import gradio as gr
import torch
import os
import gc
from typing import Optional, Tuple
from model import WanVideoGenerator, get_available_loras, LORA_CACHE_DIR
# Initialize the generator
generator = WanVideoGenerator()
def generate_video(
prompt: str,
negative_prompt: str,
image_input: Optional[dict],
lora_name: str,
lora_scale: float,
height: int,
width: int,
num_frames: int,
guidance_scale: float,
num_inference_steps: int,
fps: int,
seed: int,
progress: gr.Progress = gr.Progress()
) -> Tuple[str, str]:
"""Generate video from text or image prompt with optional LoRA."""
if not prompt.strip():
raise gr.Error("Please provide a prompt!")
progress(0, desc="Initializing...")
# Handle image input for TI2V (Text-Image-to-Video)
image = None
if image_input is not None:
image = image_input
try:
# Load LoRA if specified
if lora_name != "None":
progress(0.1, desc=f"Loading LoRA: {lora_name}...")
generator.load_lora(lora_name, lora_scale)
else:
generator.unload_lora()
progress(0.2, desc="Generating video...")
# Generate video
video_path = generator.generate(
prompt=prompt,
negative_prompt=negative_prompt,
image=image,
height=height,
width=width,
num_frames=num_frames,
guidance_scale=guidance_scale,
num_inference_steps=num_inference_steps,
fps=fps,
seed=seed if seed >= 0 else None,
progress_callback=lambda p: progress(0.2 + 0.7 * p, desc="Generating frames...")
)
progress(1.0, desc="Complete!")
# Get generation info
info = f"""Generation Info:
- Prompt: {prompt}
- LoRA: {lora_name} (scale: {lora_scale})
- Resolution: {width}x{height}
- Frames: {num_frames}
- Steps: {num_inference_steps}
- Guidance: {guidance_scale}
- FPS: {fps}
- Seed: {seed if seed >= 0 else 'Random'}"""
return video_path, info
except Exception as e:
raise gr.Error(f"Generation failed: {str(e)}")
finally:
# Cleanup
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
def refresh_loras():
"""Refresh the list of available LoRAs."""
loras = get_available_loras()
return gr.Dropdown(choices=["None"] + loras, value="None")
# Custom CSS for professional look
custom_css = """
#header-text {
text-align: center;
margin-bottom: 20px;
}
#header-text h1 {
font-size: 2.5em;
font-weight: bold;
background: linear-gradient(90deg, #667eea 0%, #764ba2 100%);
-webkit-background-clip: text;
-webkit-text-fill-color: transparent;
}
#header-text p {
color: #666;
font-size: 1.1em;
}
.lora-section {
background: linear-gradient(135deg, #667eea22 0%, #764ba222 100%);
border-radius: 12px;
padding: 15px;
margin-bottom: 15px;
}
.generation-info {
font-family: monospace;
font-size: 0.9em;
background: #1a1a2e;
color: #eee;
padding: 15px;
border-radius: 8px;
}
"""
# Build the Gradio interface
with gr.Blocks() as demo:
# Header
with gr.Row():
with gr.Column():
gr.HTML("""
""")
with gr.Row():
# Left column - Inputs
with gr.Column(scale=1):
# Prompt section
prompt = gr.Textbox(
label="Prompt",
placeholder="Describe the video you want to generate...",
lines=3,
info="Be descriptive for better results"
)
negative_prompt = gr.Textbox(
label="Negative Prompt",
placeholder="What to avoid in the video...",
lines=2,
value="blur, distortion, low quality, jittery, morphing"
)
# Image input for TI2V
with gr.Accordion("Image Input (TI2V)", open=False):
gr.Markdown("Upload an image for Text-Image-to-Video generation")
image_input = gr.Image(
label="Input Image",
type="pil",
height=300
)
# LoRA section
with gr.Group(elem_classes="lora-section"):
gr.Markdown("### 🚀 LoRA Acceleration")
with gr.Row():
lora_dropdown = gr.Dropdown(
choices=["None"] + get_available_loras(),
value="None",
label="Select LoRA",
info="Choose a LoRA for faster generation"
)
refresh_btn = gr.Button("🔄", size="sm", variant="secondary")
lora_scale = gr.Slider(
minimum=0.0,
maximum=1.0,
value=0.8,
step=0.05,
label="LoRA Scale",
info="Higher values = stronger LoRA effect"
)
gr.Markdown("""
**Available LoRAs:**
- `wan-fast-lora`: Optimized for speed (2-3x faster)
- `wan-quality-lora`: Enhanced quality
- `wan-motion-lora`: Better motion dynamics
""")
# Generation parameters
with gr.Accordion("⚙️ Advanced Settings", open=True):
with gr.Row():
height = gr.Slider(256, 1024, value=480, step=64, label="Height")
width = gr.Slider(256, 1024, value=848, step=64, label="Width")
with gr.Row():
num_frames = gr.Slider(8, 81, value=25, step=1, label="Number of Frames")
fps = gr.Slider(6, 30, value=16, step=1, label="FPS")
with gr.Row():
guidance_scale = gr.Slider(1.0, 20.0, value=5.0, step=0.5, label="Guidance Scale")
num_inference_steps = gr.Slider(4, 50, value=20, step=1, label="Inference Steps")
seed = gr.Slider(-1, 999999, value=-1, step=1, label="Seed (-1 for random)")
# Generate button
generate_btn = gr.Button("🎬 Generate Video", variant="primary", size="lg")
# Right column - Output
with gr.Column(scale=1):
video_output = gr.Video(
label="Generated Video",
height=450,
autoplay=True
)
info_output = gr.Textbox(
label="Generation Info",
lines=10,
elem_classes="generation-info",
show_copy_button=True
)
# Examples
with gr.Accordion("📝 Example Prompts", open=False):
gr.Examples(
examples=[
["A majestic dragon flying through clouds at sunset, cinematic lighting, 4K quality", None, "wan-fast-lora"],
["A serene ocean wave crashing on a beach, slow motion, golden hour", None, "wan-fast-lora"],
["A futuristic city with flying cars and neon lights, cyberpunk style", None, "wan-quality-lora"],
["A cat playing with a ball of yarn, cute, fluffy, soft lighting", None, "wan-fast-lora"],
["Time-lapse of flowers blooming in a garden, vibrant colors", None, "wan-motion-lora"],
],
inputs=[prompt, image_input, lora_dropdown],
label="Click to use example prompts"
)
# Footer
gr.HTML("""
""")
# Event handlers
refresh_btn.click(refresh_loras, outputs=lora_dropdown)
generate_btn.click(
fn=generate_video,
inputs=[
prompt,
negative_prompt,
image_input,
lora_dropdown,
lora_scale,
height,
width,
num_frames,
guidance_scale,
num_inference_steps,
fps,
seed
],
outputs=[video_output, info_output],
api_visibility="public"
)
# Launch the app
demo.launch(
theme=gr.themes.Soft(
primary_hue="indigo",
secondary_hue="purple",
neutral_hue="slate",
font=gr.themes.GoogleFont("Inter"),
text_size="md",
spacing_size="md",
radius_size="lg"
),
css=custom_css,
footer_links=[
{"label": "API", "url": "/api"},
{"label": "Gradio", "url": "https://gradio.app"},
{"label": "anycoder", "url": "https://huggingface.co/spaces/akhaliq/anycoder"}
],
allowed_paths=[LORA_CACHE_DIR] if os.path.exists(LORA_CACHE_DIR) else []
)