frames", "width": 544, "height": 320, "num_frames": 241},
-    {"label": "512x320, 249 frames", "width": 512, "height": 320, "num_frames": 249},
-    {"label": "512x320, 257 frames", "width": 512, "height": 320, "num_frames": 257},
 ]
 
-# Function to toggle visibility of sliders based on preset selection
 def preset_changed(preset):
     if preset != "Custom":
         selected = next(item for item in preset_options if item["label"] == preset)
@@ -214,7 +143,7 @@ def preset_changed(preset):
             gr.update(visible=True),
             gr.update(visible=True),
         )
-        
+
 # Load models
 vae = load_vae(vae_dir)
 unet = load_unet(unet_dir)
@@ -232,7 +161,29 @@ pipeline = XoraVideoPipeline(
     vae=vae,
 ).to(torch.device("cuda:0"))
 
-@spaces.GPU(duration=90)  # Dynamic duration
+def enhance_prompt_if_enabled(prompt, enhance_toggle):
+    if not enhance_toggle:
+        print("Enhance toggle is off, Prompt: ", prompt)
+        return prompt
+
+    messages = [
+        {"role": "system", "content": system_prompt_t2v},
+        {"role": "user", "content": prompt},
+    ]
+
+    try:
+        response = client.chat.completions.create(
+            model="gpt-4-mini",
+            messages=messages,
+            max_tokens=200,
+        )
+        print("Enhanced Prompt: ", response.choices[0].message.content.strip())
+        return response.choices[0].message.content.strip()
+    except Exception as e:
+        print(f"Error: {e}")
+        return prompt
+
+@spaces.GPU(duration=90)
 def generate_video_from_text_90(
     prompt="",
     enhance_prompt_toggle=False,
@@ -247,13 +198,17 @@ def generate_video_from_text_90(
     num_frames=60,
     progress=gr.Progress(),
 ):
+    # 프롬프트 전처리 (한글 -> 영어)
+    prompt = process_prompt(prompt)
+    negative_prompt = process_prompt(negative_prompt)
+
     if len(prompt.strip()) < 50:
         raise gr.Error(
             "Prompt must be at least 50 characters long. Please provide more details for the best results.",
             duration=5,
         )
 
-    prompt = enhance_prompt_if_enabled(prompt, enhance_prompt_toggle, type="t2v")
+    prompt = enhance_prompt_if_enabled(prompt, enhance_prompt_toggle)
 
     sample = {
         "prompt": prompt,
@@ -297,7 +252,6 @@ def generate_video_from_text_90(
         gc.collect()
 
     output_path = tempfile.mktemp(suffix=".mp4")
-    print(images.shape)
     video_np = images.squeeze(0).permute(1, 2, 3, 0).cpu().float().numpy()
     video_np = (video_np * 255).astype(np.uint8)
     height, width = video_np.shape[1:3]
@@ -305,101 +259,10 @@ def generate_video_from_text_90(
     for frame in video_np[..., ::-1]:
         out.write(frame)
     out.release()
-    # Explicitly delete tensors and clear cache
     del images
     del video_np
     torch.cuda.empty_cache()
     return output_path
-    
-@spaces.GPU(duration=90)  # Dynamic duration
-def generate_video_from_image_90(
-    image_path,
-    prompt="",
-    enhance_prompt_toggle=False,
-    img2vid_analytics_toggle=True,
-    negative_prompt="",
-    frame_rate=20,
-    seed=random.randint(0, MAX_SEED),
-    num_inference_steps=35,
-    guidance_scale=4.2,
-    height=768,
-    width=768,
-    num_frames=60,
-    progress=gr.Progress(),
-):
-
-    print("Height: ", height)
-    print("Width: ", width)
-    print("Num Frames: ", num_frames)
-
-    if len(prompt.strip()) < 50:
-        raise gr.Error(
-            "Prompt must be at least 50 characters long. Please provide more details for the best results.",
-            duration=5,
-        )
-
-    if not image_path:
-        raise gr.Error("Please provide an input image.", duration=5)
-
-    if img2vid_analytics_toggle:
-        with Image.open(image_path) as img:
-            original_resolution = f"{img.width}x{img.height}"  # Format as "widthxheight"
-            clip_embedding = compute_clip_embedding(image=img)
-
-    media_items = load_image_to_tensor_with_resize(image_path, height, width).to(device).detach()
-
-    prompt = enhance_prompt_if_enabled(prompt, enhance_prompt_toggle, type="i2v")
-
-    sample = {
-        "prompt": prompt,
-        "prompt_attention_mask": None,
-        "negative_prompt": negative_prompt,
-        "negative_prompt_attention_mask": None,
-        "media_items": media_items,
-    }
-
-    generator = torch.Generator(device="cuda").manual_seed(seed)
-
-    def gradio_progress_callback(self, step, timestep, kwargs):
-        progress((step + 1) / num_inference_steps)
-
-    try:
-        with torch.no_grad():
-            images = pipeline(
-                num_inference_steps=num_inference_steps,
-                num_images_per_prompt=1,
-                guidance_scale=guidance_scale,
-                generator=generator,
-                output_type="pt",
-                height=height,
-                width=width,
-                num_frames=num_frames,
-                frame_rate=frame_rate,
-                **sample,
-                is_video=True,
-                vae_per_channel_normalize=True,
-                conditioning_method=ConditioningMethod.FIRST_FRAME,
-                mixed_precision=True,
-                callback_on_step_end=gradio_progress_callback,
-            ).images
-
-        output_path = tempfile.mktemp(suffix=".mp4")
-        video_np = images.squeeze(0).permute(1, 2, 3, 0).cpu().float().numpy()
-        video_np = (video_np * 255).astype(np.uint8)
-        height, width = video_np.shape[1:3]
-        out = cv2.VideoWriter(output_path, cv2.VideoWriter_fourcc(*"mp4v"), frame_rate, (width, height))
-        for frame in video_np[..., ::-1]:
-            out.write(frame)
-        out.release()
-    except Exception as e:
-        raise gr.Error(
-            f"An error occurred while generating the video. Please try again. Error: {e}",
-            duration=5,
-        )
-    finally:
-        torch.cuda.empty_cache()
-        gc.collect()
-    return output_path
 
 def create_advanced_options():
     with gr.Accordion("Step 4: Advanced Options (Optional)", open=False):
@@ -440,6 +303,7 @@ def create_advanced_options():
             width_slider,
             num_frames_slider,
         ]
+
 with gr.Blocks(theme=gr.themes.Soft()) as iface:
     with gr.Row(elem_id="title-row"):
         gr.Markdown(
@@ -450,7 +314,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as iface:
         """
         )
     with gr.Row(elem_id="title-row"):
-        gr.HTML(  # add technical report link
+        gr.HTML(
             """
         <div style="display:flex;column-gap:4px;">
             <a href="https://github.com/Lightricks/LTX-Video">
@@ -471,6 +335,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as iface:
         </div>
         """
         )
+
     with gr.Accordion(" 📖 Tips for Best Results", open=False, elem_id="instructions-accordion"):
         gr.Markdown(
             """
@@ -498,168 +363,60 @@ with gr.Blocks(theme=gr.themes.Soft()) as iface:
         """
         )
 
-    with gr.Tabs():
-        # Text to Video Tab
-        with gr.TabItem("Text to Video"):
-            with gr.Row():
-                with gr.Column():
-                    txt2vid_prompt = gr.Textbox(
-                        label="Step 1: Enter Your Prompt",
-                        placeholder="Describe the video you want to generate (minimum 50 characters)...",
-                        value="A woman with long brown hair and light skin smiles at another woman with long blonde hair. The woman with brown hair wears a black jacket and has a small, barely noticeable mole on her right cheek. The camera angle is a close-up, focused on the woman with brown hair's face. The lighting is warm and natural, likely from the setting sun, casting a soft glow on the scene. The scene appears to be real-life footage.",
-                        lines=5,
-                    )
-                    txt2vid_analytics_toggle = Toggle(
-                        label="I agree to share my usage data anonymously to help improve the model features.",
-                        value=True,
-                        interactive=True,
-                    )
-
-                    txt2vid_enhance_toggle = Toggle(
-                        label="Enhance Prompt",
-                        value=False,
-                        interactive=True,
-                    )
-
-                    txt2vid_negative_prompt = gr.Textbox(
-                        label="Step 2: Enter Negative Prompt",
-                        placeholder="Describe what you don't want in the video...",
-                        value="low quality, worst quality, deformed, distorted, disfigured, motion smear, motion artifacts, fused fingers, bad anatomy, weird hand, ugly",
-                        lines=2,
-                    )
-
-                    txt2vid_preset = gr.Dropdown(
-                        choices=[p["label"] for p in preset_options],
-                        value="768x512, 97 frames",
-                        label="Step 3.1: Choose Resolution Preset",
-                    )
-
-                    txt2vid_frame_rate = gr.Slider(
-                        label="Step 3.2: Frame Rate",
-                        minimum=6,
-                        maximum=60,
-                        step=1,
-                        value=20,
-                    )
-
-                    txt2vid_advanced = create_advanced_options()
-                    txt2vid_generate = gr.Button(
-                        "Step 5: Generate Video",
-                        variant="primary",
-                        size="lg",
-                    )
-
-                with gr.Column():
-                    txt2vid_output = gr.Video(label="Generated Output")
-
-            with gr.Row():
-                gr.Examples(
-                    examples=[
-                        [
-                            "A young woman in a traditional Mongolian dress is peeking through a sheer white curtain, her face showing a mix of curiosity and apprehension. The woman has long black hair styled in two braids, adorned with white beads, and her eyes are wide with a hint of surprise. Her dress is a vibrant blue with intricate gold embroidery, and she wears a matching headband with a similar design. The background is a simple white curtain, which creates a sense of mystery and intrigue.ith long brown hair and light skin smiles at another woman with long blonde hair. The woman with brown hair wears a black jacket and has a small, barely noticeable mole on her right cheek. The camera angle is a close-up, focused on the woman with brown hair’s face. The lighting is warm and natural, likely from the setting sun, casting a soft glow on the scene. The scene appears to be real-life footage",
-                            "low quality, worst quality, deformed, distorted, disfigured, motion smear, motion artifacts, fused fingers, bad anatomy, weird hand, ugly",
-                            "assets/t2v_2.mp4",
-                        ],
-                        [
-                            "A young man with blond hair wearing a yellow jacket stands in a forest and looks around. He has light skin and his hair is styled with a middle part. He looks to the left and then to the right, his gaze lingering in each direction. The camera angle is low, looking up at the man, and remains stationary throughout the video. The background is slightly out of focus, with green trees and the sun shining brightly behind the man. The lighting is natural and warm, with the sun creating a lens flare that moves across the man’s face. The scene is captured in real-life footage.",
-                            "low quality, worst quality, deformed, distorted, disfigured, motion smear, motion artifacts, fused fingers, bad anatomy, weird hand, ugly",
-                            "assets/t2v_1.mp4",
-                        ],
-                        [
-                            "A cyclist races along a winding mountain road. Clad in aerodynamic gear, he pedals intensely, sweat glistening on his brow. The camera alternates between close-ups of his determined expression and wide shots of the breathtaking landscape. Pine trees blur past, and the sky is a crisp blue. The scene is invigorating and competitive.",
-                            "low quality, worst quality, deformed, distorted, disfigured, motion smear, motion artifacts, fused fingers, bad anatomy, weird hand, ugly",
-                            "assets/t2v_0.mp4",
-                        ],
-                    ],
-                    inputs=[txt2vid_prompt, txt2vid_negative_prompt, txt2vid_output],
-                    label="Example Text-to-Video Generations",
-                )
-
-        # Image to Video Tab
-        with gr.TabItem("Image to Video"):
-            with gr.Row():
-                with gr.Column():
-                    img2vid_image = gr.Image(
-                        type="filepath",
-                        label="Step 1: Upload Input Image",
-                        elem_id="image_upload",
-                    )
-                    img2vid_prompt = gr.Textbox(
-                        label="Step 2: Enter Your Prompt",
-                        placeholder="Describe how you want to animate the image (minimum 50 characters)...",
-                        value="A woman with long brown hair and light skin smiles at another woman with long blonde hair. The woman with brown hair wears a black jacket and has a small, barely noticeable mole on her right cheek. The camera angle is a close-up, focused on the woman with brown hair's face. The lighting is warm and natural, likely from the setting sun, casting a soft glow on the scene. The scene appears to be real-life footage.",
-                        lines=5,
-                    )
-                    img2vid_analytics_toggle = Toggle(
-                        label="I agree to share my usage data anonymously to help improve the model features.",
-                        value=True,
-                        interactive=True,
-                    )
-                    img2vid_enhance_toggle = Toggle(
-                        label="Enhance Prompt",
-                        value=False,
-                        interactive=True,
-                    )
-                    img2vid_negative_prompt = gr.Textbox(
-                        label="Step 3: Enter Negative Prompt",
-                        placeholder="Describe what you don't want in the video...",
-                        value="low quality, worst quality, deformed, distorted, disfigured, motion smear, motion artifacts, fused fingers, bad anatomy, weird hand, ugly",
-                        lines=2,
-                    )
-
-                    img2vid_preset = gr.Dropdown(
-                        choices=[p["label"] for p in preset_options],
-                        value="768x512, 97 frames",
-                        label="Step 3.1: Choose Resolution Preset",
-                    )
-
-                    img2vid_frame_rate = gr.Slider(
-                        label="Step 3.2: Frame Rate",
-                        minimum=6,
-                        maximum=60,
-                        step=1,
-                        value=20,
-                    )
-
-                    img2vid_advanced = create_advanced_options()
-                    img2vid_generate = gr.Button("Step 6: Generate Video", variant="primary", size="lg")
-
-                with gr.Column():
-                    img2vid_output = gr.Video(label="Generated Output")
-
-            with gr.Row():
-                gr.Examples(
-                    examples=[
-                        [
-                            "assets/i2v_i2.png",
-                            "A woman stirs a pot of boiling water on a white electric burner. Her hands, with purple nail polish, hold a wooden spoon and move it in a circular motion within a white pot filled with bubbling water. The pot sits on a white electric burner with black buttons and a digital display. The burner is positioned on a white countertop with a red and white checkered cloth partially visible in the bottom right corner. The camera angle is a direct overhead shot, remaining stationary throughout the scene. The lighting is bright and even, illuminating the scene with a neutral white light. The scene is real-life footage.",
-                            "low quality, worst quality, deformed, distorted, disfigured, motion smear, motion artifacts, fused fingers, bad anatomy, weird hand, ugly",
-                            "assets/i2v_2.mp4",
-                        ],
-                        [
-                            "assets/i2v_i0.png",
-                            "A woman in a long, flowing dress stands in a field, her back to the camera, gazing towards the horizon; her hair is long and light, cascading down her back; she stands beneath the sprawling branches of a large oak tree;  to her left, a classic American car is parked on the dry grass; in the distance, a wrecked car lies on its side; the sky above is a dramatic canvas of bright white clouds against a darker sky; the entire image is in black and white, emphasizing the contrast of light and shadow. The woman is walking slowly towards the car.",
-                            "low quality, worst quality, deformed, distorted, disfigured, motion smear, motion artifacts, fused fingers, bad anatomy, weird hand, ugly",
-                            "assets/i2v_0.mp4",
-                        ],
-                        [
-                            "assets/i2v_i1.png",
-                            "A pair of hands shapes a piece of clay on a pottery wheel, gradually forming a cone shape. The hands, belonging to a person out of frame, are covered in clay and gently press a ball of clay onto the center of a spinning pottery wheel. The hands move in a circular motion, gradually forming a cone shape at the top of the clay. The camera is positioned directly above the pottery wheel, providing a bird’s-eye view of the clay being shaped. The lighting is bright and even, illuminating the clay and the hands working on it. The scene is captured in real-life footage.",
-                            "low quality, worst quality, deformed, distorted, disfigured, motion smear, motion artifacts, fused fingers, bad anatomy, weird hand, ugly",
-                            "assets/i2v_1.mp4",
-                        ],
-                    ],
-                    inputs=[
-                        img2vid_image,
-                        img2vid_prompt,
-                        img2vid_negative_prompt,
-                        img2vid_output,
-                    ],
-                    label="Example Image-to-Video Generations",
-                )
-
-    # [Previous event handlers remain the same]
-    txt2vid_preset.change(fn=preset_changed, inputs=[txt2vid_preset], outputs=txt2vid_advanced[3:])
+    with gr.Column():
+        txt2vid_prompt = gr.Textbox(
+            label="Step 1: Enter Your Prompt (한글 또는 영어)",
+            placeholder="생성하고 싶은 비디오를 설명하세요 (최소 50자)...",
+            value="긴 갈색 머리와 밝은 피부를 가진 여성이 긴 금발 머리를 가진 다른 여성을 향해 미소 짓습니다. 갈색 머리 여성은 검은 재킷을 입고 있으며 오른쪽 뺨에 작고 거의 눈에 띄지 않는 점이 있습니다. 카메라 앵글은 갈색 머리 여성의 얼굴에 초점을 맞춘 클로즈업입니다. 조명은 따뜻하고 자연스러우며, 아마도 지는 해에서 나오는 것 같아 장면에 부드러운 빛을 비춥니다.",
+            lines=5,
+        )
+        txt2vid_analytics_toggle = Toggle(
+            label="I agree to share my usage data anonymously to help improve the model features.",
+            value=True,
+            interactive=True,
+        )
+
+        txt2vid_enhance_toggle = Toggle(
+            label="Enhance Prompt",
+            value=False,
+            interactive=True,
+        )
+
+        txt2vid_negative_prompt = gr.Textbox(
+            label="Step 2: Enter Negative Prompt",
+            placeholder="비디오에서 원하지 않는 요소를 설명하세요...",
+            value="낮은 품질, 최악의 품질, 변형된, 왜곡된, 손상된, 모션 번짐, 모션 아티팩트, 융합된 손가락, 잘못된 해부학, 이상한 손, 추한",
+            lines=2,
+        )
+
+        txt2vid_preset = gr.Dropdown(
+            choices=[p["label"] for p in preset_options],
+            value="768x512, 97 frames",
+            label="Step 3.1: Choose Resolution Preset",
+        )
+
+        txt2vid_frame_rate = gr.Slider(
+            label="Step 3.2: Frame Rate",
+            minimum=6,
+            maximum=60,
+            step=1,
+            value=20,
+        )
+
+        txt2vid_advanced = create_advanced_options()
+        txt2vid_generate = gr.Button(
+            "Step 5: Generate Video",
+            variant="primary",
+            size="lg",
+        )
+
+        txt2vid_output = gr.Video(label="Generated Output")
+
+    txt2vid_preset.change(
+        fn=preset_changed,
+        inputs=[txt2vid_preset],
+        outputs=txt2vid_advanced[3:],
+    )
 
     txt2vid_generate.click(
         fn=generate_video_from_text_90,
@@ -677,24 +434,5 @@ with gr.Blocks(theme=gr.themes.Soft()) as iface:
         queue=True,
     )
 
-    img2vid_preset.change(fn=preset_changed, inputs=[img2vid_preset], outputs=img2vid_advanced[3:])
-
-    img2vid_generate.click(
-        fn=generate_video_from_image_90,
-        inputs=[
-            img2vid_image,
-            img2vid_prompt,
-            img2vid_enhance_toggle,
-            img2vid_analytics_toggle,
-            img2vid_negative_prompt,
-            img2vid_frame_rate,
-            *img2vid_advanced,
-        ],
-        outputs=img2vid_output,
-        concurrency_limit=1,
-        concurrency_id="generate_video",
-        queue=True,
-    )
-
 if __name__ == "__main__":
-    iface.queue(max_size=64, default_concurrency_limit=1, api_open=False).launch(share=True, show_api=False)
+    iface.queue(max_size=64, default_concurrency_limit=1, api_open=False).launch(share=True, show_api=False)