6Morpheus6/SadTalker
0
1import os, sys2import tempfile3import gradio as gr4from src.gradio_demo import SadTalker 5# from src.utils.text2speech import TTSTalker6from huggingface_hub import snapshot_download7 8def get_source_image(image): 9 return image10 11try:12 import webui # in webui13 in_webui = True14except:15 in_webui = False16 17 18def toggle_audio_file(choice):19 if choice == False:20 return gr.update(visible=True), gr.update(visible=False)21 else:22 return gr.update(visible=False), gr.update(visible=True)23 24def ref_video_fn(path_of_ref_video):25 if path_of_ref_video is not None:26 return gr.update(value=True)27 else:28 return gr.update(value=False)29 30def download_model():31 REPO_ID = 'vinthony/SadTalker-V002rc'32 snapshot_download(repo_id=REPO_ID, local_dir='./checkpoints', local_dir_use_symlinks=True)33 34def sadtalker_demo():35 36 download_model()37 38 sad_talker = SadTalker(lazy_load=True)39 # tts_talker = TTSTalker()40 41 with gr.Blocks(analytics_enabled=False) as sadtalker_interface:42 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> \43 <a style='font-size:18px;color: #efefef' href='https://arxiv.org/abs/2211.12194'>Arxiv</a> \44 <a style='font-size:18px;color: #efefef' href='https://sadtalker.github.io'>Homepage</a> \45 <a style='font-size:18px;color: #efefef' href='https://github.com/Winfredy/SadTalker'> Github </div>")46 47 48 gr.Markdown("""49 <b>You may duplicate the space and upgrade to GPU in settings for better performance and faster inference without waiting in the queue. <a style='display:inline-block' href="https://huggingface.co/spaces/vinthony/SadTalker?duplicate=true"><img src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a></b> \50 <br/><b>Alternatively, try our GitHub <a href=https://github.com/Winfredy/SadTalker> code </a> on your own GPU. </b> <a style='display:inline-block' href="https://github.com/Winfredy/SadTalker"><img src="https://img.shields.io/github/stars/Winfredy/SadTalker?style=social"/></a> \51 """)52 53 with gr.Row(): #.style(equal_height=False):54 with gr.Column(variant='panel'):55 with gr.Tabs(elem_id="sadtalker_source_image"):56 with gr.TabItem('Source image'):57 with gr.Row():58 source_image = gr.Image(label="Source image", sources="upload", type="filepath", elem_id="img2img_image") # .style(width=512)59 60 61 with gr.Tabs(elem_id="sadtalker_driven_audio"):62 with gr.TabItem('Driving Methods'):63 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) ")64 65 with gr.Row():66 driven_audio = gr.Audio(label="Input audio", sources="upload", type="filepath", max_length=180) # 180s67 driven_audio_no = gr.Audio(label="Use IDLE mode, no audio is required", sources="upload", type="filepath", visible=False)68 69 with gr.Column():70 use_idle_mode = gr.Checkbox(label="Use Idle Animation")71 length_of_audio = gr.Number(value=5, label="The length(seconds) of the generated video.")72 use_idle_mode.change(toggle_audio_file, inputs=use_idle_mode, outputs=[driven_audio, driven_audio_no]) # todo73 74 with gr.Row():75 ref_video = gr.Video(label="Reference Video", sources="upload", elem_id="vidref") # .style(width=512)76 77 with gr.Column():78 use_ref_video = gr.Checkbox(label="Use Reference Video")79 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))")80 81 ref_video.change(ref_video_fn, inputs=ref_video, outputs=[use_ref_video]) # todo82 83 84 with gr.Column(variant='panel'): 85 with gr.Tabs(elem_id="sadtalker_checkbox"):86 with gr.TabItem('Settings'):87 gr.Markdown("need help? please visit our [[best practice page](https://github.com/OpenTalker/SadTalker/blob/main/docs/best_practice.md)] for more detials")88 with gr.Column(variant='panel'):89 # width = gr.Slider(minimum=64, elem_id="img2img_width", maximum=2048, step=8, label="Manually Crop Width", value=512) # img2img_width90 # height = gr.Slider(minimum=64, elem_id="img2img_height", maximum=2048, step=8, label="Manually Crop Height", value=512) # img2img_width91 with gr.Row():92 pose_style = gr.Slider(minimum=0, maximum=45, step=1, label="Pose style", value=0) #93 exp_weight = gr.Slider(minimum=0, maximum=3, step=0.1, label="expression scale", value=1) # 94 blink_every = gr.Checkbox(label="use eye blink", value=True)95 96 with gr.Row():97 size_of_image = gr.Radio([256, 512], value=256, label='face model resolution', info="use 256/512 model?") # 98 preprocess_type = gr.Radio(['crop', 'resize','full', 'extcrop', 'extfull'], value='crop', label='preprocess', info="How to handle input image?")99 100 with gr.Row():101 is_still_mode = gr.Checkbox(label="Still Mode (fewer head motion, works with preprocess `full`)")102 facerender = gr.Radio(['facevid2vid','pirender'], value='facevid2vid', label='facerender', info="which face render?")103 104 with gr.Row():105 batch_size = gr.Slider(label="batch size in generation", step=1, maximum=10, value=1)106 enhancer = gr.Checkbox(label="GFPGAN as Face enhancer")107 108 submit = gr.Button('Generate', elem_id="sadtalker_generate", variant='primary')109 110 with gr.Tabs(elem_id="sadtalker_genearted"):111 gen_video = gr.Video(label="Generated video", scale=1) # .style(width=256)112 113 114 115 submit.click(116 fn=sad_talker.test,117 inputs=[source_image,118 driven_audio,119 preprocess_type,120 is_still_mode,121 enhancer,122 batch_size, 123 size_of_image,124 pose_style,125 facerender,126 exp_weight,127 use_ref_video,128 ref_video,129 ref_info,130 use_idle_mode,131 length_of_audio,132 blink_every133 ], 134 outputs=[gen_video],135 )136 137 with gr.Row():138 examples = [139 [140 'examples/source_image/full_body_1.png',141 'examples/driven_audio/bus_chinese.wav',142 'crop',143 True,144 False145 ],146 [147 'examples/source_image/full_body_2.png',148 'examples/driven_audio/japanese.wav',149 'crop',150 False,151 False152 ],153 [154 'examples/source_image/full3.png',155 'examples/driven_audio/deyu.wav',156 'crop',157 False,158 True159 ],160 [161 'examples/source_image/full4.jpeg',162 'examples/driven_audio/eluosi.wav',163 'full',164 False,165 True166 ],167 [168 'examples/source_image/full4.jpeg',169 'examples/driven_audio/imagine.wav',170 'full',171 True,172 True173 ],174 [175 'examples/source_image/full_body_1.png',176 'examples/driven_audio/bus_chinese.wav',177 'full',178 True,179 False180 ],181 [182 'examples/source_image/art_13.png',183 'examples/driven_audio/fayu.wav',184 'resize',185 True,186 False187 ],188 [189 'examples/source_image/art_5.png',190 'examples/driven_audio/chinese_news.wav',191 'resize',192 False,193 False194 ],195 [196 'examples/source_image/art_5.png',197 'examples/driven_audio/RD_Radio31_000.wav',198 'resize',199 True,200 True201 ],202 ]203 gr.Examples(examples=examples,204 inputs=[205 source_image,206 driven_audio,207 preprocess_type,208 is_still_mode,209 enhancer], 210 outputs=[gen_video],211 fn=sad_talker.test,212 cache_examples=os.getenv('SYSTEM') == 'spaces') # 213 214 return sadtalker_interface215 216 217if __name__ == "__main__":218 219 demo = sadtalker_demo()220 demo.queue(max_size=10, api_open=True)221 demo.launch(debug=True)222