import gradio as gr
import torch
from diffusers import WanPipeline, UniPCMultistepScheduler
from PIL import Image
import numpy as np
import random
import os

device = "cuda" if torch.cuda.is_available() else "cpu"

pipe = WanPipeline.from_pretrained(
    "Wan-AI/Wan2.2-TI2V-5B-Diffusers",
    torch_dtype=torch.float16 if device=="cuda" else torch.float32
)
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
pipe = pipe.to(device)

FIXED_FPS = 24
MIN_DURATION, MAX_DURATION = 1, 8


def generate_video(prompt, duration, init_image=None):
    duration = max(MIN_DURATION, min(MAX_DURATION, duration)) 
    input_image = None
    if init_image is not None:
        input_image = Image.fromarray(init_image).convert("RGB")
    video = pipe(
        prompt=prompt,
        img=input_image,
        height=512,
        width=512,
        duration_seconds=duration,
        guidance_scale=1.0
    ).videos[0]
    video_path = "output.mp4"
    video.save(video_path)
    return video_path

with gr.Blocks() as demo:
    gr.Markdown("## Wan 2.2 TI2V-5B Video Generator")
    
    with gr.Row():
        prompt_input = gr.Textbox(label="Prompt", placeholder="Describe your scene")
        duration_input = gr.Slider(label="Duration (seconds)", minimum=1, maximum=8, step=1, value=4)
        init_image_input = gr.Image(label="Optional Initial Image", type="numpy")
        generate_btn = gr.Button("Generate Video")
        output_video = gr.Video(label="Generated Video")
    
    generate_btn.click(
        generate_video,
        inputs=[prompt_input, duration_input, init_image_input],
        outputs=output_video
    )

demo.launch()