app.py

import os, sys
import gradio as gr
import numpy as np

# Monkey patch for facexlib on NumPy 1.24+ / 2.0+
if not hasattr(np, 'float'):
    np.float = float

from src.gradio_demo import SadTalker  


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


import torch

# GPU Optimizations for Google Colab
if torch.cuda.is_available():
    torch.backends.cudnn.benchmark = True
    torch.backends.cudnn.deterministic = False

if os.cpu_count() is not None:
    torch.set_num_threads(os.cpu_count())

def sadtalker_demo(checkpoint_path='checkpoints', config_path='src/config', warpfn=None):

    sad_talker = SadTalker(checkpoint_path, config_path, lazy_load=True)

    with gr.Blocks() as sadtalker_interface:
        gr.Markdown("<div align='center'> <h2> 😭 SadTalker: 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; \
                    <a style='font-size:18px;color: #efefef' href='https://sadtalker.github.io'>Homepage</a>  &nbsp;&nbsp;&nbsp;&nbsp;&nbsp; \
                     <a style='font-size:18px;color: #efefef' href='https://github.com/Winfredy/SadTalker'> Github </div>")
        
        with gr.Row():
            with gr.Column(variant='panel'):
                with gr.Tabs(elem_id="sadtalker_source_image"):
                    with gr.Tab('Upload image'):
                        with gr.Row():
                            source_image = gr.Image(label="Source image", type="filepath", elem_id="img2img_image", width=512)

                        
                with gr.Tabs(elem_id="sadtalker_driven_audio"):
                    with gr.Tab('Upload OR TTS'):
                        with gr.Column(variant='panel'):
                            driven_audio = gr.Audio(label="Input audio", type="filepath")
                        
                        if False:  # Disabled TTS to avoid heavy TTS dependency
                            from src.utils.text2speech import TTSTalker
                            tts_talker = TTSTalker()
                            with gr.Column(variant='panel'):
                                input_text = gr.Textbox(label="Generating audio from text", lines=5, placeholder="please enter some text here, we genreate the audio from text using @Coqui.ai TTS.")
                                tts = gr.Button('Generate audio',elem_id="sadtalker_audio_generate", variant='primary')
                                tts.click(fn=tts_talker.test, inputs=[input_text], outputs=[driven_audio])
                            
            with gr.Column(variant='panel'): 
                with gr.Tabs(elem_id="sadtalker_checkbox"):
                    with gr.Tab('Settings'):
                        gr.Markdown("need help? please visit our [[best practice page](https://github.com/OpenTalker/SadTalker/blob/main/docs/best_practice.md)] for more detials")
                        with gr.Column(variant='panel'):
                            with gr.Row():
                                pose_style = gr.Slider(minimum=0, maximum=46, step=1, label="Estilo de movimiento de cabeza (Plantillas 0-46)", value=0)
                                exp_weight = gr.Slider(minimum=0, maximum=3, step=0.1, label="Fuerza de la expresión facial (boca/ojos)", value=1)

                            with gr.Row():
                                size_of_image = gr.Radio([256, 512], value=256, label='Resolución del modelo de rostro')
                                preprocess_type = gr.Radio(['crop', 'resize','full', 'extcrop', 'extfull'], value='crop', label='Modo de recorte de imagen')
                            
                            with gr.Row():
                                is_still_mode = gr.Checkbox(label="Modo Quieto (Congela la cabeza, ideal para avatares formales)")
                                batch_size = gr.Slider(label="Velocidad (Batch Size)", step=1, maximum=32, value=8)
                                enhancer = gr.Checkbox(label="Mejorar calidad del rostro (GFPGAN)")
                                
                            submit = gr.Button('Generar Animación 🚀', elem_id="sadtalker_generate", variant='primary')


                with gr.Tabs(elem_id="sadtalker_genearted"):
                    with gr.Tab("Result"):
                        gen_video = gr.Video(label="Generated video", width=256)

        submit.click(
                    fn=sad_talker.test, 
                    inputs=[source_image,
                            driven_audio,
                            preprocess_type,
                            is_still_mode,
                            enhancer,
                            batch_size,                            
                            size_of_image,
                            pose_style,
                            exp_weight
                            ], 
                    outputs=[gen_video]
                    )

    return sadtalker_interface
 

if __name__ == "__main__":

    demo = sadtalker_demo()
    demo.queue()
    demo.launch(server_name="0.0.0.0", server_port=7860, share=True)