app.py

import os, sys
import tempfile
import gradio as gr
from src.gradio_demo import SadTalker  
# from src.utils.text2speech import TTSTalker
from huggingface_hub import snapshot_download

def get_source_image(image):   
        return image

try:
    import webui  # in webui
    in_webui = True
except:
    in_webui = False


def toggle_audio_file(choice):
    if choice == False:
        return gr.update(visible=True), gr.update(visible=False)
    else:
        return gr.update(visible=False), gr.update(visible=True)
    
def ref_video_fn(path_of_ref_video):
    if path_of_ref_video is not None:
        return gr.update(value=True)
    else:
        return gr.update(value=False)
    
#def download_model():
#    REPO_ID = 'vinthony/LipSyncer-V002rc'
#    snapshot_download(repo_id=REPO_ID, local_dir='./checkpoints', local_dir_use_symlinks=True)

def lipsyncer_demo():
    #download_model()
    lip_syncer = SadTalker(lazy_load=True)
    # tts_talker = TTSTalker()

    with gr.Blocks(analytics_enabled=False) as lipsyncer_interface:
        gr.Markdown("<div align='center'> <h2> LipSyncer: Learning Realistic 3D Motion Coefficients for Stylized Audio-Driven Single Image Talking Face Animation (CVPR 2023) </span> </h2> \
                    <a style='font-size:18px;color: #efefef' href='https://arxiv.org/abs/2211.12194'>Arxiv</a> &nbsp;&nbsp;&nbsp;&nbsp;&nbsp; </div>")
        
        
        gr.Markdown("""
        <b></b>
        """)
        
        with gr.Row().style(equal_height=False):
            with gr.Column(variant='panel'):
                with gr.Tabs(elem_id="lipsyncer_source_image"):
                    with gr.TabItem('Source image'):
                        with gr.Row():
                            source_image = gr.Image(label="Source image", source="upload", type="filepath", elem_id="img2img_image").style(width=512)


                with gr.Tabs(elem_id="lipsyncer_driven_audio"):
                    with gr.TabItem('Driving Methods'):
                        gr.Markdown("Possible driving combinations: <br> 1. Audio only 2. Audio/IDLE Mode + Ref Video(pose, blink, pose+blink) 3. IDLE Mode only 4. Ref Video only (all) ")

                        with gr.Row():
                            driven_audio = gr.Audio(label="Input audio", source="upload", type="filepath")
                            driven_audio_no = gr.Audio(label="Use IDLE mode, no audio is required", source="upload", type="filepath", visible=False)

                            with gr.Column():
                                use_idle_mode = gr.Checkbox(label="Use Idle Animation")
                                length_of_audio = gr.Number(value=5, label="The length(seconds) of the generated video.")
                                use_idle_mode.change(toggle_audio_file, inputs=use_idle_mode, outputs=[driven_audio, driven_audio_no]) # todo

                        with gr.Row():
                            ref_video = gr.Video(label="Reference Video", source="upload", type="filepath", elem_id="vidref").style(width=512)

                            with gr.Column():
                                use_ref_video = gr.Checkbox(label="Use Reference Video")
                                ref_info = gr.Radio(['pose', 'blink','pose+blink', 'all'], value='pose', label='Reference Video',info="How to borrow from reference Video?((fully transfer, aka, video driving mode))")

                            ref_video.change(ref_video_fn, inputs=ref_video, outputs=[use_ref_video]) # todo


            with gr.Column(variant='panel'): 
                with gr.Tabs(elem_id="lipsyncer_checkbox"):
                    with gr.TabItem('Settings'):
                        gr.Markdown("")
                        with gr.Column(variant='panel'):
                            # width = gr.Slider(minimum=64, elem_id="img2img_width", maximum=2048, step=8, label="Manually Crop Width", value=512) # img2img_width
                            # height = gr.Slider(minimum=64, elem_id="img2img_height", maximum=2048, step=8, label="Manually Crop Height", value=512) # img2img_width
                            with gr.Row():
                                pose_style = gr.Slider(minimum=0, maximum=45, step=1, label="Pose style", value=0) #
                                exp_weight = gr.Slider(minimum=0, maximum=3, step=0.1, label="expression scale", value=1) # 
                                blink_every = gr.Checkbox(label="use eye blink", value=True)

                            with gr.Row():
                                size_of_image = gr.Radio([256, 512], value=256, label='face model resolution', info="use 256/512 model?") # 
                                preprocess_type = gr.Radio(['crop', 'resize','full', 'extcrop', 'extfull'], value='crop', label='preprocess', info="How to handle input image?")
                            
                            with gr.Row():
                                is_still_mode = gr.Checkbox(label="Still Mode (fewer head motion, works with preprocess `full`)")
                                facerender = gr.Radio(['facevid2vid','pirender'], value='facevid2vid', label='facerender', info="which face render?")
                                
                            with gr.Row():
                                batch_size = gr.Slider(label="batch size in generation", step=1, maximum=10, value=1)
                                enhancer = gr.Checkbox(label="GFPGAN as Face enhancer")
                            
                            submit = gr.Button('Generate', elem_id="lipsyncer_generate", variant='primary')
                            
                with gr.Tabs(elem_id="lipsyncer_genearted"):
                        gen_video = gr.Video(label="Generated video", format="mp4").style(width=256)

        

        submit.click(
                fn=lip_syncer.test,
                inputs=[source_image,
                        driven_audio,
                        preprocess_type,
                        is_still_mode,
                        enhancer,
                        batch_size,                            
                        size_of_image,
                        pose_style,
                        facerender,
                        exp_weight,
                        use_ref_video,
                        ref_video,
                        ref_info,
                        use_idle_mode,
                        length_of_audio,
                        blink_every
                        ], 
                outputs=[gen_video]
                )

        with gr.Row():
            examples = [
                [
                    'examples/source_image/full_body_1.png',
                    'examples/driven_audio/bus.wav',
                    'crop',
                    True,
                    False
                ],
                [
                    'examples/source_image/full_body_2.webp',
                    'examples/driven_audio/japanese.wav',
                    'crop',
                    False,
                    False
                ],
                [
                    'examples/source_image/full3.png',
                    'examples/driven_audio/deyu.wav',
                    'crop',
                    False,
                    True
                ],
                [
                    'examples/source_image/full4.jpeg',
                    'examples/driven_audio/eluosi.wav',
                    'full',
                    False,
                    True
                ],
                [
                    'examples/source_image/art_6.jpeg',
                    'examples/driven_audio/imagine.wav',
                    'full',
                    True,
                    True
                ],
                [
                    'examples/source_image/art_8.png',
                    'examples/driven_audio/bus.wav',
                    'full',
                    True,
                    False
                ],
                [
                    'examples/source_image/art_13.png',
                    'examples/driven_audio/fayu.wav',
                    'resize',
                    True,
                    False
                ],
                [
                    'examples/source_image/art_9.png',
                    'examples/driven_audio/news.wav',
                    'resize',
                    False,
                    False
                ],
                [
                    'examples/source_image/art_5.png',
                    'examples/driven_audio/RD_Radio31_000.wav',
                    'resize',
                    True,
                    True
                ],
            ]
            gr.Examples(examples=examples,
                        inputs=[
                            source_image,
                            driven_audio,
                            preprocess_type,
                            is_still_mode,
                            enhancer], 
                        outputs=[gen_video],
                        fn=lip_syncer.test,
                        cache_examples=os.getenv('SYSTEM') == 'spaces') # 

    return lipsyncer_interface

if __name__ == "__main__":
    demo = lipsyncer_demo()
    demo.queue(max_size=10)
    demo.launch(debug=True)