Upload 5 files

README.md

@@ -1,14 +1,13 @@
----
-title: SadTalker (Gradio 4.x)
-emoji: 😭
-colorFrom: purple
-colorTo: green
-sdk: gradio
-sdk_version: 4.40.0
-app_file: app.py
-pinned: false
-license: mit
----
-
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+---
+title: SadTalker (Gradio 4.x, latest PyTorch)
+emoji: 😭
+colorFrom: purple
+colorTo: green
+sdk: gradio
+sdk_version: 4.42.0
+app_file: app.py
+pinned: false
+license: mit
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -1,216 +1,206 @@
-import spaces
-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/SadTalker-V002rc'
-    snapshot_download(repo_id=REPO_ID, local_dir='./checkpoints', local_dir_use_symlinks=True)
-
-def sadtalker_demo():
-
-    download_model()
-
-    sad_talker = SadTalker(lazy_load=True)
-    # tts_talker = TTSTalker()
-
-    with gr.Blocks(analytics_enabled=False, theme="Hev832/Applio") 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>")
-        
-        
-        gr.Markdown("""
-        <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> \
-        <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> \
-        """)
-        
-        with gr.Row():
-            with gr.Column(variant='panel'):
-                with gr.Tabs(elem_id="sadtalker_source_image"):
-                    with gr.TabItem('Source image'):
-                        with gr.Row():
-                            source_image = gr.Image(label="Source image", type="filepath", elem_id="img2img_image")
-
-
-                    with gr.TabItem('Driving Methods',elem_id="sadtalker_driven_audio"):
-                        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", type="filepath")
-                            driven_audio_no = gr.Audio(label="Use IDLE mode, no audio is required", 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", elem_id="vidref")
-
-                        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.TabItem('Settings',elem_id="sadtalker_checkbox"):
-                        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=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="sadtalker_generate", variant='primary')
-                            
-                with gr.Tabs(elem_id="sadtalker_genearted"):
-                    gen_video = gr.Video(label="Generated video", format="mp4")
-
-        
-
-        submit.click(
-                fn=sad_talker.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_chinese.wav',
-                    'crop',
-                    True,
-                    False
-                ],
-                [
-                    'examples/source_image/full_body_2.png',
-                    '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/full4.jpeg',
-                    'examples/driven_audio/imagine.wav',
-                    'full',
-                    True,
-                    True
-                ],
-                [
-                    'examples/source_image/full_body_1.png',
-                    'examples/driven_audio/bus_chinese.wav',
-                    'full',
-                    True,
-                    False
-                ],
-                [
-                    'examples/source_image/art_13.png',
-                    'examples/driven_audio/fayu.wav',
-                    'resize',
-                    True,
-                    False
-                ],
-                [
-                    'examples/source_image/art_5.png',
-                    'examples/driven_audio/chinese_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=sad_talker.test,
-                        cache_examples=False
-                        #cache_examples=os.getenv('SYSTEM') == 'spaces'
-                       ) # 
-
-    return sadtalker_interface
- 
-
-if __name__ == "__main__":
-    demo = sadtalker_demo()
-    demo.queue()
-    demo.launch()
-    #demo.queue(max_size=10, api_open=True)
+import spaces
+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/SadTalker-V002rc'
+    snapshot_download(repo_id=REPO_ID, local_dir='./checkpoints', local_dir_use_symlinks=True)
+
+def sadtalker_demo():
+
+    download_model()
+
+    sad_talker = SadTalker(lazy_load=True)
+    # tts_talker = TTSTalker()
+
+    with gr.Blocks(analytics_enabled=False, theme="Hev832/Applio") 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>")
+        
+        gr.Markdown("""
+        <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> \
+        <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> \
+        """)
+        
+        with gr.Row():
+            with gr.Column(variant='panel'):
+                with gr.Tabs(elem_id="sadtalker_source_image"):
+                    with gr.TabItem('Source image'):
+                        with gr.Row():
+                            source_image = gr.Image(label="Source image", type="filepath", elem_id="img2img_image")
+
+                    with gr.TabItem('Driving Methods',elem_id="sadtalker_driven_audio"):
+                        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", type="filepath")
+                            driven_audio_no = gr.Audio(label="Use IDLE mode, no audio is required", 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", elem_id="vidref")
+                        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.TabItem('Settings',elem_id="sadtalker_checkbox"):
+                        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=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="sadtalker_generate", variant='primary')
+                            
+                with gr.Tabs(elem_id="sadtalker_genearted"):
+                    gen_video = gr.Video(label="Generated video", format="mp4")
+
+        submit.click(
+                fn=sad_talker.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_chinese.wav',
+                    'crop',
+                    True,
+                    False
+                ],
+                [
+                    'examples/source_image/full_body_2.png',
+                    '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/full4.jpeg',
+                    'examples/driven_audio/imagine.wav',
+                    'full',
+                    True,
+                    True
+                ],
+                [
+                    'examples/source_image/full_body_1.png',
+                    'examples/driven_audio/bus_chinese.wav',
+                    'full',
+                    True,
+                    False
+                ],
+                [
+                    'examples/source_image/art_13.png',
+                    'examples/driven_audio/fayu.wav',
+                    'resize',
+                    True,
+                    False
+                ],
+                [
+                    'examples/source_image/art_5.png',
+                    'examples/driven_audio/chinese_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=sad_talker.test,
+                        cache_examples=False
+                        #cache_examples=os.getenv('SYSTEM') == 'spaces'
+                       ) # 
+
+    return sadtalker_interface
+
+if __name__ == "__main__":
+    demo = sadtalker_demo()
+    demo.queue()
+    demo.launch()
+    #demo.queue(max_size=10, api_open=True)
     #demo.launch(debug=True)

packages.txt

@@ -1,2 +1 @@
-ffmpeg
-libsndfile1
+ffmpeg libsndfile1

pre-requirements.txt

@@ -0,0 +1 @@
+pip>=23.0.0

requirements.txt

@@ -1,24 +1,25 @@
-torch==1.13.1
-torchvision==0.14.1
-torchaudio==0.13.1
-numpy==1.23.5
-face_alignment==1.3.0
-imageio==2.19.3
-imageio-ffmpeg==0.4.7
-librosa==0.8.0
-numba==0.56.4
-resampy==0.3.1
-pydub==0.25.1 
+torch
+torchvision
+torchaudio
+numpy
+face_alignment
+imageio
+imageio-ffmpeg
+librosa
+numba
+resampy
+pydub
 scipy
-kornia==0.6.8
+kornia
 tqdm
-yacs==0.1.8
+yacs
 pyyaml  
-joblib==1.1.0
-scikit-image==0.19.3
-basicsr==1.4.2
-facexlib==0.3.0
+joblib
+scikit-image
+git+https://github.com/XPixelGroup/BasicSR
+git+https://github.com/TencentARC/GFPGAN
+facexlib
 dlib-bin
-gfpgan
 av
-safetensors
+safetensors
+TTS