import os
import shutil
from huggingface_hub import snapshot_download
import gradio as gr
import numpy as np
from PIL import Image
import soundfile as sf
import argparse
import uuid

os.chdir(os.path.dirname(os.path.abspath(__file__)))
from scripts.inference import inference_process

is_shared_ui = True if "fudan-generative-ai/hallo" in os.environ['SPACE_ID'] else False

if not is_shared_ui:
    hallo_dir = snapshot_download(repo_id="fudan-generative-ai/hallo", local_dir="pretrained_models")

def check_image_square(image_path):
    image = Image.open(image_path)
    if image.width != image.height:
        raise gr.Error("The uploaded image is not square. Please upload a square image.")
    return image_path

def convert_audio_to_wav(audio_path):
    if not audio_path.endswith('.wav'):
        audio_data, samplerate = sf.read(audio_path)
        wav_path = audio_path.rsplit('.', 1)[0] + '.wav'
        sf.write(wav_path, audio_data, samplerate)
        return wav_path
    return audio_path

def run_inference(source_image, driving_audio, pose_weight, face_weight, lip_weight, face_expand_ratio, progress=gr.Progress(track_tqdm=True)):
    if is_shared_ui:
        raise gr.Error("This Space only works in duplicated instances")
        
    unique_id = uuid.uuid4()
    
    args = argparse.Namespace(
        config='configs/inference/default.yaml',
        source_image=source_image,
        driving_audio=driving_audio,
        output=f'output-{unique_id}.mp4',
        pose_weight=pose_weight,
        face_weight=face_weight,
        lip_weight=lip_weight,
        face_expand_ratio=face_expand_ratio,
        checkpoint=None
    )
    
    inference_process(args)
    return f'output-{unique_id}.mp4' 

with gr.Blocks(theme='freddyaboulton/dracula_revamped@0.3.8') as demo:
    gr.Markdown(
        """
        # Talking Head Generation
        Upload a face image and driving audio, and adjust the weights to generate a talking head video.
        
        > **Note:**
        > - The face should be the main focus, making up 50%-70% of the image.
        > - The face should be facing forward, with a rotation angle of less than 30° (no side profiles).
        > - To make it work, duplicate the Space and run it on your own profile using a private GPU.
        > - An L4 costs US$0.80/h.
        """
    )
    
    with gr.Row():
        with gr.Column():
            avatar_face = gr.Image(type="filepath", label="Face", elem_id="face-input")
            driving_audio = gr.Audio(type="filepath", label="Driving Audio", elem_id="audio-input")
            
        with gr.Column():
            with gr.Accordion("Advanced Settings", open=False):
                pose_weight = gr.Slider(minimum=0.0, value=1.5, label="Pose Weight")
                face_weight = gr.Slider(minimum=0.0, value=1.0, label="Face Weight")
                lip_weight = gr.Slider(minimum=0.0, value=1.1, label="Lip Weight")
                face_expand_ratio = gr.Slider(minimum=0.0, value=1.2, label="Face Expand Ratio")
            
            generate = gr.Button("Generate", elem_id="generate-button")
            output_video = gr.Video(label="Your Talking Head", elem_id="output-video")

    avatar_face.change(fn=check_image_square, inputs=avatar_face, outputs=avatar_face)
    driving_audio.change(fn=convert_audio_to_wav, inputs=driving_audio, outputs=driving_audio)

    generate.click(
        fn=run_inference,
        inputs=[avatar_face, driving_audio, pose_weight, face_weight, lip_weight, face_expand_ratio],
        outputs=output_video
    )
    
demo.launch()