app.py

import spaces
import os
import shutil
from huggingface_hub import snapshot_download
import gradio as gr
os.chdir(os.path.dirname(os.path.abspath(__file__)))

hallo_dir = snapshot_download(repo_id="fudan-generative-ai/hallo", local_dir="pretrained_models")
joyhallo_dir = snapshot_download(repo_id="jdh-algo/JoyHallo-v1", local_dir="pretrained_models/joyhallo")
wav_dir = snapshot_download(repo_id="TencentGameMate/chinese-wav2vec2-base", local_dir="pretrained_models/chinese-wav2vec2-base")
print(hallo_dir, joyhallo_dir)
print(os.listdir(hallo_dir))

from scripts.inference import predict

@spaces.GPU(duration=120)
def run_inference(source_image, driving_audio, progress=gr.Progress(track_tqdm=True)):
        
    return predict(source_image, driving_audio, 1.0, 1.0, 1.0, 1.2)


css = '''
div#warning-ready {
    background-color: #ecfdf5;
    padding: 0 16px 16px;
    margin: 20px 0;
    color: #030303!important;
}
div#warning-ready > .gr-prose > h2, div#warning-ready > .gr-prose > p {
    color: #057857!important;
}
div#warning-duplicate {
    background-color: #ebf5ff;
    padding: 0 16px 16px;
    margin: 20px 0;
    color: #030303!important;
}
div#warning-duplicate > .gr-prose > h2, div#warning-duplicate > .gr-prose > p {
    color: #0f4592!important;
}
div#warning-duplicate strong {
    color: #0f4592;
}
p.actions {
    display: flex;
    align-items: center;
    margin: 20px 0;
}
div#warning-duplicate .actions a {
    display: inline-block;
    margin-right: 10px;
}
.dark #warning-duplicate {
    background-color: #0c0c0c !important;
    border: 1px solid white !important;
}
'''

with gr.Blocks(css=css) as demo:
    gr.Markdown("# JoyHallo: Digital human model for Mandarin")
    gr.Markdown("Generate talking head avatars driven by Mandarin speech. Data requirements:")
    gr.Markdown("""
Image:
1. Cropped to square shape.
2. Face should be facing forward and occupy 50%-70% of the image.

Audio:
1. Use wav format.
2. Mandarin, English or mixed, with clear audio and suitable background music.
                
! Important: Too long audio will casue a very long processing time, please keep the audio length within 5s.
                """)
    with gr.Row():
        with gr.Column():
            avatar_face = gr.Image(type="filepath", label="Face")
            driving_audio = gr.Audio(type="filepath", label="Driving audio")
            generate = gr.Button("Generate")
        with gr.Column():
            output_video = gr.Video(label="Your talking head")

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